Notes from the team.
On custom AI, the gap between capability and adoption, and what actually moves the needle for businesses.
The maintenance is the work — why AI is the first system that drifts under you
Most software is stable until you change it. AI systems drift on their own — models update, data shifts, prompts that worked six months ago start missing edge cases. The maintenance phase is where the value compounds, and skipping it is how AI projects quietly stop working.
Read →Inference got cheap — which AI projects just unlocked
The cost of running a frontier model dropped roughly an order of magnitude in 2025 alone. A whole category of work that used to be too expensive to automate is now economical on a Tuesday afternoon. Here is which use cases just crossed the line.
Read →The 1M-context window changes the shape of audit work
Frontier models now hold a million tokens of context — roughly a quarter of business emails, an entire codebase, or a multi-year document archive in a single conversation. The interesting question is not what is now possible. It is what stopped requiring a pipeline.
Read →The gap between AI capability and your business — and who closes it
Every week another frontier AI capability ships. Very few of them end up inside a business. The gap between what the models can do and what your team actually uses is where the next decade of advantage gets built.
Read →