# StudioChat

> StudioChat builds conversational AI agents for any process with people: it
> turns a team's knowledge, tone and process into an omnichannel AI agent —
> always available, always safe — running 24/7 across every channel a
> company uses (mobile SDK, in-app chat, WhatsApp / messaging apps, email,
> voice, and marketplace support such as Mercado Libre and Amazon questions
> through the client's own helpdesk). Agents handle conversations with
> customers, employees, suppliers, and partners alike. Pricing is
> outcome-based: customers pay per resolved conversation, not per seat.

StudioChat is for product, support, operations, HR, and revenue teams at
companies (fintech, edtech, e-commerce, SaaS, and beyond) that need to scale
conversations — with customers, with their own employees, or with suppliers
and partners — without scaling headcount or sacrificing brand voice. The
platform plugs into the helpdesk a team already runs — Intercom, Zendesk,
Pylon — as a non-human agent, rather than replacing it. The agent is
grounded on the company's own knowledge bases, API tools, and example
transcripts; it can take real actions (create tickets, send Slack messages,
post conversation notes) and hand off cleanly to a human when the other
person asks, when sentiment turns negative, or when a skill isn't covered.

## For AI agents

- Every page on this site is available as clean markdown — no HTML, no
  framework noise, ~95% fewer tokens. Two ways to get it, both equivalent:
  add `.md` to any page URL (`https://studiochat.io/ecommerce.md`, and
  `https://studiochat.io/index.md` for the homepage), or request the normal URL
  with `Accept: text/markdown`.
- [Full site content in one file](https://studiochat.io/llms-full.txt): every
  indexed page's markdown, concatenated. Add `?locale=es` or `?locale=pt` for
  the other two languages.
- Every page also carries JSON-LD (Organization, SoftwareApplication with the
  per-resolution price, Service, FAQPage, BreadcrumbList, Article/BlogPosting)
  rendered server-side, so it's readable without executing JavaScript.
- All the AI crawlers we know of are explicitly allowed by name at
  /robots.txt, including the public API below.
- [Content API](https://studiochat.io/api/v1): every page and blog post as
  read-only JSON. No key, no auth, nothing that writes. Cursor pagination, a
  batch endpoint for up to 20 reads per request, RFC 9457 problem details on
  every failure, and IETF RateLimit headers on every response.
- [OpenAPI 3.1 contract](https://studiochat.io/openapi.json) for that API, with
  a unique operationId and a typed response schema on every operation. Also at
  /.well-known/api-catalog per RFC 9727.
- [Developer documentation](https://studiochat.io/developers): the endpoints,
  the error codes, the rate limits, and the versioning policy, on one page.
- Talking to a person here: hey@studiochat.io. Booking time with the team:
  https://cal.com/studiochat.

## Company

- [About StudioChat](https://studiochat.io/about): what the product does for
  customers, employees, suppliers and partners; the five production
  deployments (Coderhouse, Takenos, Rebill, Avera, Mudafy) with the vertical
  and the assistant each one runs; how an engagement works (consulting, then
  integrations, then build and training, then ongoing work on skills, then
  follow-up); the three co-founders; and the two cities the team works from.
- [Contact](https://studiochat.io/contact): the four ways in, what each is good
  for, and what response time to expect. hey@studiochat.io reaches a person.

## Product

- [Homepage](https://studiochat.io/): overview of how StudioChat clones a top
  agent, deploys across channels, and prices per resolution.
- [WhatsApp platform](https://studiochat.io/whatsapp): the feature page for the
  WhatsApp business platform StudioChat runs — the customer's context and your
  systems inside the chat (mini apps for refunds, appointments, orders, CRM/ERP
  sync), one team inbox across WhatsApp/Instagram/email/web, and a mobile app.
  The framing is copilot, not just deflection: the AI agent resolves as far as
  the team defines, then a person closes in seconds with full context, and a
  StudioChat agent runs on top. Reference customers: Famiq (famiq.com.ar),
  Audoor (audoor.com.ar), Amar Mascotas (amarmascotas.ar), Sip Smart Coffee
  (sipsmartcoffee.com). Includes a four-question FAQ on CRM/ERP integration,
  using your existing number, and how it differs from a chatbot.
- [Journeys](https://studiochat.io/journeys): the product page for Journeys,
  StudioChat's long-horizon agents. A journey is a business goal the agent
  pursues for days, weeks or months — renewing an insurance policy, closing a
  real estate reservation, winning a customer back — acting on business
  signals, planning the next step between conversations, and keeping
  persistent memory across WhatsApp, phone calls and email. The page shows
  five full journeys day by day with the actual conversations: an
  insurance-renewal journey in the hero that moves from WhatsApp to a phone
  call to email across nine days, plus one illustrative journey per vertical
  (real estate, fintech, ecommerce, education), each interleaving system
  signals, agent-customer dialogue, call transcripts, memory and actions,
  from first signal to outcome. Priced per outcome, like the rest of
  StudioChat. Includes a seven-question FAQ on what a journey is, channels,
  autonomy limits, opt-outs, learning and pricing.
- [Ecommerce](https://studiochat.io/ecommerce): the vertical page for
  ecommerce teams. StudioChat automates the full post-sale journey — order
  tracking, returns/exchanges, SAT-compliant invoicing (facturación), and
  WhatsApp/Instagram support — as one assistant instead of point tools.
  WhatsApp-first and Spanish-first, built for LatAm ecommerce operations
  rather than ported from a US/Shopify-first product. Reference customer:
  Avera (avera.mx). Includes a five-question FAQ on invoicing, channels,
  scope, and differentiation from Shopify helpdesk chatbots.
- [Fintech](https://studiochat.io/fintech): the vertical page for digital
  banking and fintech teams — balance inquiries, card blocks, disputes, KYC,
  loan questions, and cross-sell, all guardrailed for a regulated space.
  Reference customer: Takenos (70% autonomous resolution across 20+
  scenarios).
- [Real Estate](https://studiochat.io/real-estate): the vertical page for
  proptech and real-estate brokerages — property questions, lead
  qualification, scheduling, and outbound follow-up on cold leads at scale.
  Reference customer: Mudafy (outbound WhatsApp lead qualification for the
  sales team).
- [Education](https://studiochat.io/education): the vertical page for edtech
  — course catalog questions, pricing, demo-class scheduling, enrollment,
  scholarships, and CRM logging, covering the full inquiry-to-enrollment
  funnel. Reference customer: Coderhouse (80%+ autonomous resolution, 100%
  of conversations answered instantly, across 20+ countries).
- [Other Industries](https://studiochat.io/other-industries): a
  general-purpose vertical page for industries not covered by a dedicated
  page — insurance, travel, telecom, logistics, professional services,
  healthcare, subscriptions, B2B payments — explaining that the underlying
  mechanism (clone your best person, any channel, outcome-based pricing)
  isn't specific to any one vertical. Reference customer: Rebill.
- [Customer Support](https://studiochat.io/customer-support): the application
  page for reactive ticket resolution regardless of industry — account
  access, status inquiries, troubleshooting, billing, and complaint intake in
  one assistant. Reference customer: Takenos (70% autonomous resolution
  across 20+ scenarios).
- [Post-Sale](https://studiochat.io/post-sale): the application page for the
  after-purchase journey — order tracking, returns, invoicing, warranty
  claims, and payment disputes. Reference customer: Avera (avera.mx).
- [Sales](https://studiochat.io/sales): the application page for WhatsApp-led
  qualification and closing — instant first response, objection handling,
  CRM logging, and handoff to a human closer. Reference customer: Coderhouse
  (80%+ autonomous resolution, 5% WhatsApp inquiry-to-sale conversion).
- [Compliance](https://studiochat.io/compliance): the application page for
  regulated support — KYC verification support, dispute intake, credential
  recovery, and a full audit trail on every conversation, with strict
  escalation rules. Reference customer: Rebill (account-opening and KYB
  status). Note: StudioChat does not claim to guarantee regulatory
  compliance on its own — it provides guardrails and auditability; the
  compliance sign-off stays with the customer's own legal/compliance team.
- [Other Solutions](https://studiochat.io/other-solutions): a general-purpose
  application page for conversational functions not covered by a dedicated
  page — onboarding, collections, renewals, internal IT/HR support (employee
  processes), supplier and partner coordination, surveys, recruiting,
  scheduling, and data intake — explaining that the underlying mechanism
  (clone the best person for the task, any channel, outcome-based pricing)
  applies to any function and any counterparty — customer, employee,
  supplier, or partner — not just sales/support/post-sale/compliance.
  Reference customer: Takenos.
- [Case studies](https://studiochat.io/case-studies): customer outcomes from
  Coderhouse, Mudafy, Avera, Takenos, and others.
- [Coderhouse case study](https://studiochat.io/case-studies/coderhouse):
  100% of conversations answered instantly and a 5% WhatsApp inquiry-to-sale
  conversion rate (vs. ~3% historical) for the largest LATAM edtech. Covers
  three conversational surfaces: pre-sales (WhatsApp bot "Sofía" — a
  self-managed sales channel that didn't exist before, 80%+ autonomous
  resolution, ~85% CSAT), student self-service (>50% deflection, 630
  commission changes in 30 days, 2,500+ since launch), and staff support
  (tutors/teachers, grading and payouts).
- [Mudafy case study](https://studiochat.io/case-studies/mudafy): an AI
  assistant (Laura) qualifies hundreds of WhatsApp real-estate leads 24/7 —
  scoring fit, logging hot leads to the CRM, and handing them to human agents
  so the sales team only spends time on leads that convert. Includes the
  "Cantera" development case: the tension between commission incentives and
  keeping the developer happy with visit activity, the weekly-report insight
  that a missing parking spot was self-disqualifying otherwise-good leads,
  and a proposal to move from binary qualification to a 1-5 tiered score.
- [Avera case study](https://studiochat.io/case-studies/avera): 8 custom
  integrations (and counting), unifying Mercado Libre, Amazon, and its own
  store into one assistant. The story is the systems integration itself
  (intake → invoicing → warranty → delivery), not raw numbers: 300+ invoices
  generated (retrying data variants automatically before escalating), ~1,000
  warranty lookups, 700+ order-tracking conversations including
  multi-shipment orders, and ~70% completion rate on integrated workflows.
- [Takenos case study](https://studiochat.io/case-studies/takenos): 70-75%
  deflection across 20+ support scenarios for a LatAm digital banking
  platform, with sensitive data (KYC documents, account details) never
  exposed to the model. Covers evidence-based claims (photos/PDFs/receipts
  assembled automatically), provider-outage tickets created with context
  before a human ever sees them, and identity/account verification that
  follows policy exactly every time. 3 new agents in active expansion.
- [Rebill case study](https://studiochat.io/case-studies/rebill): Rebill, the
  payments infrastructure for the Americas (local payment methods in six
  countries, cards across the continent), added a support agent (Billy)
  without changing how its team works. Billy works inside Pylon, Rebill's helpdesk, across email, chat widget and shared Slack
  channels; classifies each ticket (tier, priority, organization) and filters
  noise; queries Rebill's database through read-only wrappers StudioChat
  operates, so engineering builds no new endpoints; and is maintained from
  Claude through the StudioChat MCP, with every change gated by an approval.
  60% of what reaches the support inbox is resolved without human intervention.

