About Us
We've been hands-on with AI longer than most people have been paying attention to it. Before tab-tab autocomplete in your IDE. Before GPT-4. Before "tool calling" was a phrase anyone used. We were building chains by hand, stitching together early RAG pipelines, wiring up the kind of multi-step workflow graphs nobody was calling agents yet because the word hadn't settled on them. A lot of it was duct tape and a lot of it broke. But the thing we kept noticing, the longer we spent in prompt engineering, was that you get what you give. Same model, different context, wildly different output. Context was going to be the differentiator. And almost as quickly it became clear it was really two problems stacked on top of each other - the context you feed the model, and the harness you give it to operate inside. Both had to get solved.
We started Buildforce a year ago with a thesis that's evolved a lot since. That AI was about to change software development. That the missing piece wasn't going to be the model - it was going to be context. We had no idea how right and how wrong we were at the same time.
We were right that context is the bottleneck. We've watched it play out hundreds of times in our own work. The agent reads the code, misses the reasoning, ships something that compiles cleanly and is still confidently wrong.
We were wrong about what to do with it. The first version of Buildforce was a CLI with slash commands, plan files, session artifacts, the whole thing. We used it every day and it worked for us. Almost nobody else used it. We learned, slowly and painfully, that you can't ask developers to do extra work to capture what they're already producing. The capture has to ride on infrastructure that's already there. So we threw out a lot of code and started thinking again.
What we have now is closer to right. Engraph is the open-source foundation - a git-native layer that materializes codebase expertise as part of how you already work, not as an additional thing you have to maintain. Buildforce is the intelligence platform built on top of it, where planning, building, and review all happen with your team's accumulated reasoning actually present in the room. Underneath both is a conviction we keep hardening: context layer, not model capability, is the real bottleneck for AI productivity. The teams that solve it first will pull ahead in ways the rest find hard to copy.
But the bigger thing we're working toward isn't a product. We believe the people who craft and build software are the ones responsible for shaping what AI in software looks like, and we believe that work has to happen in the open. Not behind closed doors at a handful of large labs. Not with your team's reasoning locked inside someone else's vendor cloud. The expertise that lives inside an engineering team is the most valuable thing that team produces, and it has to belong to the team - portable, version-controlled, readable by any tool, theirs.
The version of the future we're betting on is one where humans don't get smaller. Where what makes you irreplaceable - your thinking, your taste, your accumulated judgment, the memory of why you decided what you decided - doesn't evaporate at the end of every AI interaction. It becomes a layer you carry with you. It compounds. The next agent stands on it instead of starting from zero. The next teammate inherits it instead of waiting for you to be in a meeting. That's the work. That's why we're here. We're not done figuring it out, and we're going to keep doing it in public until we are.
If any of this resonates, or you're working on the same problem from a different angle, get in touch.