Blog
Updates, deep dives, and stories.
Pretty please won't stop your agent from deleting production. Here's what will.
PocketOS lost its production database to a single API call in nine seconds. The standard responses — evals, prompts, permission popups — all miss the actual problem. Here's why the fix has to be structural, and what that looks like in Monk.
Capsules as audit targets
AI-generated code is shipping faster than security teams can audit it. Monk capsules turn every git branch into a production-shaped cluster on your own cloud — pinned to a commit, disposable, reproducible. Here's why we think that's interesting for security work, and an open invitation to come break it.
Why giving your coding agent shell access is a bad idea
Most coding agents get shell access by default — a terminal, your cloud CLIs, your .env file. This worked when the operator was a human. It's the wrong architecture for an agent. Here's why, and what a narrower interface looks like.
Monk Capsules: 0 config, full-stack, per-branch previews
Every git branch gets a real production environment on your cloud, with its own HTTPS URL — no manifests, no cluster, no setup. Here's how capsules work, why we built them, and what changes the day you turn them on.
State vs context — the difference in 5 examples
Coding agents have a lot of context and almost no state. That's the architectural difference between a very smart assistant and a system you can leave to operate infrastructure. Five common situations where context runs out and state has to take over.
The other half of the harness
OpenAI's Codex team named a new discipline: harness engineering. Their harness stops at the git boundary. Here's what we built for everything past it — and why every git branch now gets its own production.
A New Category
Autonomous software.
Software that builds, deploys, and manages itself. No human in the execution loop. You define the intent. You set the guardrails. Your coding agent and Monk handle everything between idea and production.
5-day free trial.