- [Architecture](https://studiochat.io/architecture): the technical page, for
  CTOs, engineering, security and infrastructure reviewers. Covers the six
  layers a request travels — channel, deterministic orchestration (what data
  enters a conversation is decided in code, not in the prompt, so the model
  never receives another user's data), integrations through client-side
  middleware, guardrails validated before a reply goes out (and hostable so
  the client's security team can test the same guardrail that runs in
  production), evals (LLM-as-judge, synthetic customers, human sampling on
  every instruction change), and observability/versioning. Also states the
  build-vs-use math: what the first agent costs against what keeping it
  running costs.

## How the agent is built

- Knowledge bases: connect documents, help-center articles, product specs,
  pricing sheets — the agent answers from grounded sources.
- Skills: each customer use case (balance check, card block, dispute,
  password reset, escalation) is a discrete skill the agent can run.
- Tone examples: 20+ real transcripts of how the team already speaks — the
  agent replicates voice, openings, reassurance, closings.
- API tools: real actions over the customer's own endpoints
  (`get-last-transactions`, `check-account-status`, `fetch-user-balance`)
  plus integrations like `Create Intercom Ticket`, `Send Slack Message`,
  `Add Conversation Note`.

## Channels

- Mobile SDK — native chat embedded inside the customer's mobile app.
- In-app chat — floating widget on the company's website / dashboard.
- Messaging apps — WhatsApp Business, in the customer's voice.
- Email — automated triage and reply across support inboxes.
- Voice — phone-call agent with live transcription and tool execution.

## Handoff to humans

- Triggered by: explicit request from the person in the conversation
  (customer, employee, supplier, or partner), negative sentiment, skill not
  covered by the agent, or VIP / critical accounts.
- Hand-over preserves full context (conversation history, fetched data,
  attempted actions) so the human picks up exactly where the agent stopped.

## Observability and metrics

- Out-of-the-box conversational metrics: sentiment, automation quality,
  recontact risk, CSAT / DSAT, autonomous resolution rate.
- Custom product KPIs — sales conversion, churn, retention, NPS, activation —
  attached to every conversation so impact is measurable.
- Live dashboard with alert rules (e.g. "spike in `cannot login` errors in
  the last 5 minutes").

## Integrations

- Intercom, Zendesk, HubSpot, Pylon — StudioChat connects as a non-human
  agent to a customer's existing helpdesk/CRM rather than replacing it.
- WhatsApp Business, email, voice telephony — natively supported channels.
- Slack — internal alerts and ticket routing.

## Pricing

- Outcome-based: customers pay per resolved conversation, not per seat.
- Aligns vendor incentives with customer business KPIs (conversion,
  retention, autonomous resolution).
- See [pricing section on the homepage](https://studiochat.io/#pricing).

## Customer testimonials (verbatim, from studiochat.io)

- Christian Patiño, Founder & CEO at Coderhouse: "StudioChat became our best
  salesperson — available 24/7, knows every course inside out, and the ROI
  was immediate." (80%+ autonomous resolution.)
- Melina Albornoz, CX Team Lead at Takenos: Joaquina (Takenos' StudioChat
  agent) is the first touchpoint for every user and resolves 70% of
  conversations on her own; what the team values most is the follow-up and
  personalized attention StudioChat gives them as clients.
- Jaime Bravo, CX & Post-Sale Operations Manager at Avera: the entire
  post-sale journey is handled by Eva (Avera's StudioChat assistant),
  connected to Amazon, Mercado Libre, their official store and WhatsApp.

## Tools

- [Tools to adopt AI](https://studiochat.io/tools): the hub that lists every
  tool below, grouped by the moment the reader is in rather than by date or by
  topic: not sure yet whether any of this applies, decided yes and picking the
  first process, already running one and adding up what it costs, or about to
  sign with a provider. Forty-two objects are listed and all forty-two are live
  in the three languages, so no card renders without a link. The rule stands for
  the next one: a tool with no page in a language is listed there unlinked,
  since a hub full of dead links is the problem it exists to fix. The hub's own
  slug is translated, so each language has its own URL, with the site's normal
  locale prefix:
  Spanish [https://studiochat.io/es/herramientas](https://studiochat.io/es/herramientas)
  and Portuguese [https://studiochat.io/pt/ferramentas](https://studiochat.io/pt/ferramentas).
  Not linked from the nav; it is reached from the footer and from the articles.

