What we are researching. What we shipped. What failed.
Monthly, no marketing.
We do not write to fill a schedule. Each issue is sent when we have something worth saying. Usually once a month, sometimes more, sometimes less.
Summaries of completed experiments with methodology, results, and honest assessment of what worked and what did not.
What we are running right now: hypothesis, current state, early observations. We write these before we know the outcome.
First notice of new repositories, major updates to existing tools, and deprecations. With the reasoning behind each.
Where we are presenting next, what we are planning to say, and links to slides and recordings after the event.
When we have something long to say that does not fit a standard article format — a detailed post-mortem, a technical tutorial, a research design document.
We document experiments that did not work. These are sometimes the most useful issues — a documented failure prevents someone else from repeating the same mistake.
A sample of what the last few issues covered. The full archive is available to subscribers.
Issue #001 through #019 available in the archive after subscribing.
“The Labs Letter is the only AI newsletter I read where the author admits when something didn't work. That alone makes it worth subscribing.”
“I've implemented three patterns from the Labs research directly in our production systems. The observability middleware alone saved us hours of debugging.”
“Short, dense, no padding. Every issue has something I can use in the week it arrives. That's rare.”
Research, experiments, releases, and failures — once a month, directly from the engineers building and studying AI systems in production.
We never sell your email. One email per month. Unsubscribe anytime — no dark patterns, no re-subscribe tricks.