lemon Lesson 5 of 7 · 4 minutes

Watch the work

Let the agent move while keeping scope changes, risky actions, and surprises visible.

Once the brief and plan are clear, let the agent work. You do not need to approve every file read or harmless check. Constant supervision turns the agent back into an autocomplete tool.

Instead, agree on the moments that require a stop.

Stop conditions

The agent should pause when it encounters:

  • A genuine design fork you have not chosen.
  • A need to expand beyond the agreed scope.
  • A destructive or difficult-to-reverse action.
  • Credentials, private data, paid calls, or production systems.
  • A check that cannot run or a result that contradicts the plan.

Everything else can usually continue through the agreed loop.

Surprises become follow-up work

Agents often notice nearby problems while reading broadly. That can be useful, but discovery is not permission. Ask the agent to record unrelated findings and finish the current task without folding them into the diff.

This gives you small results that are easy to review and a separate list of possible next tasks. It also prevents a two-line fix from becoming a dependency upgrade, a refactor, and a new architecture.

Autonomy is not “do anything.” It is freedom to move inside a boundary that both sides understand.

Try it

Make the call

Pick an option to see how the rest of the session plays out.

you

Approved: fix the broken label on the contact form. Just that.

agent

On it. While reading the form code I noticed the form library is two years out of date.

Mid-task, the agent notices an unrelated dependency is outdated. What is the best move?

agent

I fixed the label and upgraded the library while I was in there. The upgrade changed how validation behaves, so I also adjusted three other forms.

you

The label fix I can check. The rest is a second project I never approved, tangled into the first one.

not yet

The task just gained a new risk surface and a new definition of done. Nearby is not the same as necessary. AF-15

agent

Label fixed: one file, here is the diff. Separately, I wrote down what the library upgrade would involve. Want that as its own task?

you

Yes. Today’s change I can review in a minute, and the upgrade gets its own session with its own checks.

good call

The current result stays reviewable, while the useful discovery is preserved for a task that can be scoped on its own.

you

There’s a security notice going around about our form library. Did we know it was out of date?

agent

I noticed it during the label fix a few weeks ago, but I didn’t record it anywhere.

close, but

Avoiding scope creep is good; losing a potentially useful finding is not. Capture it without acting on it.

Take it to your tool

Try this in your agent

With the project open in your coding agent, paste:

Carry out the approved plan. Stay inside the agreed scope. If you discover a larger problem, a destructive action, or a genuine design fork, stop and ask instead of expanding the task.
Got the idea?

Mark it complete on this device. You can return or reset the course at any time.