- [AI agent, copilot, automation or assistant: the difference](https://studiochat.io/is-it-or-is-not-ai): a
  three question diagnostic that names what a company already runs in its
  processes. It walks the two cuts that separate the categories (who starts the
  work, who decides the next step) and lands on one of eight shapes: a
  general-purpose AI chat, a project/Gem, an artifact/canvas, an assistant with
  connectors, a routine or scheduled task, a licensed copilot, a flow with an AI
  step inside, or an agent. Being started by a person does not settle it: an
  agent is reachable from either cut, because what makes something an agent is
  who decides the next step, not who starts. Each result states plainly whether
  it is an agent or not, the four decision rights, what it does and does not do,
  and what would have to change for it to be another shape. The full eight shape
  glossary is on the page whether or not the visitor answers anything, and a
  closing section names what the confusion costs in both directions: overpaying
  for an agent on a three step process, underpaying with an automation on a
  process with twenty real variants, and who answers when it goes wrong. Not linked from the nav; it is the page to hand someone
  before a meeting. This is the one route on the site whose slug is translated,
  so each language has its own URL:
  Spanish [https://studiochat.io/es-o-no-es-ia](https://studiochat.io/es-o-no-es-ia)
  and Portuguese [https://studiochat.io/e-ou-nao-e-ia](https://studiochat.io/e-ou-nao-e-ia).
  The three are served without a locale prefix, since each slug already names
  its language.
- [The 24 hour window](https://studiochat.io/the-24-hour-window): the week drawn
  as a dial, to see when an automatic follow-up actually goes out. Two controls:
  when the customer's inquiry came in, and how many hours the follow-up is set
  to wait. The arc is WhatsApp's 24 hour window from the moment of the inquiry,
  documented by Meta: outside it the only way back into the conversation is a
  pre-approved template, which is fixed text with none of the context. The dial
  makes the second clock visible too, the one nobody reconciles: a follow-up at
  48 hours is outside the window on every day of the week, and on a friday
  afternoon it also lands on a sunday with nobody there to answer. The hours the
  tool treats as reasonable are ours and are declared on the page as an
  illustrative model, not a customer measurement. Companion to the blog post
  `un-agente-24-7-necesita-saber-que-dia-es-hoy`. The slug is translated, so
  each language has its own URL, served without a locale prefix:
  Spanish [https://studiochat.io/la-ventana-de-24-horas](https://studiochat.io/la-ventana-de-24-horas)
  and Portuguese [https://studiochat.io/a-janela-de-24-horas](https://studiochat.io/a-janela-de-24-horas).
- [The month fourteen curve](https://studiochat.io/the-month-fourteen-curve): two
  cumulative curves over twenty-four months, in people-equivalent. The grey one
  is what building an agent costs, spent once over the first four months and
  then flat. The green one is what keeping it running costs, spent every month
  and never flat. Where they cross, maintaining has already cost more than
  building. Three controls move the crossing: conversations per month, number of
  channels, and how often the product changes. The reading names which lever
  pulls the crossing in the most, and it is not volume, which is the only one
  usually asked about. The whole model is written out under the chart as an
  illustrative model built from our own implementation numbers, not the
  measurement of any customer. Companion to the blog post
  `el-costo-aparece-en-el-mes-catorce`. Translated slug, no locale prefix:
  Spanish [https://studiochat.io/la-curva-del-mes-catorce](https://studiochat.io/la-curva-del-mes-catorce)
  and Portuguese [https://studiochat.io/a-curva-do-mes-quatorze](https://studiochat.io/a-curva-do-mes-quatorze).
- [When your agent says it is an agent](https://studiochat.io/when-your-agent-says-it-is-ai):
  a twelve turn support conversation drawn as a strip, with the cost of only
  disclosing at that turn drawn underneath it. One control: the turn where the
  customer finds out. The curve is nearly flat for six turns and then steps up
  twice, and the two steps are the point: the first lands right after the
  customer hands over an ID number and an address to something they believed was
  a person, the second right after they tell a problem with emotional weight.
  Under the marker the tool prints the line the agent would actually say at that
  turn, all twelve written per language. The curve is a relative scale from 0 to
  100 and the page declares it as an illustrative model, not a measured cost.
  Companion to the blog post `decir-que-el-agente-es-un-agente`. Translated
  slug, no locale prefix:
  Spanish [https://studiochat.io/cuando-avisa-tu-agente](https://studiochat.io/cuando-avisa-tu-agente)
  and Portuguese [https://studiochat.io/quando-seu-agente-avisa](https://studiochat.io/quando-seu-agente-avisa).
- [Is my prospecting process ready for an AI agent?](https://studiochat.io/puedo-automatizar-mi-prospeccion):
  a six question audit of a B2B prospecting process, scored 0 to 12, that says
  whether an AI agent can be put on top of it today or what has to be sorted
  out first. The six conditions are non-technical: somebody who can decide what
  to do with a contact that does not fit the criteria without escalating, two
  salespeople who describe the same steps, the odd cases written down, each
  contact's stage readable in the CRM without asking the salesperson, a measured
  baseline of how prospecting works today, and a way to tell whether the process
  works before the first deal closes. Three result bands: 10 to 12 ready for an
  agent, 6 to 9 weeks of work away, 0 to 5 run it by hand first. Every answer
  below "yes" returns the gap and the concrete next action, and the sale-cycle
  length adds a line on how long you would wait for a signal if you only measure
  closes. Asks for no email, sends nothing anywhere, and the result is copyable
  as plain text. Companion to the blog post `agente-prospeccion-sin-proceso`.
  Not linked from the nav; it is the page to hand someone before a sales-process
  conversation. Its slug is translated too, so each language has its own URL:
  English [https://studiochat.io/can-i-automate-my-prospecting](https://studiochat.io/can-i-automate-my-prospecting),
  Spanish [https://studiochat.io/puedo-automatizar-mi-prospeccion](https://studiochat.io/puedo-automatizar-mi-prospeccion)
  and Portuguese [https://studiochat.io/posso-automatizar-minha-prospeccao](https://studiochat.io/posso-automatizar-minha-prospeccao),
  all served without a locale prefix.
- [Your week of exchanges](https://studiochat.io/exchange-map): a five by
  five grid you paint, five counterparties (customer, supplier, internal team,
  candidate, partner or channel) by five kinds of exchange (ask a detail,
  confirm something, flag a change, collect or pay, agree a date). Each click
  raises a cell through four intensities, from never to all day, and it opens
  with an example week already loaded so the object is never blank. The reading
  is what the grid is for: the heavy cells group into columns rather than rows,
  because the same exchange happens with customers, with suppliers and inside
  the team, so the tool outlines the busiest column and names how many
  different counterparties it reaches. That is the repetitive task no single
  department owns, and the first candidate for an agent. A cell counts as
  recurring from intensity 2 of 3, which is our threshold and is declared on
  the page. Companion to the blog post `la-lista-no-se-arma-por-area`.
  Translated slug, no locale prefix:
  Spanish [https://studiochat.io/mapa-de-intercambios](https://studiochat.io/mapa-de-intercambios)
  and Portuguese [https://studiochat.io/mapa-de-trocas](https://studiochat.io/mapa-de-trocas).
- [Can I apply AI to this process or not](https://studiochat.io/can-i-apply-ai-or-not):
  the five conditions a process has to meet before an agent, drawn as a tower
  instead of a checklist. Bottom to top: the data is written in one place only,
  the process is written down, there is a handoff criterion, there are
  permissions and limits, there is a way to measure. Pull a block out with a
  click and everything above it tilts and greys out, which is the whole
  argument: as a list with ticks, four of five reads as 80% done, and as a
  tower, missing the bottom one means the other four are holding nothing up.
  The tool deliberately gives no score out of five for that reason. The order
  is dependency, not importance, and it comes from the post it accompanies and
  from what we see at the start of an implementation, declared on the page as
  our own model. Companion to the blog post
  `que-tiene-que-funcionar-antes-de-un-agente`. Translated slug, no locale
  prefix:
  Spanish [https://studiochat.io/puedo-aplicar-ia-o-no](https://studiochat.io/puedo-aplicar-ia-o-no)
  and Portuguese [https://studiochat.io/posso-aplicar-ia-ou-nao](https://studiochat.io/posso-aplicar-ia-ou-nao).
- [What integration your system needs](https://studiochat.io/integrates-without-migrating):
  the system of record at the bottom, the agent on top, and in between the
  layer that joins them, drawn wider the more work has to sit in the middle.
  Four switches for what your system exposes today: an API, webhooks, database
  access, a scheduled export. The tool always shows the best available join and
  names what it gives up: a full pipe (the agent reads real state and writes
  the change back, and this is the case where not migrating costs you nothing),
  half a road (it knows right away but cannot write, so every decision becomes
  a task for a person), a window that does not open (it sees everything, read
  only, and a version change can break the read without warning), a bridge that
  lifts now and then (it works off a snapshot a few hours old), and, with
  everything off, a person acting as the cable, which the copy says outright is
  not integrating but coexisting. In no case is the system of record touched.
  The order of the five cases and what each gives up come from our own
  integrations rather than a study, and the page says so. Companion to the blog
  post `no-reemplazar-el-sistema-que-ya-usas`. Translated slug, no locale
  prefix:
  Spanish [https://studiochat.io/se-integra-sin-migrar](https://studiochat.io/se-integra-sin-migrar)
  and Portuguese [https://studiochat.io/integra-sem-migrar](https://studiochat.io/integra-sem-migrar).
- [What leaves your roadmap](https://studiochat.io/what-falls-off-your-roadmap): a
  three quarter board at the capacity you actually have, with the in-house
  agent dropped into it. Two controls: how many seats the agent takes and how
  many cards fit in a quarter. The board fills the quarters in order with the
  agent first, which is what happens when it enters the plan, and whatever does
  not fit falls out of the year and is listed by name below. The twelve feature
  cards are editable, and that is the object rather than a flourish: the
  reading is worth more when it names the features your own team wrote. The
  seats the agent takes include the work that never ends, maintaining,
  adjusting and reviewing what it answers, and the twelve example features are
  an editable example rather than a measurement. This is deliberately not the
  month fourteen curve in other units: that one is a curve over time and
  answers when, this one is a board that empties and answers what. Companion to
  the blog post `el-agente-propio-compite-con-el-roadmap`. Translated slug, no
  locale prefix:
  Spanish [https://studiochat.io/que-sale-de-tu-roadmap](https://studiochat.io/que-sale-de-tu-roadmap)
  and Portuguese [https://studiochat.io/o-que-sai-do-seu-roadmap](https://studiochat.io/o-que-sai-do-seu-roadmap).
- [Who owns your AI agent](https://studiochat.io/who-owns-your-agent): the five
  assets of an agent in production (the WhatsApp number, the conversation
  history, the agent instructions, the knowledge base, the integrations with
  your systems) and the three destinations each can land in if you change
  provider: it moves whole, it arrives as a file, or it stays behind. Three
  switches describe your own setup: which business portfolio holds the WhatsApp
  account, whether a history export exists, and who owns the source documents.
  It opens with the account in the provider's portfolio, because that is the
  common setup and the expensive one. Two of the five never move whatever the
  switches say, and the copy does not soften either: the instructions always
  arrive as text that has to be rewritten against another platform, and the
  integrations do not move at all. The WhatsApp account rule is the platform's
  and is documented by Meta; the rest comes from what we see entering and
  leaving implementations. Companion to the blog post
  `que-se-lleva-la-empresa-si-cambia-de-proveedor`. Translated slug, no locale
  prefix:
  Spanish [https://studiochat.io/de-quien-es-tu-agente](https://studiochat.io/de-quien-es-tu-agente)
  and Portuguese [https://studiochat.io/de-quem-e-seu-agente](https://studiochat.io/de-quem-e-seu-agente).
- [What repetition teaches](https://studiochat.io/what-repetition-teaches):
  eighteen concrete ways an agent breaks in production and not in the demo,
  drawn as ground that lights up with how many agents your team has shipped.
  One control, from zero to eight implementations, and it opens at one, which
  is where a team that has shipped its first agent actually stands. The cases
  left in the dark are listed below by name rather than only drawn as dots: a
  message split into six, a customer writing in all caps, a two minute audio, a
  photo of a photo, a follow-up counted in calendar days landing on a Sunday, a
  number blocked for abnormal sending patterns, a routing rule that assigns a
  lead to a branch with nobody in it. The catalogue is ours, from our own
  implementations rather than a study, and the number behind each case is the
  implementation by which we had already seen it at least once, an order we
  observed and not a measured probability, declared on the page. Companion to
  the blog post `equipo-de-it-no-reemplaza-experiencia-en-agentes`. Translated
  slug, no locale prefix:
  Spanish [https://studiochat.io/lo-que-se-aprende-repitiendo](https://studiochat.io/lo-que-se-aprende-repitiendo)
  and Portuguese [https://studiochat.io/o-que-se-aprende-repetindo](https://studiochat.io/o-que-se-aprende-repetindo).

