We’d been seeking “fully autonomous production systems” long before Claude Code ever came out. I knew it would happen someday - I just didn’t think that the shift would feel so imperceptibly natural.
Sometimes it’s easy to forget how different a day with my laptop is today than it was back then.
We ran an experiment. We pointed Antimetal at our own production environment, emitting 3,000 error logs a day, with no prior guidance.
It cut our error volume by 76% in a single day.
It found six unique problems, including one we'd misdiagnosed for weeks. It diagnosed and triaged each one, resolved it, and monitored the fix to confirm stability.
Production that runs itself is no longer hypothetical.
https://t.co/JGpd3N8kMO
@JustJerry121 When the checklist started nesting and branching it became obvious this needed to be backed by a workflow software. When it became clear that the workflow needed some level of dynamism to adapt for specific customers, it made sense to introduce agentic parts.
Most companies building complex software products scale technical implementation by hiring forward-deployed engineers. We built an internal application to automate most of it first.
We break this down in our newest blog post, including how it works, what it's meant for our team, and where we see the role of FDEs.
https://t.co/zINdFGmvHM
I've been working on this with Sai. I love this kind of work. It really scratches an itch that's down in my soul; never to be satisfied. I'll take this, and every opportunity I have, to suggest this now very dated book as well.
https://t.co/TlREa6PSrC
The way we write, ship, and maintain software is changing fast.
Join @cursor_ai, @modal, @AntithesisHQ & @antimetal in NYC on Feb 26 to see how teams are building the AI dev stack.
RSVP required
https://t.co/9eDvWY3gUi
As software systems grow, understanding them becomes fragmented across people and tools. No single artifact captures how the system actually behaves.
We describe the model we’ve been building at Antimetal to make this implicit understanding explicit and machine-legible.
https://t.co/fJPEbqObEX