AI agents decouple strategy from execution, and that changes the business of anyone who lives off their experience
Experts who live off their experience design well and execute expensively. An agent can execute a defined strategy without a team, but can't replace its designer.
Iván Itzcovich
Co-founder
Over the last few months we've talked with a lot of people who make a living from knowing how to do something. Consultants for small and midsize businesses, for supply chain, legal processes, debt collection, payment reconciliation and mediation. Some work alone. Others built whole companies around their experience, and in every case they sell the same thing: knowing how a process gets solved.
Almost all of them have the same problem, even if they describe it differently. They enjoy strategy and don't want to go back to executing, but execution is part of solving the client's problem, because nobody pays for a plan that stays in a document. And what they end up asking for is some version of the same thing: an agent that executes what they have in their head.
I think that request points at the most underrated use of agents today. Strategy was 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 without that team. What it can't do is bring the experience, so the strategy still belongs to whoever has it. What changes is that they can now take it to many more clients.
Why were strategy and execution stuck together?
Strategy is deciding what to try and what to do. That's where experience weighs more than anywhere else, for an almost dictionary reason: having experience means having been through the situation before. Someone who has already seen what worked and what didn't, with which client and with what consequence, designs better than someone thinking about it for the first time.
Execution is the other half, the most repetitive and least attractive part of any implementation. Until now there were two ways to handle it, and agents add a third.
The strategist executes it
It arrives intact, but it has to fit in one calendar.
- Designs the strategy
- Executes it personally
- Result for the client
- points where the strategy can get bent
- 0
- What caps it
- One person's hours
- Where a deviation shows up
- Right away, because it's the same person
A team executes it
It multiplies, and it gets bent along the way.
- Designs the strategygets summarized into a proposal
- Project plangets prioritized against other work
- Team coordinationeach person interprets it
- People executingwhoever understood it moves on
- Result for the client
- points where the strategy can get bent
- 4
- What caps it
- What it costs to hire and coordinate people
- Where a deviation shows up
- In the result, weeks later
An agent executes it
It multiplies with a single handoff.
- Designs the strategywhatever wasn't defined, or was left ambiguous
- Instructions with the defined strategy
- Agent connected to the systems
- Result for the client
- points where the strategy can get bent
- 1
- What caps it
- How well defined the strategy is
- Where a deviation shows up
- In the conversations, and it gets fixed in a text
Illustrative route: the number of steps changes with every project. The number under each column counts its red arrows, the points where the strategy can get bent.
The first is for the strategist to execute it. It works on small projects and the strategy arrives intact, but the business is capped by one person's hours.
The second is building a team. That multiplies capacity and brings what any team brings: salaries, coordination, months before it gets going, and the risk that what gets executed is no longer what was designed. If the implementation is also technical, add development, integrations and timelines of their own. With that math, plenty of good strategies never left the document.
What changes when an agent executes it?
An agent is good at executing instructions. It takes instructions, connects to the systems where the process lives (the CRM, the ERP, the WhatsApp channel), resolves the cases that come in and logs what it did. It doesn't get moved to another project, and it doesn't have rough weeks.
For whoever designs the strategy, that changes the math. The capacity to execute stops depending on how many people can be hired and coordinated, and starts depending on how well defined the strategy is.
Where the strategy gets lost changes too. With a team, the deviation shows up in the result, weeks later, and reconstructing which handoff it happened in is archaeology. With an agent, the deviation shows up in the conversations and gets fixed in a text. The agent makes mistakes too, but when it does, the first thing you check is an instruction, and the fix stays in the instructions for next time. With real cases saved as evaluations you can also verify that the fix didn't break something else (I went into more detail in the cost that arrives in month fourteen).
Until now, a strategy was only as good as the team that could execute it. With agents, it's only as good as its definition.
Why not ask the agent for the strategy too?
Because it's the part it struggles with most today.
Caught up in the moment, many companies use agents to replicate the whole package: have the model figure out what to do and then do it. A model can propose a strategy in seconds, and it will sound reasonable. Reasonable isn't proven. It lacks exactly what defines experience: having seen the outcome of those decisions in a specific market, with a specific kind of client, and having corrected course afterwards. And it struggles with the creative part, the one that departs from what any playbook says because someone has already watched the playbook fail.
