The Service-Led Diagnostic Agent.

The Hidden Killer: The Fact-Find

In professional services, "Discovery" is often just a polite term for a 2-hour interrogation. You ask the same questions ("What's your budget?", "Who are your competitors?") to filter clients.

I decided to automate myself out of this loop by building a Google Gemini Gem to handle the entire intake process—not just collecting data, but *reasoning* about it.

The Logic: Integrity Testing

The challenge wasn't just making it chat; it was making it *honest*. The first version was a sycophant that agreed with everything. I had to build an "Integrity Gate" to force the agent to reject bad fits.

The Integrity Gate: Hard-coding ethics into the logic flow.

I taught the system three core behaviors:

• 1. Sequential Inquiry
  Don't wall-of-text the user. Ask one thing, wait, then adapt.
• 2. Skepticism
  If a user says "I don't know my revenue," provide a base rate estimate instead of accepting ignorance.
• 3. Bedside Manner
  Contextualize scary financial terms (COGS, OpEx) so creatives don't feel alienated.

The Velocity: 24 Hours

Traditional development (React/Typeform) would take 5 days. With AI, I compressed the R&D cycle into a single day.

The Output: Structured Intelligence

The agent doesn't just chat; it produces a tangible deliverable. A structured Confidential Analyst Brief that outlines the client's current state and gaps.

The goal isn't a chat. The goal is a structured deliverable.