Deal Timing & AI Workforce

This call was different from previous sessions. There was no set topic. The first twenty-five minutes were spent on lie-flat airline seats, round-the-world ticket hacking, 3D printing, and whether Bamboo Labs printers are worth buying.

None of that belongs in a synthesis document about founder lessons, except that it does. This is what it looks like when a founder peer group starts functioning as an actual peer group rather than a structured meeting. The banter is the relationship infrastructure that makes the harder conversations possible. When one founder shared devastating news about a deal collapse minutes later, the trust was already there. That doesn't happen in a group that skips the 3D printers and goes straight to the agenda.

The substantive threads that emerged are below, but the session itself is evidence of something the group has been circling for months: founders need spaces where they can just be people first and founders second.

One founder had been in due diligence with the Yale Endowment for ten months. On Friday, they said no. The reason given: his references didn't provide enough detail about how he executes the strategy day-to-day.

This was the most exhaustive DD process he'd ever experienced: fourteen references, every conceivable document, no objections to his track record (122 deals, 10 years, never lost money). The rejection came down to: the Yale analyst forgot what the strategy was by the time he was calling references, and therefore didn't ask the right questions.

The Takeaway: Prep the Questioner, Not Just the References

His self-assessment was specific and actionable: he should have asked Yale what they wanted to hear from his references. He'd never had a bad reference in his career, so he assumed the process would take care of itself. It didn't.

The practice going forward: When someone asks for references, ask them explicitly what they're looking to learn. Then prep your references to address those specific questions. People forget — even people who work alongside you daily will forget the time you solved a critical problem.

The Deeper Lesson: Time Kills Deals

The group converged on a principle: extended timelines almost always indicate something else is going on beneath the surface. Extended DD creates the illusion of progress. Every month feels like you're just around the corner.

After two to three months, if it hasn't closed, you should step back and let them come to you. Continuing to push past that point starts to look like desperation.

The Information Inflection Point

There's a point in any pitch where you've said everything that needs saying. If the person is going to buy, they're going to buy. Anything you add past that point actually increases the chance they won't — it creates new objections, introduces complexity, or makes you look like you're trying too hard.

Knowing where that crossover lives, and stopping just before it, separates effective selling from overselling.

The Argument

The finance founder offered a framing that stuck with the group: AI is trained on all human knowledge. Therefore, by construction, it represents the average human. It will make decisions no better and no worse than the average informed person.

In finance, this is a fatal flaw. The price the market shows you on any given day is, roughly, the consensus of all informed participants — what the average qualified person thinks the price should be. If you trade on what AI tells you, you're trading on market consensus. You're buying at market price. That's not how you make money.

Why He's Still Hiring Humans

The same founder had started the year determined not to hire anyone. He's now accepted that he's going to keep hiring. Three reasons:

1. The context-switching problem. Managing multiple projects simultaneously, each with its own context window and accumulated decisions, is draining. Handing a project to a person and saying "own this, build it out" is still more efficient than managing that context across several AI sessions.

2. The agency problem. He doesn't want someone (or something) that does exactly what he says. He wants someone who can take a direction, internalize it, and make independent decisions about how to execute. AI can build what you tell it to build. It doesn't yet reliably decide what to build next when you're not looking.

3. Human labor is cheap. When employees stay, the ongoing cost is low relative to the value they generate. The expensive part is when they leave and you have to rebuild the institutional knowledge.

Where AI Does Replace Work

There are parts of the workflow where AI has genuinely replaced human effort: QA and testing, requirements documentation, routine coding tasks. The whole first-pass layer of the software development lifecycle is compressible now.

What remains is the person who understands how systems need to run at scale, who can have a conversation about what actually needs to be built, and who can exercise judgment about tradeoffs that the AI doesn't know to surface.

"A person with AI is a much better investment than either a person without AI or AI without a person."

The Value Corollary

If AI can do it, it will cease to have value. By definition, anything that can be fully automated becomes commodity-priced. The things that generate value are, by construction, the things that require something AI doesn't provide. That category shifts over time, but it doesn't disappear.

This thread doesn't have a clean business lesson, but it illustrates something the group keeps returning to: the value of building physical things as a counterpoint to building software.

Two founders are now actively 3D printing. One is building custom toys with his seven-year-old, using AI to convert children's drawings into printable 3D models through a software pipeline (sketch → SVG → CAD extrusion → STL → slicer). He has no CAD background. The whole pipeline was built by asking ChatGPT and Claude to write the conversion software.

The other is printing components for robotics projects, sourcing open-source drone kits, and exploring how to connect AI agents to microcontrollers.

The broader observation: 3D printing sits at the intersection of several themes the group has been tracking — AI-assisted design, distributed manufacturing, physical-digital product hybrids, and the collapsing barrier to entry for hardware. It's also one of those areas where the learning is genuinely fun, which connects back to the stress management discussion: find the thing that gives you a complete context switch.