True North: Outcome-maxxing, judgment capture, and agent guardrails

Connected Stack is back. The most influential enterprise AI founders and firms rarely share a room. On September 10 in San Francisco , we're bringing back Connected Stack — the room for the builders, buyers, and backers defining the future of enterprise AI. Join us →

Signal

Let's try outcome-maxxing. The win isn't tokens burned — it's value per token. First find the thing worth burning tokens on; spend thousands a day to crack it if you must, then drop to a cheaper model once it's working. Not everything needs Fast and Max.

Price to the human, not the seat. Vertical AI agents don't price like software. They price like the role they replace. When your agent charges a salary, the ceiling stops being the software budget and becomes the wage bill — often 10 to 50x larger. The founders pulling this off aren't selling a tool: they're selling the job and its ROI.

If it's repeatable, it belongs in an AI loop. All repeatable work should loop. Stop thinking of “members of technical staff” and ask where human motivation and creativity are actually necessary and truly deserving of human ingenuity. The humans you add should amplify creativity and revenue.

The compounding asset is captured judgment — not captured knowledge. Don't just “document” — record the decisions that shaped the work, the constraints that mattered, and the specific ways the loop failed. Facts are portable; taste and failure logs are earned. That's a moat that compounds because it's the byproduct of running the system longer than anyone else.

True Team & Network

150 founders and builders. We hosted our AI Workflow Summit last week to showcase the flows that are actually working for those at the frontier. Check out our recap →

Monthly AI spend per team member ~equivalent to a generous coffee budget. Our latest portfolio AI Reality Check survey results are in. Nearly half of teams spend $25–250 per team member per month on AI. One in five founders spend $1,000+ per team member per month.

Founders Building

Sovereign intelligence from space. It started with a prototype on a rooftop, a white tent, and a team waiting for the ISS to clear the horizon. It now looks like a €1B Series F funding round at a €10B valuation. ICEYE continues to take off →

Every LLM needs a layer that says no. We backed Ari Jacoby and Todd Lieberman before, and we just did it again at Concentrate AI — the governance layer that decides what your models are actually allowed to do. Adam on why these founders are worth betting on twice →

Your agents are wrong more often than they let on. Chips and models hog the headlines, but the unglamorous behavioral assurance layer is what quietly decides whether enterprise agents survive contact with production. Puneet shares why we backed ChatSee →

Better robots don't have to mean bigger machines. Surgical robotics spent years assuming more capability meant more hardware. Channel Robotics put the intelligence in the surgeon's hands instead — same skill, a lot less machine. More from Rohit →

Open Tabs

Tab 1 The “oozy” look of consumer brands right now. Read more →

Tab 2 Evals: the strategic IP that will define the next era of AI. Read more →