So happy for my friends @kevinweil and @dpatil. Plenty of reasons to cheer these two on over the years. Helping scientists develop new medicines faster is a particularly good one. Congrats to the whole Warp Sciences team!

RT @GuildAI: This team's been shipping hard all summer long. And our first end of summer party is when we finally took a beat to celebrate.…

We surveyed hundreds of U.S. IT decision-makers about how they’re managing AI agents. 96% say they’re confident their organization has a complete and accurate inventory of its agents. At the same time, two-thirds of organizations using agents have experienced an agent-related incident in the past year. Only 40% have logging or audit trails. So maybe companies know their agents exist. But many still don’t know what those agents are doing, what they can access, who owns them, or how to stop them. Companies are putting agents into production much faster than they’re figuring out how to manage them. That’s the Agent Management Gap, and it’s the problem we’re building Guild to solve. guild.ai/ai-agent-management…

Of everything we’ve built on the Guild platform, Software Factory has had the biggest impact inside our own company. We use it every day, and it’s changed how our engineering team works. Now you can try the same Factory for free. We’ll even give you 100M tokens to get started.

I don’t think enterprises are going to pick one agent framework. They’re going to have OpenClaw, LangChain, goose, and probably ten things we haven’t heard of yet. Trying to force all of that onto one framework feels backwards to me. Give people freedom to build with what they want, then give the enterprise one secure, governed place to run it all. That’s the bet we’re making at Guild.

Great video on setting up a software factory on Guild. Try it for free and be up and running in minutes.

Tag an issue. Get a review-ready PR. The Guild Software Factory is pretty simple to get going: 1. Set up a factory in a few minutes 2. Connect it to your issue tracker 3. Tag an issue guild-auto From there it picks up the issue, plans the work, writes the code, runs the tests, and gives you a PR to review. And if you don’t like exactly how our factory works, change it. Swap models. Add or remove agents. Change the tools. Add your own rules, approvals, or workflow. Or build a completely different factory. That’s really the idea behind Guild. The factory is something we built on the platform, not a black box you’re stuck with. If you’re looking for something to mess around with over the long weekend, set one up for free here: guild.ai/product/software-fa… We’ll throw in about $200 of tokens too.

Our Software Factory now writes 34% of the PRs we merge at Guild and 56% of all our code fixes. Today, we’re releasing it for other engineering teams. We built Software Factory for ourselves first. It’s now doing a meaningful amount of the work that used to sit in backlogs or interrupt engineers. Production bugs. Flaky tests. Noisy logs. Small fixes. Maintenance. The stuff that matters, but often doesn’t get done because pulling an engineer away from something else costs more than living with the problem. The model cost to create a merged Factory PR is about $12. That means a lot of work that wasn’t worth interrupting an engineer for suddenly becomes worth fixing. A few numbers from our own codebase: - 70% of fixes for auto-detected issues come from the Factory - 81% of log-triage fixes are Factory-written - 91% of merged Factory PRs need no additional engineer commits - In some workflows, time-to-fix is down 97%, from roughly a month to a day This is probably the thing I’m most excited about that we’ve built at Guild. Software Factory is a system of specialized agents that can find work, figure out what needs to be done, implement and test the change, review it, and hand an engineer a finished PR. And importantly, it can start the work itself. Our production systems can detect an issue and feed it directly into the Factory without someone first creating a ticket, putting it in a backlog, and waiting for an engineer to pick it up. Humans still decide what gets merged. For us, that’s been the real shift. The Factory isn’t just helping engineers write code faster. It’s getting work done that otherwise might never get prioritized. We depend on ours every day now. Starting today, you can run your own Software Factory on Guild. Try it. → guild.ai/product/software-fa…

We just shipped Guild Optimizer. In our internal testing, we’ve reduced the cost of some agents by as much as 87% with no measurable reduction in quality. What surprised me was where a lot of those savings came from. Not some magical new model. Mostly from going back and fixing decisions that were completely reasonable when the agent was first built. The first job is to get the thing working. So you pick a model, write some prompts, give it the tools it needs, and ship it. Maybe the model is more powerful than you need. Maybe some of the work should just be code. Maybe you added a bunch of tools that never actually get used. It works, so you move on. Going back later is harder than it sounds. You need to know that the cheaper version still does the job just as well, which means having good evals. Most teams don’t. That’s what Optimizer does. It takes your actual production history and automatically builds an eval set. Then it starts testing cheaper models, looking for unused tools, evaluating the prompts, and finding places where predictable work can be moved out of the model and into code. You can see the proposed changes and the impact of each one before approving anything. Nothing just magically changes in production. One thing this has made pretty clear to me: there’s a lot more hiding in the cost of an agent than which model you picked. As we run more and more agents ourselves, I think continuously optimizing them is just going to become part of operating them. Optimizer is live in Guild today. Would love to hear what people think. guild.ai/blog/product/optimi…

Open source is about to get a new kind of contributor: agents. That’s one reason I’m so excited that Guild is partnering with the @linuxfoundation . A lot of us at Guild come from open source. Back when I was at Netscape, when we open-sourced Mozilla, I saw firsthand what an open community could build. I’ve believed in the power of that model ever since. And few organizations have done more for open source than the Linux Foundation, which supports some of the world's most important open-source projects and is now home to the Agentic AI Foundation. The Linux Foundation team has already pushed us in good ways. Their requirements drove our first support for an outside agent framework: goose, the open source framework that now lives under the AAIF. Over the coming months, we’ll be working with their team to build internal agents for the Linux Foundation running on Guild. Ultimately, their product team will be able to build, run, and manage those agents themselves. I’m personally so grateful for this partnership. And we can’t wait to see what this unlocks. I wrote more about why I think agents are going to change open source, and what we’re building together with the Linux Foundation: guild.ai/blog/news/linux-fou…