A publication for the builders.
Alchemyst Press exists to make hard-won engineering knowledge common ground — told by the people who were there, in enough detail to be useful.
Most writing about AI describes what a system is supposed to do. The interesting part is what happens after: what should an agent remember, what should it forget, who can it trust, and who was on call when it broke. Those questions begin where the demo ends, and they are answered by people rather than benchmarks.
How we work
Firsthand, not secondhand
We publish what someone actually did: the decision, the constraint, the failure and the fix. Anyone can summarise a release note. Very few people can tell you why the system fell over at 3am.
Leave the rough edges in
A story that resolves too neatly is usually missing its most useful part. We keep the trade-offs visible, including the ones that did not work out.
Context over conclusions
A model sees tokens. A system needs context. We care about the memory, permissions and knowledge that hold a system together, not just the benchmark it scored.
Common ground
One engineer's hard problem should become another's starting point. Every piece is written to be read once for the story and kept for the detail.
What we publish
Conversations
Long-form interviews with the people who built and operate the system.
Articles
Reported and edited pieces that follow a single problem to its end.
Field manuals
Practical, structured write-ups meant to be used while you work.
If you have a story that fits, we would like to hear it. Read the publications to see the shape of the work, then get in touch.