Q10Leadership

Q10. Product wants autonomous agents, Sales wants a demo next week. What do you do?

How I handle competing pressure from Product and Sales on an AI agent feature: a guided flow with human approval in two weeks, evals from real usage next, and autonomy only where accuracy proves it.

Rajendra Kumar · · 2 min read

Q10. Product wants autonomous agents, Sales wants a demo next week. What do you do?

I don't pick a side. I break the work into phases so nobody is blocked and we never ship something that might misbehave in front of a customer.

Three phases: weeks 1 to 2 ship a guided flow with an approve button so Sales has a demo; weeks 3 to 4 build evals from real traces so the golden set grows; after that, remove approval gates per action type where accuracy is proven.
Architecture · Each phase earns the next one with real usage data.

The phases

  • Weeks 1–2: guided version. A fixed workflow with a human approval button on every action. Sales gets a working, reliable demo right away, and we start collecting real usage traces.
  • Weeks 3–4: harden. Use those traces to fix edge cases and build the automated eval suite.
  • After that: autonomy, one action at a time. Remove the approval gate only for the specific action types where the numbers show consistently high accuracy. Risky actions, like money movement, may keep a human forever, and that's fine.

Why everyone ends up happy

Sales gets a demo in two weeks. Product gets the autonomous feature, built on evidence instead of hope. And engineering doesn't ship an unpredictable liability. The approval button isn't a weakness in a demo either. For finance and operations buyers it's usually a selling point.

Key takeaways

  1. 01A guided flow with approvals is a great first demo.
  2. 02Real traces become your test suite.
  3. 03Autonomy is earned per action type.
Product thinkingAI agentsStakeholder managementDeliveryInterview prep