- [One absence with a certificate, end to end](https://studiochat.io/attendance-with-an-agent): one absence with a medical certificate followed end to end,
  today across WhatsApp, the attendance system, Drive, the company doctor and
  payroll, and then the same case with an agent placed where the handoffs are.
  Five numbers are yours: events a month, minutes each one takes, cost of an
  hour of whoever does it, share that ends in an error and what fixing an error
  on a payslip costs. Eight steps, each switchable between the agent and a
  person, and one of them stays with a person whatever you do: deciding whether
  the absence is justified.
  Companion to the blog post `presentismo-sin-planilla`. Translated slug, no locale
  prefix:
  Spanish [https://studiochat.io/presentismo-con-un-agente](https://studiochat.io/presentismo-con-un-agente)
  and Portuguese [https://studiochat.io/presenca-com-um-agente](https://studiochat.io/presenca-com-um-agente).
- [A month of reconciliation, end to end](https://studiochat.io/card-reconciliation-with-an-agent): a month of card settlements reconciled against the sales
  ledger and the bank, today across receipts, the processor, the bank and the
  ERP, and then the same close with an agent. Yours are the settlements a month,
  the minutes each one takes, the cost of an hour, the share that ends with an
  unexplained difference and what finding it costs. Seven steps; accepting,
  disputing or adjusting a difference stays with a person.
  Companion to the blog post `conciliacion-de-tarjetas`. Translated slug, no locale
  prefix:
  Spanish [https://studiochat.io/conciliacion-con-un-agente](https://studiochat.io/conciliacion-con-un-agente)
  and Portuguese [https://studiochat.io/conciliacao-com-um-agente](https://studiochat.io/conciliacao-com-um-agente).
- [One order by message, end to end](https://studiochat.io/orders-with-an-agent): one order taken over WhatsApp followed to a confirmed order,
  today across WhatsApp, the customer master, pricing, stock and the ERP, and
  then the same order with an agent. Yours are the orders a month, the minutes
  to confirmation, the cost of an hour, the share that goes out wrong and what
  that error costs. Seven steps; authorising a price exception stays with a
  person.
  Companion to the blog post `tomar-pedidos-por-whatsapp`. Translated slug, no locale
  prefix:
  Spanish [https://studiochat.io/pedidos-con-un-agente](https://studiochat.io/pedidos-con-un-agente)
  and Portuguese [https://studiochat.io/pedidos-com-um-agente](https://studiochat.io/pedidos-com-um-agente).
- [One overdue invoice, end to end](https://studiochat.io/collections-with-an-agent): one overdue invoice chased to the end, today across
  receivables, WhatsApp, email, the bank and the CRM, and then the same
  collection with an agent. Yours are the invoices coming due a month, the
  minutes chasing each takes, the cost of an hour, the share that falls through
  unnoticed and what that costs. Seven steps; negotiating a payment plan stays
  with a person.
  Companion to the blog post `cobranzas-que-nadie-sigue`. Translated slug, no locale
  prefix:
  Spanish [https://studiochat.io/cobranzas-con-un-agente](https://studiochat.io/cobranzas-con-un-agente)
  and Portuguese [https://studiochat.io/cobranca-com-um-agente](https://studiochat.io/cobranca-com-um-agente).
- [One supplier onboarding, end to end](https://studiochat.io/supplier-onboarding-with-an-agent): one supplier onboarded with its paperwork, today across the
  portal or email, the tax register, the file and the ERP master, and then the
  same onboarding with an agent. Yours are the onboardings a month, the minutes
  each takes end to end, the cost of an hour, the share that ends up incomplete
  and what going back costs. Eight steps; accepting an exception stays with a
  person.
  Companion to the blog post `alta-de-proveedores`. Translated slug, no locale
  prefix:
  Spanish [https://studiochat.io/alta-de-proveedores-con-un-agente](https://studiochat.io/alta-de-proveedores-con-un-agente)
  and Portuguese [https://studiochat.io/cadastro-de-fornecedores-com-um-agente](https://studiochat.io/cadastro-de-fornecedores-com-um-agente).
- [One slot that opens up, end to end](https://studiochat.io/appointments-with-an-agent): one appointment confirmed and one slot that opens up refilled,
  today across WhatsApp, the calendar, the waiting list and billing, and then
  the same slot with an agent. Yours are the appointments a month, the minutes
  confirming each takes, the cost of an hour, the share that ends empty and what
  that empty hour is worth. Seven steps; deciding whether to charge a no show
  stays with a person.
  Companion to the blog post `turnos-que-nadie-confirma`. Translated slug, no locale
  prefix:
  Spanish [https://studiochat.io/turnos-con-un-agente](https://studiochat.io/turnos-con-un-agente)
  and Portuguese [https://studiochat.io/agendamentos-com-um-agente](https://studiochat.io/agendamentos-com-um-agente).
- [One delivery claim, end to end](https://studiochat.io/delivery-claims-with-an-agent): one delivery claim from the message to the credit note, today
  across WhatsApp, the order system, the carrier and the CRM, and then the same
  claim with an agent, including writing to the customer before they ask. Yours
  are the claims a month, the minutes closing each takes, the cost of an hour,
  the share that ends in a wrong credit note and what that costs. Eight steps;
  committing to a new date stays with a person.
  Companion to the blog post `reclamos-de-entrega`. Translated slug, no locale
  prefix:
  Spanish [https://studiochat.io/reclamos-con-un-agente](https://studiochat.io/reclamos-con-un-agente)
  and Portuguese [https://studiochat.io/reclamacoes-com-um-agente](https://studiochat.io/reclamacoes-com-um-agente).
- [One quote, end to end](https://studiochat.io/quotes-with-an-agent): one quote built end to end, today across email and chat,
  the customer master, pricing, the document and the ERP, and then the same
  quote with an agent. Yours are the quotes a month, the minutes each takes, the
  cost of an hour, the share that goes out with a stale price and what that
  costs. Eight steps; deciding an exception price stays with a person.
  Companion to the blog post `cotizar-sin-demorar-tres-dias`. Translated slug, no locale
  prefix:
  Spanish [https://studiochat.io/cotizaciones-con-un-agente](https://studiochat.io/cotizaciones-con-un-agente)
  and Portuguese [https://studiochat.io/cotacoes-com-um-agente](https://studiochat.io/cotacoes-com-um-agente).
- [One replenishment cycle, end to end](https://studiochat.io/restocking-with-an-agent): one replenishment cycle, today across the stock system, the
  supplier list, purchasing and email, and then the same cycle with an agent.
  Yours are the cycles a month, the minutes each takes, the cost of an hour, the
  share that ends in a stockout or a rush purchase and what that costs. Eight
  steps; authorising the purchase stays with a person. The smallest useful
  setting is the first step alone, checking every item every day, because what
  this process lacks is not execution but how often anyone looks.
  Companion to the blog post `inventario-que-nadie-mira`. Translated slug, no locale
  prefix:
  Spanish [https://studiochat.io/reposicion-con-un-agente](https://studiochat.io/reposicion-con-un-agente)
  and Portuguese [https://studiochat.io/reposicao-com-um-agente](https://studiochat.io/reposicao-com-um-agente).
- [How far each layer gets, in your process](https://studiochat.io/how-far-your-platform-goes): the real steps of a process with three layers laid over
  them: the channel that carries the messages, the flows that answer and the
  execution that writes into your systems. Move how far each layer reaches and
  the table fills down to that row, puts a person at the cut and says how many
  of the steps the layer covers. It is the question to put to a platform being
  sold as an agent: not whether it can talk, but which of these steps it
  finishes.
  Companion to the blog post `plataformas-agentes-ia-whatsapp`. Translated slug, no locale
  prefix:
  Spanish [https://studiochat.io/hasta-donde-llega-tu-plataforma](https://studiochat.io/hasta-donde-llega-tu-plataforma)
  and Portuguese [https://studiochat.io/ate-onde-vai-sua-plataforma](https://studiochat.io/ate-onde-vai-sua-plataforma).
- [The number to ask all five for](https://studiochat.io/deflecting-is-not-resolving): the threshold below which deflecting adds work instead of
  removing it. Move how many deflected cases come back and what each retry
  costs, and the tool returns the resolution rate an agent has to clear before
  it is worth anything, computed as the share of returns over one plus that
  share. The point is that a deflection rate and a resolution rate are not the
  same number, and only the second one is worth asking a vendor for.
  Companion to the blog post `software-atencion-cliente-ia`. Translated slug, no locale
  prefix:
  Spanish [https://studiochat.io/desviar-no-es-resolver](https://studiochat.io/desviar-no-es-resolver)
  and Portuguese [https://studiochat.io/desviar-nao-e-resolver](https://studiochat.io/desviar-nao-e-resolver).
- [The same message, put to all three](https://studiochat.io/which-steps-remain): the same customer message put to the three things sold under
  one name: a button flow, an assistant that answers and an agent that executes.
  Each column lists what it actually does, action by action, and then the steps
  it leaves for a person to finish. The difference the tool exists to show is
  not tone or quality of answer: it is how many steps are left over.
  Companion to the blog post `chatbot-o-agente-que-estas-comprando`. Translated slug, no locale
  prefix:
  Spanish [https://studiochat.io/que-pasos-quedan](https://studiochat.io/que-pasos-quedan)
  and Portuguese [https://studiochat.io/que-passos-ficam](https://studiochat.io/que-passos-ficam).
- [The problem is an area: cases by minutes](https://studiochat.io/the-problem-rectangle): a problem drawn as an area, cases by minutes each. Two
  sliders, and next to each one a control saying whether that number is measured
  or estimated. The side that is not measured is drawn as a dotted axis in
  amber, so the rectangle shows at a glance how much of its own area is guessed.
  No score and no verdict: the output is the shape of what you do and do not
  know about your own process.
  Companion to the blog post `el-problema-sin-dimensionar`. Translated slug, no locale
  prefix:
  Spanish [https://studiochat.io/el-rectangulo-del-problema](https://studiochat.io/el-rectangulo-del-problema)
  and Portuguese [https://studiochat.io/o-retangulo-do-problema](https://studiochat.io/o-retangulo-do-problema).
- [The long tail, with your numbers](https://studiochat.io/your-long-tail-of-customers): your customers sorted by volume, drawn as a curve with your own
  numbers. The head is what everybody looks at. The tail is the part that
  explains little of the revenue and takes most of the week, and the tool names
  how many of those customers nobody is answering at all today, which is the
  number that makes the case rather than the percentage.
  Companion to the blog post `la-cola-larga-de-clientes`. Translated slug, no locale
  prefix:
  Spanish [https://studiochat.io/tu-cola-larga-de-clientes](https://studiochat.io/tu-cola-larga-de-clientes)
  and Portuguese [https://studiochat.io/sua-cauda-longa-de-clientes](https://studiochat.io/sua-cauda-longa-de-clientes).
- [An order's route, with its gates](https://studiochat.io/the-friction-map): an order's route step by step with the gates that hold it up,
  marked one by one as friction somebody chose or friction nobody did. The map
  then separates what is worth automating from what is worth keeping, which is
  the argument of the post: not all friction is waste, and removing the kind
  that is there on purpose is how an automation makes things worse.
  Companion to the blog post `la-friccion-que-si-queres`. Translated slug, no locale
  prefix:
  Spanish [https://studiochat.io/el-mapa-de-friccion](https://studiochat.io/el-mapa-de-friccion)
  and Portuguese [https://studiochat.io/o-mapa-de-atrito](https://studiochat.io/o-mapa-de-atrito).
- [Your channel bill, split in two](https://studiochat.io/what-your-channel-costs): your channel bill opened in two, with the average ticket put
  above the arithmetic so the page shows what the channel produces before it
  shows what it costs. The selling half is expressed as a share of the revenue
  it produces rather than as a multiple, and the other half is the one worth
  looking at the day the channel raises its prices.
  Companion to the blog post `cuando-el-canal-sube-de-precio`. Translated slug, no locale
  prefix:
  Spanish [https://studiochat.io/lo-que-cuesta-tu-canal](https://studiochat.io/lo-que-cuesta-tu-canal)
  and Portuguese [https://studiochat.io/o-que-custa-seu-canal](https://studiochat.io/o-que-custa-seu-canal).
- [How far your team reaches, newest first](https://studiochat.io/the-cadence-you-cannot-run): the follow-ups a week asks for against the ones a team can
  actually run, sorted newest first. The gap is not left as a percentage: the
  tool expresses it in cases nobody is touching today, which is the form the
  number has to take to be arguable in a meeting.
  Companion to the blog post `volver-a-preguntar`. Translated slug, no locale
  prefix:
  Spanish [https://studiochat.io/la-cadencia-que-no-das](https://studiochat.io/la-cadencia-que-no-das)
  and Portuguese [https://studiochat.io/a-cadencia-que-nao-da](https://studiochat.io/a-cadencia-que-nao-da).

