Notes, essays, and research from the work.
What we're thinking and learning building agentic AI and robotics into the industries the world depends on — written as we go, not after the fact.
- Playbook2026.08.24
Re-writing the data-driven playbook for 2026
Give a smart new hire full access to your systems and nobody to ask — how far do they get? That's roughly how far your first agent gets, except the new hire tells you when they're confused. What AI actually needs from your data that BI never did, why an unopenable vendor is a procurement problem rather than a data problem, and a 30-day sequence that ends with something you can hand a builder.
- Essay2026.08.17
The five phases of AI adoption
A map of the route from zero to AI native — tools, experiments, systems, infrastructure, native — and why the value is in the transitions, not the phases. Most companies are one phase behind where they think they are, and every phase has its own trap, its own failure mode, and its own exit signal.
- Field Note2026.08.17
Why your AI pilot never shipped
The pilot worked. It demoed well, people liked it, and a year later it's still a pilot. Five reasons that happens — four of them not technical — plus how to tell which one killed yours, and what a salvage actually looks like when the prototype is still sitting there working.
- Playbook2026.08.17
7 things to do before your first AI build
Not a checklist — a funnel. Generate candidate jobs from your goals and the constraints blocking them, work out what harness each would need, spot-check which ones you can actually reach in your systems of record, qualify the survivors on evaluability and a real metric, and only then design the one that's left. Carried end to end through one worked example, including what gets deferred, what gets disqualified, and the difference.
- Essay2026.08.17
Buy, build, or partner
The resourcing question every company hits once it has decided AI matters — and why it's the wrong question as asked. It isn't one decision, it's a decision per layer of the stack. An honest account of what each path is actually good at, what each one costs, and the cases where you shouldn't hire a partner at all.
- Playbook2026.08.11
Where AI actually pays in manufacturing
Real-world AI use cases in manufacturing, ranked by time-to-value — not the robot-arm fantasy, but the paperwork around the plant: PO chasing, downtime investigations, quality reporting, supplier exposure, and the tribal knowledge walking out the door. What each one looks like today, what an agent changes, and what it needs to connect to.
- Research Note2026.08.11
Ego, exo, and the robot data engine
The most important video for robotics isn't of robots — it's of human hands. A close look at how the field turns egocentric and exocentric human video into robot policies: the capture spectrum from paired rigs to handheld grippers, the three pipeline families (retarget, repaint, latent actions), and what the mid-2026 results say actually matters.
- Research Note2026.08.11
How robots learn now
A field guide to the leading edge of AI in robotics, mid-2026: the three-layer stack behind the new generation of robots, the end-to-end pipeline that turns internet knowledge and human video into motor control, and the roadblocks — data, evaluation, and the last nines of reliability — the whole industry is working against.
- Field Note2026.08.11
Three agents, one job
The most common thing we find inside companies adopting AI isn't failure — it's duplication. Multiple agents built independently for the same job, each one a success story, none of them the system. Call it agent sprawl: experimentation that never gets consolidated, and the feeling of progress without the results.
- Essay2026.06.30
Every company needs an AI team
Most companies are standing up an AI function — and getting it wrong, either as a silo that hoards AI or a figurehead who ships nothing. The version that works is a real team whose job is to make AI a capability every other role wields.
- Essay2026.06.29
Your company is becoming an optimizer
As every business becomes a full-stack AI firm, the edge won't come from the tools. It comes from running the organization as one coherent, instrumented system aimed at a clear objective — because at machine speed, coherence and correct aim compound, and incoherence compounds against you.
- Product2026.06.29
We built Damon for the factory floor, not the inbox
Almost every AI agent being built today is pointed at knowledge work. Damon goes the other way — an agentic operating system for manufacturing that meets a plant wherever it is, from pen-and-paper and spreadsheets to a dozen systems that don't talk, and grows it into one intelligence layer that compounds.