The bottleneck moved.
Cheap building makes bad asks more expensive.
The work looks trivial, so the “yes” feels free. The tail is not.
expensive build
cheap yes
feels free
But every tiny yes creates a surface area: edge cases, expectations, politics, support, and future asks.
Scope
Maintenance
Stakeholders
Dependency
Communication
Five ways the easy yes compounds.
Scope creep
Maintenance tail
Stakeholder surprise
De facto employee
Communication ceiling
Replace the easy yes with a boundary system.
“Sure, I can do that.”
Momentum without visibility.
“Yes, and that’s phase two.”
Acknowledges the idea. Protects the scope.
“Here’s who owns it.”
Turns delivery into handoff.
Scope the promise, not the task list.
fast
Get read-only access. Validate what is already in their CRM, then show the output. An afternoon, nothing to run.
clean
The owned, monitored, documented build. Only after they have seen it work.
Price the tail before it becomes a favor.
Build
Ship
API changes
Prompt drifts
They call you
Map the people before you automate the work.
A tool changes someone’s job. If that person was not in the scope, they arrive later as rework.
Ask: who else has to say yes before this is done?
Treat communication as the scarce resource.
Not build time. Attention.
One request path. One cadence. One owner.
Faster delivery does not make the outcome worth less.
Design the exit before you say yes.
Done means someone else can run it.
Three moves that keep AI work bounded.
Scope the promise.
Say whether this is learning, production, or an operating commitment.
Name the tail.
Maintenance, support, approvals, communication, and monitoring are real work.
Design the exit.
Done means docs, walkthrough, owner, and a clean handoff.
If you cannot answer them clearly, you are not looking at a quick task. You are looking at unscoped ownership.