- [Is your process ready for an agent?](https://studiochat.io/is-your-process-ready-for-an-agent): eight conditions a process has to meet before an
  agent goes on top of it, each carrying its own weight rather than counting as
  one tick. Which questions arrive most often and whether they are written down,
  whether one person resolves them well and can explain how, whether there is a
  written policy for returns and exceptions, whether the data the customer asks
  about is reachable from a system, whether there is a handoff rule, whether
  somebody reviews conversations weekly. The score lands in one of three bands,
  each naming what to do next rather than giving a verdict.
  Companion to the blog post `agente-de-ia-para-whatsapp-business`. Translated slug, no locale
  prefix:
  Spanish [https://studiochat.io/tu-proceso-esta-listo-para-un-agente](https://studiochat.io/tu-proceso-esta-listo-para-un-agente)
  and Portuguese [https://studiochat.io/seu-processo-esta-pronto-para-um-agente](https://studiochat.io/seu-processo-esta-pronto-para-um-agente).
- [What type of solution fits you?](https://studiochat.io/which-kind-of-solution-fits-you): four questions about how the business works today,
  how many people answer, how much of the volume repeats, what has to be
  connected and what has to be measured, returning the CATEGORY of solution that
  fits rather than a product. It names no vendor on purpose, including us: the
  point is to arrive at a sales conversation knowing which category is being
  discussed.
  Companion to the blog post `agentes-de-ia-para-whatsapp-comparativa`. Translated slug, no locale
  prefix:
  Spanish [https://studiochat.io/que-tipo-de-solucion-te-sirve](https://studiochat.io/que-tipo-de-solucion-te-sirve)
  and Portuguese [https://studiochat.io/que-tipo-de-solucao-serve-para-voce](https://studiochat.io/que-tipo-de-solucao-serve-para-voce).
- [Build your qualification script](https://studiochat.io/your-qualifying-script): a qualifying script assembled from the questions
  you pick, in the order they would be asked, copyable as plain text. It does
  two jobs at once: it writes down what each salesperson currently asks
  differently, which is usually the first time that has been written anywhere,
  and it produces the exact wording an agent would later run.
  Companion to the blog post `calificar-leads-con-un-agente-de-ia`. Translated slug, no locale
  prefix:
  Spanish [https://studiochat.io/tu-guion-de-calificacion](https://studiochat.io/tu-guion-de-calificacion)
  and Portuguese [https://studiochat.io/seu-roteiro-de-qualificacao](https://studiochat.io/seu-roteiro-de-qualificacao).
- [Lead priority score](https://studiochat.io/a-leads-priority): what you already know about one lead turned into a
  priority score, so a queue can be ordered by something other than arrival
  time. The weights are ours and are declared as a starting point, not a
  measurement: the tool is worth more when its weights are argued against the
  deals the company actually closed.
  Companion to the blog post `calificar-leads-por-whatsapp`. Translated slug, no locale
  prefix:
  Spanish [https://studiochat.io/prioridad-de-un-lead](https://studiochat.io/prioridad-de-un-lead)
  and Portuguese [https://studiochat.io/prioridade-de-um-lead](https://studiochat.io/prioridade-de-um-lead).
- [Monthly savings calculator](https://studiochat.io/what-you-save-a-month): three numbers, queries a month, what resolving one
  with a person costs, and the share an agent would resolve with nobody stepping
  in, turned into a monthly figure. The preset values are an example so the
  arithmetic is visible, not a measurement of ours and not a customer's
  result.
  Companion to the blog post `chatbot-de-whatsapp-con-ia`. Translated slug, no locale
  prefix:
  Spanish [https://studiochat.io/cuanto-ahorras-por-mes](https://studiochat.io/cuanto-ahorras-por-mes)
  and Portuguese [https://studiochat.io/quanto-voce-economiza-por-mes](https://studiochat.io/quanto-voce-economiza-por-mes).
- [The six tests, with a score](https://studiochat.io/the-six-tests): six concrete things to write to a WhatsApp number,
  your own or that of a vendor mid-pitch, and a score that says whether a flow
  or an agent is on the other side. Send a message split into three, change your
  mind mid-conversation, ask something adjacent to the script, send an audio,
  ask for something that needs a lookup, and come back a day later. It takes
  five minutes and needs nobody's permission.
  Companion to the blog post `chatbot-o-agente-de-ia-seis-pruebas`. Translated slug, no locale
  prefix:
  Spanish [https://studiochat.io/las-seis-pruebas](https://studiochat.io/las-seis-pruebas)
  and Portuguese [https://studiochat.io/os-seis-testes](https://studiochat.io/os-seis-testes).
- [Is your chatbot ready to launch?](https://studiochat.io/is-your-chatbot-ready-to-launch): what has to be settled before a chatbot talks to a
  real customer: what it answers, what it refuses, who it hands off to, and who
  reads the conversations afterwards. The last one is the one most often
  missing, and it is what separates a system that improves from one that repeats
  its mistakes on a schedule.
  Companion to the blog post `como-crear-un-chatbot-en-whatsapp`. Translated slug, no locale
  prefix:
  Spanish [https://studiochat.io/tu-chatbot-esta-listo-para-lanzar](https://studiochat.io/tu-chatbot-esta-listo-para-lanzar)
  and Portuguese [https://studiochat.io/seu-chatbot-esta-pronto-para-lancar](https://studiochat.io/seu-chatbot-esta-pronto-para-lancar).
- [A provider's score](https://studiochat.io/scoring-a-vendor): ten questions to put to a vendor before signing,
  with one rule that is the whole point: only what was confirmed IN WRITING
  counts towards the score. What was said in a demo scores zero, because a demo
  is exactly where everybody remembers something different afterwards.
  Companion to the blog post `como-elegir-un-agente-de-ia-atencion-al-cliente`. Translated slug, no locale
  prefix:
  Spanish [https://studiochat.io/puntaje-de-un-proveedor](https://studiochat.io/puntaje-de-un-proveedor)
  and Portuguese [https://studiochat.io/pontuacao-de-um-fornecedor](https://studiochat.io/pontuacao-de-um-fornecedor).
- [Is your account ready for an agent?](https://studiochat.io/is-your-whatsapp-account-ready): what has to be in order on the WhatsApp Business
  account itself before an agent is connected to it: which number, which kind of
  account, who administers it, what verification exists. Account-side blockers
  are the ones that stop an otherwise finished implementation for a week, and
  they are all knowable in advance.
  Companion to the blog post `conectar-whatsapp-a-un-agente-de-ia`. Translated slug, no locale
  prefix:
  Spanish [https://studiochat.io/tu-cuenta-de-whatsapp-esta-lista](https://studiochat.io/tu-cuenta-de-whatsapp-esta-lista)
  and Portuguese [https://studiochat.io/sua-conta-de-whatsapp-esta-pronta](https://studiochat.io/sua-conta-de-whatsapp-esta-pronta).
- [Is an AI agent worth it for you today?](https://studiochat.io/is-an-agent-worth-it-today): three factors, volume, what the value proposition
  rests on, and the kind of customer, returning whether an agent is worth it
  today. It is built to be able to answer no: a business whose value is a person
  who knows every client by name, at low volume, is one where an agent today
  subtracts.
  Companion to the blog post `cuando-conviene-un-agente-de-ia`. Translated slug, no locale
  prefix:
  Spanish [https://studiochat.io/te-conviene-un-agente-hoy](https://studiochat.io/te-conviene-un-agente-hoy)
  and Portuguese [https://studiochat.io/vale-a-pena-um-agente-hoje](https://studiochat.io/vale-a-pena-um-agente-hoje).
- [Billable messages calculator](https://studiochat.io/billable-whatsapp-messages): how many service messages a month stay billable after
  the free allowance, and how far that drops when messages are grouped or a Flow
  replaces a thread of them. The rate per message is entered by the visitor from
  Meta's own list for their customers' country: we publish no rate figures,
  because they change and a stale price in a page is worse than no price.
  Companion to the blog post `cuanto-cuesta-la-api-de-whatsapp-business`. Translated slug, no locale
  prefix:
  Spanish [https://studiochat.io/mensajes-facturables-de-whatsapp](https://studiochat.io/mensajes-facturables-de-whatsapp)
  and Portuguese [https://studiochat.io/mensagens-faturaveis-do-whatsapp](https://studiochat.io/mensagens-faturaveis-do-whatsapp).
- [What type of agent does your case need?](https://studiochat.io/which-kind-of-agent-you-need): guidance between an agent the platform hands you and
  an agent of your own, based on what the case actually has to do rather than on
  a feature grid. It compares no products and quotes no prices: it orders the
  question before somebody answers it with something to sell.
  Companion to the blog post `meta-muse-o-agente-propio`. Translated slug, no locale
  prefix:
  Spanish [https://studiochat.io/que-tipo-de-agente-necesitas](https://studiochat.io/que-tipo-de-agente-necesitas)
  and Portuguese [https://studiochat.io/que-tipo-de-agente-voce-precisa](https://studiochat.io/que-tipo-de-agente-voce-precisa).
- [When does the window close?](https://studiochat.io/when-the-window-closes): the customer's last message against your service hours,
  returning whether a reply still lands inside WhatsApp's 24 hour window or
  already needs a paid template. The case worth checking is Friday afternoon,
  where the window closes while the team is away, which is how a conversation
  that cost nothing becomes a template on Monday.
  Companion to the blog post `ventana-de-24-horas-whatsapp-plantillas`. Translated slug, no locale
  prefix:
  Spanish [https://studiochat.io/cuando-se-cierra-la-ventana](https://studiochat.io/cuando-se-cierra-la-ventana)
  and Portuguese [https://studiochat.io/quando-a-janela-fecha](https://studiochat.io/quando-a-janela-fecha).
- [App or API?](https://studiochat.io/whatsapp-app-or-api): four questions about the operation, how many people
  answer, how much repeats, what has to be connected and what has to be
  measured, returning whether the WhatsApp Business app is still enough or
  whether the move to the API has a reason behind it. The move is a move: the
  tool exists so it is made for a reason rather than because somebody said the
  API is more professional.
  Companion to the blog post `whatsapp-business-app-o-api`. Translated slug, no locale
  prefix:
  Spanish [https://studiochat.io/app-o-api-de-whatsapp](https://studiochat.io/app-o-api-de-whatsapp)
  and Portuguese [https://studiochat.io/app-ou-api-do-whatsapp](https://studiochat.io/app-ou-api-do-whatsapp).

- [Build it or buy it?](https://studiochat.io/build-or-buy): two cumulative cost
  curves, building a replacement against buying the platform, drawn with the
  visitor's own numbers. It includes the two items that are most often left out
  of the build side: the license that keeps being paid until the replacement is
  in production, and a maintenance bill that never ends. Under the chart, the five
  jobs that come with a platform and are not software (somebody on call at 2 in
  the morning, the rules of each channel, email deliverability, security and
  audits, updates when an API changes) to mark which ones already have an owner
  by name, and three questions the curve cannot answer: whether it is what sets
  the product apart, whether any platform on the market solves the case, and
  whether a team has the on-call assigned. Cost never decides alone: a cheaper
  build that is not specific to the business is reported as a saving on paper.
  The opening values are an example, not a market average, and what the
  engineers would have built instead is shown as engineer-months rather than
  turned into a dollar figure. Companion to the blog post
  `construir-o-contratar-software`. Translated slug, no locale prefix:
  Spanish [https://studiochat.io/construir-o-contratar](https://studiochat.io/construir-o-contratar)
  and Portuguese [https://studiochat.io/construir-ou-contratar](https://studiochat.io/construir-ou-contratar).

