Augment Code

Big news from Augment Code. We've agreed to sell Cosmos, Auggie CLI, the Code Context Engine & related tech to Harness, the Autonomous SDLC Platform. Harness's integrated ecosystem provides the ideal home to deliver these capabilities at true enterprise scale.  Best of all, our team is moving to Harness to continue advancing these products with our new strategic alignment. We started Augment with a mission to help enterprises build software faster. As AI generates more code, the challenge has shifted. The bottleneck is no longer writing code. It’s orchestrating, testing, and deploying AI-generated code reliably across the entire software lifecycle. Augment Code is joining Harness to bring our software factory capabilities into its Autonomous SDLC Platform. Together, we can make orchestration part of how software gets delivered: from generating code to governing every pipeline. To our customers, partners, investors, and team: thank you for trusting us, pushing us to solve harder problems, and helping us build Augment. We're excited to keep working on this mission with Harness. Read the announcement: https://lnkd.in/eTSsix2k

Your PR sat for three days. Not because the code was hard. CI went red. A reviewer left eight comments. Main moved on without you. And when your reviewer finally opened it, they didn't have enough context to be confident hitting approve — so it sat a little longer. Writing code stopped being the bottleneck a while ago. Getting it merged is where teams are stuck now. So we pointed a fleet of agents at the PR itself. You flip one switch — Ready for review — and they fan out. They clear the busywork: • Failing CI gets diagnosed and fixed • Review comments get addressed with a reply explaining what changed • Merge conflicts get resolved And they build the case for correctness: • A line-by-line review of the diff, tuned for real bugs over noise • A risk classification of the change • End-to-end verification that actually exercises the behavior and reports evidence • A briefing for the human on the design and security calls worth a second opinion That last part is the point. The three-day delay usually isn't laziness — it's a reviewer who doesn't have enough signal to say yes. Give them a real review and real evidence, and the decision gets fast. The human still approves. The human still merges. No agent merges your code, and that's deliberate — judgment stays where it belongs. Short demo below. Curious what your team's actual PR-to-merge time looks like right now.

Spend a morning in NYC building your AI software factory — not hearing about one. On September 17, we're running a half-day, hands-on workshop for engineering and technical leaders on tightening the loops between spec, code, review, and ship. This isn't a webinar with a slide deck. Two full hours are yours to build: → Real loop engineering examples from live customer deployments  → Test environment access, set up on the spot  → Build your own loops with our team in the room  → Present what you built, get it reviewed by your peers 9:00 AM – 2:00 PM at Convene, 530 Fifth Ave. Lunch, snacks, and drinks included. Giveaways for the best loops. Seats are limited and registration is by approval — apply in the comments.

A quick start guide: setting up a software factory in Cosmos takes two screens and one prompt. 1. Connect your code. Install the GitHub app, or connect GitLab with a group access token, and choose which repositories your factory can work in. 2. Create the environment. Name it, pick a base image — Cosmos Default, TypeScript, Go — and your experts get a shared cloud sandbox to work in. 3. Describe the factory you want. Open Cosmos Advisor and say it in plain English. Advisor does step 3 as work, not as a form. It decides the expert roster, writes each configuration, scopes seven triggers to your repository, and applies them itself. It asks you one clarifying question along the way, and it honors the constraint you set. You describe the outcome. The platform configures itself. Full walkthrough in the video. https://lnkd.in/gK_3Nbi9

Everyone wants a software factory. Nobody wants the wiring it takes to get one — the wiring up, the config screens, the dashboard you go check afterward. Cosmos Advisor collapses all of it into one prompt box. "Build a software factory: ticket → PR → review → merge → deploy." That one sentence stands up the factory. The same box is where you go when something inside it breaks, and where you find out what the whole thing returned this month. No click-ops, no manual configuration. Just ask. https://lnkd.in/gK_3Nbi9

Augment Code’s Cosmos is a single platform for running specialized agents across the SDLC, from specs and implementation to code review and verification. Agents share context and improve with every run, while engineers remain in control at the points that require the most judgment. Teams can also build and tune their agents using natural language. We’re excited to welcome Augment Code as a Headline Partner for #LDX3NewYork! 🙌 Learn more: https://lnkd.in/d-mD8DCU 📅 September 15 & 16 📍 Javits Center, New York 🎟️ Join us at LDX3: https://lnkd.in/eJQN5pWS

Earlier this year we rebuilt the Auggie CLI harness from scratch. The result: on SWE-bench Pro, at the same pass rate, Auggie v2 completes a task for $1.27 where Claude Code spends $2.70. Three months of prototyping led us to fork Pi, an excellent minimal open-source coding harness built by Mario Zechner, and plug our context engine into it as an extension. The biggest lever was subtraction. V1 had a specialized tool for everything: launch-process, read-process, kill-process, str-replace-editor, and more. V2 gives the model one bash tool and three file tools: read, edit, write. Every tool schema gets serialized into every LLM call, so a wide toolset is a standing tax on cache-read tokens for the life of a session and it invites the model to spend output tokens choosing between narrow tools instead of just doing the work. Read the full breakdown of the rebuild, including why two earlier prototypes (Vercel AI SDK, Mastra) didn't make the cut in the link below 👇 by Tulga Tsogtgerel and Chris Kelly.

AI coding tools can boost organizational productivity by 30-40%. Beyond that, the bottleneck becomes all of the steps after code is generated: review, verification, bug fixing, and human decision-making. The answer? Build a software factory that optimizes human input to just judgment calls and knowledge transfer. Here's how we rebuilt Augment's PR to Merge loop on Cosmos to improve our internal software factory. See the blog link in the comments.

Less than one month after Grok 4.5 came to Augment, Grok 4.6 is now available in Cosmos, our agent orchestration platform for building software factories. Together with Augment's Context Engine, Grok models have become a powerful option for working across large codebases. We're excited to see what our customers build with the most intelligent Grok model to date!

Engineering manager Alex Ding talks about how we solved our code review bottleneck with a fleet of agents 👇️