I was doing some digital housekeeping last week when I got to my Claude conversation history. Scrolling through months of prompts, I realized I wasn’t just using AI to get things done—I was accidentally prototyping products.
This might be hyperbole, but consider the anatomy of any AI prompt you’ve written recently. You’ve likely structured it around five key elements:
Context (“I’m a venture capitalist reviewing deal flow…”)
Job (“Help me analyze this startup’s competitive positioning…”)
Constraints (“…using only public information and focusing on Southeast Asia…”)
Output (“…in a structured memo format…”), and
Success criteria (“…that I can share with my investment committee.”).
This isn’t just good prompting—it’s product specification. You’ve defined a user, use case, scope, format, and success metrics. That’s a product brief.
Take the fintech startup I was evaluating last month. Information was scattered everywhere—email threads with the founder in Gmail, market research and pitch deck in Drive, notes from the pitch meeting in Evernote, Slack conversations with other team members, deal tracking data in Airtable. Everything sort of glued together using Zapier. Instead of manually hunting through each system, I asked Claude to connect to all my data sources and build me a comprehensive analysis. The result wasn’t just better than what I could have assembled manually—it was a completely different class of analysis, output and speed. The recently launched (recent is relative; probably lightyears as measured in AI years by now!) Claude Connectors totally changed the game for me.
Clayton Christensen’s jobs-to-be-done theory becomes incredibly literal in the AI era: every prompt is literally hiring an AI worker to complete a task. What makes this particularly interesting for founders is that AI interactions generate unprecedented visibility into the jobs people actually need done. Your prompt history becomes market research. Your workflow optimisations become product features.
The most valuable prompts aren’t just instructions—they’re crystallised domain expertise, refined through iteration. A well-crafted prompt for investment analysis represents years of experience in deal evaluation and risk analysis. This expertise, once encoded in effective prompts, becomes scalable, transferable, and often more consistent than human execution. We’re moving toward what I’m calling the “prompt economy”—an ecosystem where the ability to effectively instruct AI becomes a core competitive advantage, and where the most valuable prompts become products themselves.
What jobs are you hiring AI to do? Take a look at your prompt history—your next startup idea might already be there.
Leave a Reply