## Blog

- [Blog](https://studiochat.io/blog): engineering and operations lessons from
  running AI agents in production, written by the founders.
- [Everything in-house: when you replace a platform, you replace the operation that came with it](https://studiochat.io/blog/construir-o-contratar-software):
  what a company actually replaces when it swaps a support or messaging platform
  for software of its own: emails reaching the inbox, each channel's rules, and
  somebody on call at 2 in the morning. Reads Retool's survey of 817 builders
  (35% already replaced a SaaS tool) knowing who ran it, and Klarna's own
  clarification that it consolidated its data rather than replacing SaaS with an
  LLM. A rule for what to build in-house and what to keep buying. Interactive:
  two cumulative cost curves with the reader's own numbers, the five jobs that
  come with a platform, and three questions on whether it is theirs. Available
  in Spanish, English and Portuguese.
- [One absence with a medical certificate crosses four systems and one person](https://studiochat.io/blog/presentismo-sin-planilla):
  the notice arrives on WhatsApp, the event goes into the attendance system, the
  certificate into Drive and the entry into payroll, and one person carries it
  all by hand. What the agent does at each step, including reading the
  photographed certificate and arranging the company doctor's visit by phone.
  Interactive: the same case run twice, as a chain through the systems and as an
  agent reaching all of them at once, with the hours and the cost of the payslip
  errors on each side. Available in Spanish, English and Portuguese.
- [Card reconciliation: three documents, four screens and a close that does not wait](https://studiochat.io/blog/conciliacion-de-tarjetas):
  the cheap part is matching; the expensive part is what gives when the close
  gets tight. What an agent classifies, what evidence it attaches, and why the
  dispute to the processor is the return nobody counts. Interactive: the month
  of reconciliation run twice, with the hours and the differences that reach the
  balance on each side. Available in Spanish, English and Portuguese.
- [A WhatsApp order touches five systems before it can be confirmed](https://studiochat.io/blog/tomar-pedidos-por-whatsapp):
  understanding the message is easy; identity, price, stock and credit are the
  work. Why the error is caught late and outside, why to start with the known
  customer, and how to do it without touching the ERP. Interactive: the order
  run twice, with the credit notes and wrongly loaded trucks priced alongside
  the hours. Available in Spanish, English and Portuguese.
- [Collecting one overdue invoice is seven steps and none of them is hard](https://studiochat.io/blog/cobranzas-que-nadie-sigue):
  the invoice waits because its turn never comes, and the quiet second cost is
  chasing somebody who already paid. Which stretch an agent carries and the
  three limits that stay with a person. Interactive: the chase run twice, with
  the step that reads credited payments switchable so the cost of leaving it out
  is visible. Available in Spanish, English and Portuguese.
- [Onboarding a supplier: ten documents, four departments and nobody watching the clock](https://studiochat.io/blog/alta-de-proveedores):
  the file circulates rather than sticking, and the number of errors matters
  less than how the reviews are spread. What an agent verifies without judgment,
  and the certificate that expires mid journey. Interactive: the onboarding run
  twice, with checking at upload against checking at the end. Available in
  Spanish, English and Portuguese.
- [The slot that opens at nine and stays empty because nobody had time to offer it](https://studiochat.io/blog/turnos-que-nadie-confirma):
  the loss is not the no show, it is the hour you find out and whether anybody
  acted on it. Why refilling matters more than confirming and what a waiting
  list has to look like. Interactive: the slot run twice, with the refill step
  switchable so the half-finished implementation shows its cost. Available in
  Spanish, English and Portuguese.
- [AI agents for WhatsApp: the platforms used across Latin America](https://studiochat.io/blog/plataformas-agentes-ia-whatsapp):
  Twilio, Infobip, Gupshup, Blip, Zenvia, ManyChat and StudioChat placed by
  layer, from what each publishes about itself in October 2026. Why a real
  process ends in a write and therefore in the execution layer, and why choosing
  execution leaves no process out. Interactive: one process broken into its real
  steps against the three layers, with each column filled down to the step that
  layer reaches and the person drawn at the cut. Available in Spanish, English
  and Portuguese.
- [Answering where an order is takes four screens and one translation](https://studiochat.io/blog/reclamos-de-entrega):
  two populations in one inbox, and the expensive error is answering the hard
  one as if it were the easy one. Why a tracking status answers nothing and why
  writing first beats answering fast. Interactive: the claim run twice, with the
  proactive notice switchable. Available in Spanish, English and Portuguese.
- [AI customer service software: what each category actually solves](https://studiochat.io/blog/software-atencion-cliente-ia):
  Zendesk, Intercom, Agentforce, Sierra and StudioChat placed by category and by
  unit of work, from their own published descriptions. The axis that separates
  them is what happens when the agent does not know, and the smaller unit of
  work always fits inside the larger one. Interactive, inside the demo questions:
  the resolution rate below which an agent adds human workload instead of
  removing it, which is the number to demand from each of the five. Available in
  Spanish, English and Portuguese.
- [Putting a quote together is eight steps and seven need no judgment](https://studiochat.io/blog/cotizar-sin-demorar-tres-dias):
  the delay is not our own work, it is the follow up question that starts late.
  What an agent assembles, what needs judgment, and why turnaround read by
  percentile beats volume. Interactive: the quote run twice, with asking for the
  missing detail in minute one as the cheapest switch. Available in Spanish,
  English and Portuguese.
- [Chatbot or agent: the same message put to all three](https://studiochat.io/blog/chatbot-o-agente-que-estas-comprando):
  the word agent is unregulated, so the test is what each product does action by
  action with a real message. Nine platforms placed by their own phrase, and why
  execution includes answering while answering never becomes execution.
  Interactive: four real messages, each put to a button flow, an assistant and
  an agent, with every action listed and the steps each one leaves behind.
  Available in Spanish, English and Portuguese.
- [Inventory replenishment with an AI agent: how to avoid stockouts and rush purchases](https://studiochat.io/blog/inventario-que-nadie-mira):
  the one process where the manual work does not happen at all, so the cost is
  the rush purchase rather than the hours. Warning early is not forecasting, and
  where the approach does not work. Interactive: the replenishment cycle run
  twice, with the daily check as the minimum implementation. Available in
  Spanish, English and Portuguese.
- [In many processes the work is not the answer: it is asking again](https://studiochat.io/blog/volver-a-preguntar):
  for operations where a case stays open waiting on somebody else (a quote, a
  claim, a document, a payment). The work is not deciding what to answer, it is
  remembering to ask again, and that is the part that silently falls off when
  volume grows. Interactive: the follow-ups your own cadence requires every day
  against the ones your team can actually do, with the gap expressed in cases
  nobody is touching right now and in money parked rather than lost. Three ways
  to close it: hire, stretch the cadence, or let an agent carry the asking.
- [The goal is crystal clear and the process does not exist: how it shows before the meeting](https://studiochat.io/blog/el-problema-sin-dimensionar):
  many companies know exactly what result they want from an AI agent and have
  never measured the problem. The problem is an area: cases a month by minutes a
  case. Interactive: a rectangle the reader sizes with their own numbers, where
  turning off a side they have not measured draws a second dashed rectangle that
  is the range of what they do not know. When the dashed one is larger than the
  filled one, the problem is not big or small, it is unsized, and measuring the
  missing side is worth more than any proposal.
- [When the channel raises its price, the question stops being how much you automate](https://studiochat.io/blog/cuando-el-canal-sube-de-precio):
  a channel that charges per conversation charges the same for a conversation
  that sells and one that handles a complaint, which is why the bill cannot be
  judged whole. Interactive: the month's conversations split into the two
  streams, showing what the channel generates before what it costs, with the
  selling half expressed as a percentage of the revenue it produces. Moving
  support to web or email cuts the bill without touching a single closed sale.
- [Not everything that can be automated is worth automating](https://studiochat.io/blog/la-friccion-que-si-queres):
  some friction is accidental and some was put there on purpose by the business.
  Interactive: an order's route with seven gates, each cycling between as it
  works today, kept on purpose, and handed to the agent. Three of the seven are
  control gates (discount, approval, credit limit): removing all three does not
  automate the process, it changes commercial policy without anybody deciding it
  in a meeting.
- [80% of your customers give you 20% of the volume, and eat your week](https://studiochat.io/blog/la-cola-larga-de-clientes):
  small customers give little revenue and almost all of the administrative work,
  because the admin work is paid per order and not per amount. Interactive: two
  bars over the same customer base, revenue and hours, which move in opposite
  directions as the reader sets their own concentration and order counts. The
  obvious way out, handing the tail to a distributor, is paid for with the right
  to decide what gets sold to those customers.
- [AI agents decouple strategy from execution, and that changes the business of anyone who lives off their experience](https://studiochat.io/blog/desacoplar-estrategia-de-ejecucion):
  written for consultants and firms whose value is knowing how to run a process
  (collections, payment reconciliation, supply chain, legal processes, small
  business operations). Strategy used to be tied to execution because executing
  required a team, and a team costs money, takes time and bends the strategy at
  every handoff. An agent can execute a well defined strategy without that team,
  but it cannot supply the experience, so strategy stays with the person. Two
  illustrative components: three routes a strategy takes to a client (the
  strategist, a team, an agent) with where it can get bent on each, and the same
  decision at three levels of definition with what an agent does at each. Closes
  with a test before delegating: hand the defined strategy to someone who never
  talked to you and see which of ten real cases they decide differently.
- [Why an in-house IT team does not replace experience with agents](https://studiochat.io/blog/equipo-de-it-no-reemplaza-experiencia-en-agentes):
  the list of things that only break in production is published nowhere,
  because it is not knowledge about the technology: it is the collision between
  the technology and how a company actually works. Five real ones, all seen in
  implementations: a follow-up counted in calendar days that goes out on a
  Sunday, a WhatsApp number blocked for abnormal sending patterns, a definition
  of what counts as a resolved task that nobody wrote down and that surfaces on
  the day of the first invoice, a customer who sends a two minute audio or a
  photo of a photo, and a routing rule by zone that assigns the lead to a branch
  with nobody there that day. An internal team learns exactly the same list, in
  a different order and with a real customer in front of it each time. What to
  ask a partner instead, and the one case where no partner is needed: a single
  channel, no integrations, no state, low volume.
- [The in-house agent competes with the product roadmap, and usually loses](https://studiochat.io/blog/el-agente-propio-compite-con-el-roadmap):
  building the agent in-house does not compete with the budget, it competes with
  what the product team was going to build. The project has an end date and the
  queue it leaves does not, so a team that reserved three sprints ends up
  reserving a slice of every sprint, forever, without having decided to. The
  reason it stays implicit: the agent enters the organization as a project and
  stays as a system, and those have different budgets. Two quarters later the