Asking an agent for the strategy when there's someone on the other side who already knows how to solve the process misuses both. The agent does the part it's worst at, and the person gets left out of the one part where they can't be replaced.
How defined does a strategy need to be before you can delegate it?
The more defined it is, the more you can delegate. The same decision can be defined at very different levels of detail, and the difference shows up in how an agent executes it.
The same decision, at three levels of definition
Undefined
“If someone always paid and falls behind, don't push them.”
What the agent does: It knows what not to do, but not what to do instead. It ends up writing to them like any other debtor.
Half defined
“With good payers, a softer tone.”
What the agent does: It has to guess who counts as a good payer, and it guesses differently in every conversation.
Defined as a criterion
“If they paid on time for the last 6 months and are less than 15 days late: a friendly reminder, with no mention of arrears or payment plans. If there's no reply within 5 days, they move to the standard flow.”
What the agent does: It has a rule it can apply the same way in conversation one and in conversation one thousand. If it fails, you can see which part.
Undefined
“Those small differences aren't worth chasing.”
What the agent does: It doesn't know whether “not chasing” means applying the payment, leaving it pending or telling someone.
Half defined
“If the difference is minimal and the client is known, apply it.”
What the agent does: “Minimal” and “known” are left to its interpretation, and not always the same one.
Defined as a criterion
“If the difference is under 1% of the amount and the client has more than 12 reconciled payments: apply it and notify them by email. Otherwise, ask for the receipt before applying it.”
What the agent does: It applies what fits the criterion, asks for a receipt for the rest and logs why it did each thing.
Undefined
“When the usual supplier runs late, you have to move before running out of stock.”
What the agent does: “Move” could mean flagging it, requesting a quote or switching suppliers, and it doesn't know which or when.
Half defined
“If the supplier doesn't confirm in time, get a quote from another one.”
What the agent does: “In time” changes from one order to the next, and so does which supplier it asks.
Defined as a criterion
“If the main supplier doesn't confirm the order within 48 hours and stock covers less than 10 days of sales: request quotes from the two alternates and send all three options to purchasing.”
What the agent does: It requests the quotes and leaves the decision to purchasing, with the numbers on the table.
Illustrative examples: the thresholds are not a recommendation for any business. What matters is how much the agent's behavior changes between the first level of definition and the third.
Between the first level and the third there's no extra experience. There's the same experience turned into criteria: thresholds, exceptions, what to do when no condition is met and who the case gets handed to. Defining that can't be delegated, and for anyone who lives off what they know it's the interesting part, because it's still strategy. Just defined in enough detail for someone else to execute it.
What's left for the person who knows how to solve the process?
Everything that was always theirs, without the ceiling execution put on it. The consultant who took on few clients because there weren't enough hours, or who turned down big projects because they didn't want to become a team manager, can multiply what they know without changing trades. What does change is where the time goes: less executing, more defining criteria and watching results to adjust the strategy when the market moves.
A test before delegating
Hand the strategy, as you defined it, to someone who has never talked to you and ask them to decide ten real cases. Every case they resolve differently from how you would marks something that isn't defined yet. That's exactly what the agent will be missing.
For a long time, knowing how to solve a process and being able to solve it for many clients were two different businesses, and the second one needed a team. Not necessarily anymore.
Frequently asked questions
What does decoupling strategy from execution mean?
That whoever designs how to solve a process can put it into practice without depending on a team to execute it. The strategy still belongs to the person with the experience. An agent does the execution based on that well defined strategy, connected to the systems where the process lives.
Can an AI agent design the strategy for a process?
It can propose one that sounds reasonable, but it lacks what makes a strategy valuable: having seen what worked and what didn't in that market and with that kind of client. It's far more useful executing the strategy of someone who already has that evidence.
What does a consultant need to have defined before delegating execution to an agent?
Criteria instead of intuitions: concrete thresholds, what to do in each known exception, what to do when no criterion applies and who the case gets handed to. A good test is to give that definition to someone who has never talked to the consultant and see which cases they decide differently.
Iván Itzcovich · Co-founder, StudioChat
Want agents like these working for your team?
Talk to us now