The hardest button to build wasn't the report generator — it was the one that says "I don't know"

Quick build note for anyone following along.

When I started AnchorStrategy, I assumed the hard part would be getting the AI to produce good strategic analysis — SWOT, Porter's Five Forces, that kind of thing. It wasn't. Models are already pretty good at that.

The hard part was building the opposite: a system that's willing to stop and say "insufficient basis" instead of filling a gap with a plausible-sounding number. Every instinct in how these models are trained pushes toward completing the pattern, not admitting it doesn't have enough to go on. Getting the guardrails to actually catch that — reliably, not just in the demo — took longer than the entire rest of the analysis pipeline combined.

Small proof it's working: I posted about this on IH last week, and the comments that got the most traction weren't about the frameworks or the output format. They were specifically about that boundary — people recognizing that "confident but wrong" is a worse failure mode than "honest but incomplete." One conversation from that thread is now turning into an actual business discussion.

Still in private beta, still learning what people actually want to pay for. More updates as I figure that out.