  agent loses priority to the product, which is what should happen, and an agent
  in maintenance does not break, it degrades quietly. How to do the arithmetic
  before deciding: estimate the operation as a percentage of one person per
  month, as a fixed line in the plan, and ask the committee whether that number
  is zero.
- [Building an AI agent is the cheap part: the cost shows up in month fourteen](https://studiochat.io/blog/el-costo-aparece-en-el-mes-catorce):
  a competent product team gets a prototype running in two or three weeks. What
  never ends is five fronts at once: new cases that appear in production,
  evaluations to know whether a change improved or broke something, channel
  changes (WhatsApp prices and policies Meta updates on its own), model
  deprecations on the provider's calendar, and business knowledge that goes
  stale by itself. Month fourteen and not month two because the first year is
  covered by enthusiasm: around the year mark the person who built it rotates
  away, the model version enters a deprecation schedule, and the instructions
  carry twelve months of undocumented patches. Three cases where building it
  in-house is right, and the question that settles the decision: who is on call
  the day the model version is deprecated, and with which frozen cases will they
  verify the migration.
- [Why an AI agent available 24/7 needs to know what day it is](https://studiochat.io/blog/un-agente-24-7-necesita-saber-que-dia-es-hoy):
  a reminder set to 48 hours goes out on a Sunday at ten in the morning, the
  customer answers right away, and there is nobody until Monday. Almost every
  automation counts in calendar days and almost every business runs on business
  days: two different clocks that nobody reconciles at configuration time. This
  compounds with WhatsApp's 24 hour window (documented by Meta): once it closes
  without activity, the company can only reopen with a pre-approved template,
  which is a fixed text without the context of the conversation. The five things
  to define on day one, none of them technical: real staffed hours, the actual
  holiday calendar, whether Saturdays are worked, the unit follow-ups are
  counted in, and what the agent does out of hours. Being available to answer
  and available to resolve are not the same thing.
- [What a company takes with it if it changes AI agent provider](https://studiochat.io/blog/que-se-lleva-la-empresa-si-cambia-de-proveedor):
  five assets, each with its own owner, and all five are settled when the
  contract is signed rather than when somebody wants out: the phone number and
  the account it is registered in, the conversation history, the instructions
  and knowledge base, the integrations, and the audit log plus the test cases.
  The number is the delicate one because it does not live loose: Meta's
  documentation is explicit that a WhatsApp Business account belongs to a single
  business portfolio and cannot be migrated between portfolios, while moving a
  number between providers is supported and account access can be shared with
  up to two partners. Same amount of work on day one, an enormous difference on
  the day you leave. What is honestly not portable: the integrations and the
  test runs. Four questions to ask before signing.
- [What tasks an AI agent can take on: the list is not built by department](https://studiochat.io/blog/la-lista-no-se-arma-por-area):
  the scope of an agent is defined by the kind of exchange, not by the org
  chart: any task with an exchange of conversation, files or data with a person
  qualifies, and that person is not always a customer. It can be an employee who
  needs something from the company, a supplier sending an invoice, a partner
  asking for a report. A list built by department always comes up short because
  work does not respect the reporting lines: each area answers for its own leg
  and the handoff between legs, which is the real problem, goes unnamed. The
  supplier processes that qualify (delivery date confirmations, updated price
  lists, a missing delivery note) and the exercise that builds the list in an
  afternoon: classify one real week of conversations by counterparty and by what
  was exchanged, then count.
- [Why an AI agent does not have to replace the system you already use](https://studiochat.io/blog/no-reemplazar-el-sistema-que-ya-usas):
  three sales meetings in unrelated industries ended on the same condition: we
  are not changing systems for this. An industrial plant on its time-clock
  system, a real estate firm whose agents use their own WhatsApp numbers, a
  payments fintech whose support team lives inside Salesforce. The shape of the
  answer is the same all three times: the agent does not replace the system of
  record, it sits on top as an interface layer that reads and writes through it,
  with a closed menu of allowed operations decided before it touches anything.
  Choosing what to do is the model's; deciding whether the action is allowed is
  the code's. Three real costs of not migrating, the one case where asking for a
  migration is right, and the question a buyer should ask instead of "does it
  integrate with our tools".
- [Saying the agent is an agent: when it pays off and when it does not](https://studiochat.io/blog/decir-que-el-agente-es-un-agente):
  presenting an AI agent as an agent, or as somebody from the team, changes what
  the person on the other side expects, and that expectation is set by the first
  message. An agent that declares itself earns permission to behave like what it
  is: it can say it does not know and hand off, answer in seconds without
  seeming suspicious, and state up front what it resolves and what goes to a
  person. The cost is real too: some people give less detail once they know
  there is software on the other side. What breaks when the customer finds out
  later is the handoff, because they re-read the whole conversation backwards.
  When a proper name works just as well, the two cases where disclosure is not
  up for discussion, and the concrete test: write the handoff script and read it
  out loud.
- [They asked us for an agent that prospects on its own, and the sales process did not exist yet](https://studiochat.io/blog/agente-prospeccion-sin-proceso):
  a B2B company with a decade-old product asked for an AI agent that would find
  companies, find the right contact and start the conversation. The answer to
  "how do you do this today, step by step, without the agent" was a list of
  loose actions, not a process: each salesperson does it differently, nothing is
  written down and nobody measures which step loses most prospects. An agent
  does not invent a process, it copies one, so automating it would only move the
  improvisation faster. What was recommended instead: run it by hand for a few
  weeks to find the odd cases with a person rather than with a lost customer,
  and use copilot tools (a contact finder, scraping, the existing CRM) to give
  the same person speed meanwhile. Also covers what to do when the sale cycle
  runs for months, where measuring only by closes gives no signal for a year.
- [Why telling an AI agent "don't do this" is not a security strategy](https://studiochat.io/blog/prompt-no-es-seguridad):
  AI agents are non-deterministic — a prompt instruction is a strong
  suggestion, not a barrier. Real security is deterministic: least-privilege
  permissions, guardrails in code outside the model, human-in-the-loop for
  irreversible actions, and full traceability. Grounded in the documented
  PocketOS/Cursor (April 2026) and Replit/SaaStr (July 2025) incidents.
  Available in Spanish, English and Portuguese.
- [Why a customer repeats the same problem on every support channel](https://studiochat.io/blog/un-ticket-no-es-un-cliente):
  ticket-based support systems reset context on every contact, which is why
  74% of consumers find repeating themselves to different agents extremely
  frustrating (Zendesk CX Trends 2026). The fix is a conversation that carries
  its own context across channels instead of resetting per ticket. Available
  in Spanish, English and Portuguese.
- [Why internal IT and HR support costs more than anyone measures](https://studiochat.io/blog/soporte-interno-nadie-mide):
  Internal support (IT, HR) almost never runs through a ticketing system, so
  the same repeated questions cost real time nobody tracks. Cites Gartner's
  Digital Worker Experience survey (47% struggle to find information) and
  HappySignals' Global IT Experience Benchmark (13% of incidents cause 80% of
  lost productivity). Available in Spanish, English and Portuguese.
- [Why a customer's silence after the sale isn't a good sign](https://studiochat.io/blog/el-silencio-no-es-conformidad):
  a customer who stops contacting post-sale support isn't necessarily happy —
  going silent is what unhappy customers do before churning (only 1 in 26
  complains — Kolsky/ThinkJar), and NPS/CSAT miss them because they've already
  left the sample. The fix is watching each account's contact frequency against
  its own history and reaching out proactively, since a 5% retention lift raises
  profit 25–95% (Bain & Company). Available in Spanish, English and Portuguese.
- [How to tell if an AI agent is real or just a rebranded chatbot](https://studiochat.io/blog/chatbot-con-otro-nombre):
  Gartner estimates only about 130 of thousands of vendors using the "AI agent" label
  actually have agentic capability ("agent washing," June 2025). The real difference
  isn't the model, it's whether the vendor documented the full process, exceptions
  included, instead of just an FAQ list. Available in Spanish, English and Portuguese.
- [Why the customer who uses your product the most isn't always the one who renews](https://studiochat.io/blog/uso-alto-no-es-valor):
  high usage and real value look identical on an activity dashboard. What predicts
  renewal is whether the customer reached the outcome that made them buy, not how
  often they log in; cites Gainsight's own renewal-prediction research and its
  "adoption isn't the same as value" caveat. Available in Spanish, English and
  Portuguese.
- [What the percentage of resolved conversations actually measures](https://studiochat.io/blog/metrica-de-resueltas):
  containment/deflection rate only measures whether a human got involved, not whether
  the problem went away; Gartner found only 14% of self-service issues get fully
  resolved (Aug 2024 survey of 5,728 customers). The metric that actually matters is
  the recontact rate. Available in Spanish, English and Portuguese.
- [Why a flawless demo doesn't convince a buyer who's never used AI](https://studiochat.io/blog/demo-no-convence):
  vendor marketing material ranks last among resources B2B tech buyers actually
  consult (TrustRadius 2026 Buying Disconnect Report, 74% check customer reviews
  instead); a real customer reference convinces more than another demo. Cites
  Takenos's already-published 70% autonomous resolution figure as an example.
  Available in Spanish, English and Portuguese.
- [Why an AI agent can't find an answer that's already in its knowledge base](https://studiochat.io/blog/rag-no-encuentra-respuesta):
  most wrong answers from an AI agent aren't a model failure, they're a retrieval
  failure that happened before generation. Cites Deakin University's "Seven Failure
  Points When Engineering a RAG System" (2024) and "Lost in the Middle" (Liu et al.,
  2023) as technical evidence. Available in Spanish, English and Portuguese.
- [Speaking Spanish isn't the same as serving Spanish-speaking customers](https://studiochat.io/blog/atender-en-espanol-no-es-traducir):
  75% of consumers are more likely to repurchase from a brand that supports them
  in their own language (CSA Research 2020, 8,709 consumers in 29 countries), but
  "neutral Spanish" still fails on local payment methods, tax receipts, and
  carriers, information that lives in the CRM/ERP/billing system, not the model.
  Available in Spanish, English and Portuguese.
- [Why 80% of AI projects don't pay off](https://studiochat.io/blog/ya-probamos-un-chatbot):
  RAND Corporation's meta-analysis of 65 enterprise AI projects found 80.3% failed
  to deliver the expected value; the "we already tried a chatbot" sales objection
  is almost always about the vendor training on a generic base instead of the
  team's real instructions, not about whether AI works. Cites Takenos's
  already-published 70% autonomous resolution as a counterexample. Available in
  Spanish, English and Portuguese.
- [Why an AI agent that passes the entire demo can fail with the first real customer](https://studiochat.io/blog/demo-vs-produccion):
  a demo and production test different things; benchmarks (TheAgentCompany/CMU,
  WebArena) show agents completing 8-30% of real tasks end to end, far below
  human performance, and the real-world Cursor support hallucination incident
  (The Register, April 2025) as a case study. Covers three evals that catch
  demo-to-production gaps before a customer does. Available in Spanish, English
