Notes for autumn 2026: what we are testing, carefully
Two betas open, five things in testing, three things we are deliberately not doing. A plain account of where the work is this autumn, with no promises attached.
The trace view, one year on
We shipped the trace view expecting associates to check every step. They open it in one query in five, and almost always for the same reason. What we expected, what happened and what we changed.
Why we onboard firms in cohorts (and sometimes say not yet)
Firms sometimes ask to start on a Tuesday and are told the next cohort opens in three weeks. This is why, what a cohort actually does, and the three reasons we occasionally decline altogether.
On not announcing things
Most of what we shipped in the past year arrived without an announcement. That was deliberate. A short account of why legal software should change quietly, and what quiet costs.
A voice agent that is actually built for law: what jurisdiction-aware has to mean
Most voice agents treat jurisdiction as a language setting. For legal intake it has to govern what the agent asks, what it computes and what it refuses to say.
2025, told quietly: what shipped, what didn’t, what we learned
Ten version numbers, three general availabilities, one model. A short account of the year, with the slips left in.
Why we bet on a small model for contracts
Latency, residency, reproducibility and the sealed laptop: the four constraints that made a 7-billion-parameter model the right size for clause-level work, and what we gave up.
A 500-millisecond budget: why review latency is a product decision
Inline review either keeps pace with drafting or it interrupts it. We set a hard budget of half a second at the 95th percentile, and most of our architecture follows from it.
Why every flag needs a citation
A flag without a reason cannot be checked, cannot be disagreed with and cannot be defended later. The audit argument for citations is stronger than the usability argument, and it is the one that matters.
340 firms, 28 countries: what changed in a year, and what didn’t
A year ago we had just over a hundred accounts. Now there are 340 in 28 countries. Some of what we expected to happen did. Much of it didn’t, and that is the more useful list.