Complete archive of published writing across all domains.
For a month my local validation and the competition leaderboard disagreed about whether I was improving at all. They were computing two different statistics. Here is how I found out, and the one-submission experiment that fixed it.
A summary of the arc so far from inside the ARC Prize 2026: the three failures that taught me how the competition actually works, and why the plain agent still beats every cleverer thing I have built. Almost none of it was about the model.
For fifty years software executed instructions. Cognitive systems pursue objectives instead: they perceive, plan, act, verify, and adapt under uncertainty. Here is what actually separates a cognitive system from a traditional application, and from bolting an LLM onto one.
An ARC Prize 2026 builder's log entry. I spent two work sessions reasoning from two "facts" about my own code that were both false. Checking them against the artifact took an hour. The highest-risk sentence in any project is "we can't, because."
Inference got dramatically cheaper, yet enterprise AI bills keep climbing. The token tax is not a pricing problem, it is an architecture problem, and the missing middle layer is quietly becoming enterprise debt.
Most AI systems fail not because the model is weak but because we wrap probabilistic cognition in architecture built for deterministic machines. Cognitive orchestration is the layer that manages the uncertainty: synthesized state, adaptive planning, verification, and persistent memory.
An ARC Prize 2026 builder's log entry. I taught my agent to solve more kinds of games and the leaderboard score fell from 0.09 to 0.04. The capability was real. The cost of using it was the problem. More powerful is not better.
Bigger context windows were supposed to solve AI. Instead they exposed the real bottleneck: not the model, but the architecture around it. Why the monolithic era of AI is giving way to orchestration.
An ARC Prize 2026 builder's log entry. A change that gained on my home bench lost three times as much on the real, hidden leaderboard, and my own tooling couldn't even reproduce the failure. Why the leaderboard is the only oracle.
Companies that cut jobs for AI are quietly rehiring, and most of them didn't even save money doing it. The AI boomerang, and what it reveals about the real price of human judgment.
A new thread in the builder's log: I'm competing in the ARC Prize 2026 (ARC-AGI-3), the benchmark built to measure the one thing today's AI is worst at, learning something it was never trained for. What the comp is, why it's the field's uncomfortable mirror, what it would mean to crack it, and where I'm standing.
Most companies think they have a sales problem. They have a strategy problem in disguise. A decade of auditing startups and OEM manufacturers, and the pattern is always the same: activity went up, revenue did not, because no system connected market demand to the sale.