  and Portuguese.
- [Why Latin American buyers already use AI, but don't trust your company to](https://studiochat.io/blog/conocer-ia-no-es-confiar):
  65% of Latin American consumers already use AI (Bain & Company, 2025), so the
  gap in a sales meeting is trust, not knowledge — 44% fear it spreads false
  information and 79.3% want signals of legitimacy before using an automated
  channel (WhatsApp Business with Kantar, 11,056 consumers). Argues for showing
  the control mechanism before the demo, and lays out the meeting order that
  does it. Available in Spanish, English and Portuguese.
- [What it actually means for an AI agent to integrate with your systems](https://studiochat.io/blog/leer-escribir-decidir):
  "it integrates with your tools" covers three capabilities with different risk
  profiles — reading data, writing it, or deciding with it. Projects stall
  between writing and deciding, over permission scope and approval authority,
  not over model capability. Gartner expects at least 80% of unauthorized AI
  agent transactions through 2028 to come from internal policy violations rather
  than malicious attacks. Available in Spanish, English and Portuguese.
- [Assistant, automation, copilot and agent: what each one actually means](https://studiochat.io/blog/asistente-automatizacion-copiloto-agente):
  the four words are sold as synonyms and describe four different arrangements
  of three decision rights: who decides the next step, who executes it, and who
  answers for the result. An automation follows a path drawn in advance, an
  assistant answers and the person still does the work, a copilot proposes and a
  human approves, and an agent decides and executes within a defined scope, with
  the company that deployed it answering for the outcome. Also maps the seven
  shapes a company usually already has installed (an AI chat used as a search
  engine, a project/Gem, an artifact/canvas, a routine or scheduled task, a
  licensed copilot, a flow with an AI step inside, an agent) onto those four
  patterns, using a second cut: who starts the work. The shape names are
  platform-neutral on purpose, since the same feature is called something
  different on each product. Also answers the three things buyers actually say
  ("we already have agents", "we already use AI", "we have an internal team
  building things with AI") and the three non-technical questions that tell
  which of the four a vendor demo is showing. First of an educational series
  written from StudioChat's own sales conversations.
  Available in Spanish, English and Portuguese.
- [What has to work before putting an AI agent into a process](https://studiochat.io/blog/que-tiene-que-funcionar-antes-de-un-agente):
  an agent multiplies the process it finds, it does not fix it: it clones a
  criterion, and where there is no criterion it clones improvisation and makes
  it consistent. The four non-technical things that have to be in place first
  (a named process owner who can decide judgment calls, the criteria written
  down rather than the sequence, the data queryable in a system that already
  exists, and a measured starting number), plus the three that do not need
  doing (changing the CRM/ERP/spreadsheet, documenting the whole company,
  making the process optimal), what to do when the process does not exist yet,
  and why optimization comes after ordering: half the steps of a manual process
  exist only to compensate for the limits of the human executing it.
  Available in Spanish, English and Portuguese.
- [Do you have to clean up all company data before deploying an AI agent](https://studiochat.io/blog/ordenar-los-datos-antes-de-un-agente):
  no, and the long answer is the exact minute an agent answers wrong. One
  order, one day: the real status gets written in a spreadsheet at 11:40, the
  agent reads the management system at 12:05 and answers confidently that it
  ships today. The failure is silent because there was no technical error, so
  no alert fires and the conversation is logged as resolved. The distinction
  the post draws is between messy data (which announces itself) and
  contradictory data (which does not), and what actually has to be sorted
  first: one write path for every field the agent writes, or reads in order to
  decide. Not unifying systems, not migrating history. Also when
  consolidation genuinely is the project (reporting, traceability, financial
  close) and the cost comparison between the two paths.
  Available in Spanish, English and Portuguese.
- [The shadow spreadsheet: how a backup copy becomes the real system](https://studiochat.io/blog/la-planilla-en-la-sombra):
  the twelve week route a backup spreadsheet takes to become the system of
  record, step by step (a copy in week 1, a case that lives only there in week
  3, the team's source in week 5, the weekly report in week 8, first entry
  point in week 12), and why the week somebody looks is what decides the
  price: ten minutes of conversation in week 2, days of manual work in week 6,
  a reconstruction plus lost credibility in week 12, a re-implementation if
  nobody ever looks. The four signals that show up before anyone raises it in
  a meeting, the five question script for the conversation, why banning the
  spreadsheet is the worst reflex, and the adoption metric to watch instead of
  satisfaction: the share of the process's cases that went through the system.
  Available in Spanish, English and Portuguese.
- [WhatsApp Business API pricing in 2026: what each message costs](https://studiochat.io/blog/cuanto-cuesta-la-api-de-whatsapp-business):
  WhatsApp Business API pricing from October 1, 2026: which messages are
  billed, what stays free, what a Flow costs and how to spend less on
  messages. Available in Spanish, English and Portuguese.
- [WhatsApp templates and the 24-hour window: which message you can send and what it costs](https://studiochat.io/blog/ventana-de-24-horas-whatsapp-plantillas):
  A template is not a regular message. How the WhatsApp 24-hour window works,
  when you need a template and what changed with the October pricing.
  Available in Spanish, English and Portuguese.
- [WhatsApp Business app vs API: when to switch to the API and how it's done](https://studiochat.io/blog/whatsapp-business-app-o-api):
  There are two ways to move from the WhatsApp Business app to the API. What
  changes in each, which features you lose and how StudioChat helps you
  through it. Available in Spanish, English and Portuguese.
- [How to build a WhatsApp chatbot step by step, and when to move to an agent](https://studiochat.io/blog/como-crear-un-chatbot-en-whatsapp):
  Four ways to build a WhatsApp chatbot, what each one costs, the step-by-step
  to set it up and when it makes sense to move to an AI agent connected to
  your systems. Available in Spanish, English and Portuguese.
- [How to get your WhatsApp account ready for an AI agent and connect it with StudioChat](https://studiochat.io/blog/conectar-whatsapp-a-un-agente-de-ia):
  Guide to getting your WhatsApp account ready for an AI agent: Meta
  portfolio, number, official connection, templates and tests. Plus how to
  connect with StudioChat. Available in Spanish, English and Portuguese.
- [AI agent for WhatsApp Business: what it needs to be able to do and how it connects](https://studiochat.io/blog/agente-de-ia-para-whatsapp-business):
  Before hiring an AI agent for WhatsApp Business, know what it needs to be
  able to do, how it connects with StudioChat and whether your process is
  ready. Available in Spanish, English and Portuguese.
- [AI WhatsApp chatbot: what it is, how it differs from an agent and how much it saves](https://studiochat.io/blog/chatbot-de-whatsapp-con-ia):
  A chatbot follows a script, an AI agent resolves. What each one is, how many
  inquiries it can close without a person and what it changes in your monthly
  cost. Available in Spanish, English and Portuguese.
- [WhatsApp chatbot or AI agent: six tests to tell which one you have](https://studiochat.io/blog/chatbot-o-agente-de-ia-seis-pruebas):
  Plenty of chatbots are sold as AI agents. Six tests you can run in ten
  minutes on your own WhatsApp to find out which of the two you actually have.
  Available in Spanish, English and Portuguese.
- [Meta Muse, Meta Business Agent or your own agent: which one fits and when](https://studiochat.io/blog/meta-muse-o-agente-propio):
  Meta has two AI agents: Muse for the owner and Meta Business Agent for
  talking to customers. What each one does, per Meta, and when your own agent
  makes sense. Available in Spanish, English and Portuguese.
- [AI agents for WhatsApp: how Meta, Grok Bot, Botmaker and StudioChat compare](https://studiochat.io/blog/agentes-de-ia-para-whatsapp-comparativa):
  Meta, Grok Bot, Botmaker and StudioChat serve different purposes. What each
  one is, how it is priced and which case it fits, written by one of the four.
  Available in Spanish, English and Portuguese.
- [How to choose an AI agent for customer service in Argentina: ten questions](https://studiochat.io/blog/como-elegir-un-agente-de-ia-atencion-al-cliente):
  There is no single best AI agent for everyone. Ten questions to compare
  customer service providers in Argentina, and the answers worth getting in
  writing. Available in Spanish, English and Portuguese.
- [Is an AI agent worth adding to your customer service? When yes and when no](https://studiochat.io/blog/cuando-conviene-un-agente-de-ia):
  An AI agent serves customers by text or voice, on any channel. The real
  question is when it pays off: volume, your brand's promise and your customer
  type. Available in Spanish, English and Portuguese.
- [How to qualify leads on WhatsApp and unify the conversation across all your channels](https://studiochat.io/blog/calificar-leads-por-whatsapp):
  You can qualify leads in any industry on WhatsApp. How a StudioChat agent is
  set up, how it connects to other channels, and how the conversation gets
  unified. Available in Spanish, English and Portuguese.
- [How to qualify leads with an AI agent: make the process visible before you deploy it](https://studiochat.io/blog/calificar-leads-con-un-agente-de-ia):
  An AI agent qualifies well a process that already exists and is visible.
  Which questions to ask, an example for a services company and why to
  document it first. Available in Spanish, English and Portuguese.
- [What does it mean to orchestrate AI agents across SMS, WhatsApp, web chat and voice? Four levels](https://studiochat.io/blog/que-significa-orquestar-agentes-de-ia):
  Four levels of what orchestrate means across SMS, WhatsApp, web chat and
  voice, which one your process needs, and seven questions for any vendor.
  Available in Spanish, English and Portuguese.
- [SMS, WhatsApp, web chat or voice: which channel for which conversation](https://studiochat.io/blog/sms-whatsapp-chat-o-voz):
  Which channel suits which conversation: SMS, WhatsApp, web chat or voice,
  with the limits and rules of each. Available in Spanish, English and
  Portuguese.
- [Before an AI agent texts or calls your customers: the consent rules in Australia, the US, the UK and the EU](https://studiochat.io/blog/reglas-antes-de-que-un-agente-escriba-o-llame):
  Consent, opt-out, sender registration and AI disclosure in Australia, the
  US, the UK and the EU. Sourced, not legal advice. Available in Spanish,
  English and Portuguese.
- [Voice is not chat with a microphone: what changes when the AI agent speaks](https://studiochat.io/blog/la-voz-no-es-un-chat-con-microfono):
  Why a voice agent is not a chat agent with a microphone: turn-taking,
  latency and what changes in the design. Available in Spanish, English and
  Portuguese.
- [When the AI agent hands off: what has to travel from chat to voice to a person](https://studiochat.io/blog/traspaso-a-una-persona-entre-canales):
  What has to travel with a handoff to a person when the conversation moves
  between channels. Available in Spanish, English and Portuguese.
- [Platforms for AI agents across SMS, WhatsApp, web chat and voice: Twilio, Infobip, Sinch, Intercom Fin, Genesys Cloud and StudioChat compared](https://studiochat.io/blog/comparativa-plataformas-agentes-sms-whatsapp-chat-voz):
  Twilio, Infobip, Sinch, Intercom Fin, Genesys Cloud and StudioChat compared
  on the channels each publishes, with sources and what could not be verified.
  Available in Spanish, English and Portuguese.
- RSS feed: https://studiochat.io/rss.xml

## Company

- Founders: Ignacio Puig Moreno (Industrial engineer, ITBA — founder of
  Acamica, acquired); Cristian Pereyra (Software engineer, ITBA — founder
  of Herolens, acquired, ex-CTO emblue); Iván Itzcovich (Software engineer,
  ITBA — ex-Microsoft, ex-ASAPP, AI engineer at multiple startups).
- Contact: hey@studiochat.io
- Schedule a meeting with the team: https://cal.com/studiochat
- Languages: English, Spanish, Portuguese.

## Optional

- [Spanish version](https://studiochat.io/es): same content in Spanish.
- [Portuguese version](https://studiochat.io/pt): same content in Brazilian Portuguese.