Solutions

Automate the toil around the code

PR triage, release notes, incident summaries, dependency watch — agents handle the process work that surrounds shipping.

What changes

First-pass code review

Every diff gets read against your standards before a human opens it — correctness, security, missing tests — so review time is spent on design rather than on catching the same omission again.

Release notes from merged work

Written from what actually shipped, in the format you publish, rather than reconstructed from memory on release day.

Incident paper trail

Timeline assembled from the systems involved, with the sequence of events and the changes that preceded them recorded while it is still fresh.

The work around the work

Engineering time disappears into process: triaging issues, chasing reproduction steps, writing release notes, summarising incidents, checking dependency advisories. It is genuinely necessary and it is nobody’s reason for being an engineer. It is also unusually well suited to agents, because the inputs are structured, the systems have good APIs, and the judgment required is real but bounded.

Local-first matters more here than anywhere

Source code is the asset engineering organisations are most careful about, and rightly so. Because AGNT runs on your machine and can point at a local model through Ollama, first-pass review can happen without the diff leaving your hardware. That turns "we cannot send our code to a third party" from a blocker into a configuration choice.

Start where being wrong is cheap

Issue triage and release notes are the right first deployments: useful immediately, and a mistake costs a correction rather than an outage. Code review comes next, in comment-only mode. Anything touching deployment stays behind an approval gate for as long as it takes the receipts to justify moving it — which may be forever, and that is a legitimate answer.

Common questions

Does our source code leave the machine?

Only if you choose a hosted model. With a local model through Ollama or LM Studio, review runs entirely on your own hardware.

Can agents merge or deploy?

They can be given that permission and almost nobody should. Comment, draft and prepare are where the value is; merging stays human.

How does this fit with our existing CI?

Alongside it. CI enforces what is expressible as a rule; agents read for the problems that are not.

Give AI a job. Get the proof.