Augment Code

Our team has become 3x more productive by building a software factory. What excites me most is that we can now go from customer feedback to a deployed fix within an hour. The old path would take 2-3 weeks: collect feedback and bug reports, synthesize and prioritize them, plan the work, then wait for it to be built. With Augment Cosmos, I now have agents automating every step of that process: - A customer agent answers questions, triages bug reports, and creates the ticket. - A builder agent starts with a spec, then pairs with an engineer to iterate on it. - A PR agent drafts the PR, subscribes to comments and test failures, and babysits it until it’s mergeable. The human is still in the loop for the spec and the judgment calls. But I barely spend time managing the tickets themselves. The real advantage is that every piece of feedback becomes shared memory, so the workflow gets smarter over time. I talk about our full ticket-to-PR loop in this video 👇 Comment with any questions!

Reviewing code isn't the goal. Merging it is. Most code-review bots optimize for the wrong output. They review line by line, leave comments, then stop — handing the work back to a human who still can't tell if the PR is ready. The PR-to-merge loop doesn't get shorter. It gets busier. With Augment Cosmos, we built a fleet of agents that drives each PR toward merge, pulling in humans only for the highest-leverage judgment calls. Here's one PR's journey. It starts with an engineer commenting "fix it" on a Slack bug thread. Augment Cosmos takes it from there: 🔹 A risk analyzer assesses the change and pulls a human in when necessary. 🔹 A deep code-review agent checks correctness at the line level. 🔹 An end-to-end verifier runs the tests (including UI integration tests). 🔹 A PR fixer agent responds to comments, fixes CI issues, and keeps the PR moving. If a human needs to weigh in, Cosmos pairs them with an agent to review the PR at the intent and design level — so they can understand the change deeply without wading through all of it. The system also learns from that feedback, so review gets sharper over time. Full loop: under 30 minutes. I walk through the whole workflow in this video ⬇️ Comment with questions!

One of the most useful loops we've set up with Augment Cosmos is for incident management. By the time I look at an alert, an agent has already completed the first pass of an investigation. Cosmos has understood the alert and the surrounding code, reviewed the runbook, pulled relevant logs, correlated the metrics, looked at previous incidents, and surfaced a likely cause and tells me where I should be looking. I can then ask follow-up questions to understand the impact, rather than starting from zero. This leaves me to make the mitigation and judgement calls, and spend more time doing proactive work and preventing issues before they start. I talk about this workflow a little more in the video below

Individual agents make engineers faster. Engineering loops make that work repeatable across the organization. Cosmos coordinates agents, code, tools, policy, and memory in one system.

Individual agents make engineers faster. Engineering loops make that work repeatable across the organization. Cosmos coordinates agents, code, tools, policy, and memory in one system.

Individual agents make engineers faster. Engineering loops make that work repeatable across the organization. Cosmos coordinates agents, code, tools, policy, and memory in one system.