At its best, programming is a creative endeavor, but in this age of shifting everything left, much of a developer’s day is filled with what All Hands AI co-founder and CEO Robert Brennan calls the “toil-oriented task” like writing unit tests, managing dependencies and keeping documentation up to date. AI, on the other hand, may not be creative, but it is pretty good at exactly those routine tasks.
All Hands AI, which announced a $5 million seed funding round led by Menlo Ventures on Thursday, aims to build model-agnostic open source AI agents that can handle most of this toil and allow developers to focus more of their time on doing what they do best.
A few months ago, AI Cognition showed off Devin, an AI agent that could plan and execute complex engineering tasks — and maybe more importantly: build and deploy new applications end-to-end.
“The Cognition folks came out with their Devin demo and I — and I think every other software engineer in the world — was amazed at that video,” Brennan said in an interview ahead of Thursday’s announcement. “I think it really catalyzed in our imagination what the future of development is going to look like, but also kind of scared us that it was being developed as closed source and that it was being kept in this walled garden we couldn’t see and contribute to and really own as a development community.”
This open source project, which started out as OpenDevin earlier this year and is now called OpenHands, started with a text file on GitHub and now has over 30,000 stars and more than 150 contributors.
The idea is for the OpenHands agent to become a proactive pair programmer who works hand-in-hand with the developer and who can handle much of the toil of a developer’s day-to-day work. That may involve writing tests and deploying and application, but also recognizing that a change in one file (maybe the name of a function) might influence how other parts of the application function and asking the developer if it should adjust the affected files accordingly.
“AI is going to completely change how developers work. But it’s not going to change their preference for adopting open source, especially when it comes to technology that affects their day-to-day work,” said Joff Redfern, a partner at Menlo Ventures and former chief product officer at Atlassian. “By building in the open, All Hands is helping the software engineering community work toward an ideal AI-powered development experience.”
Brennan and his two co-founders, Xingyao Wang (chief AI officer) and Graham Neubig (chief scientist), have extensive experience in working in natural language processing and building agents. Brennan previously worked on document summarization at Google and then in executive roles at a number of startups, working on machine learning and infrastructure projects. Neubig is an associate professor at Carnegie Mellon with extensive experience in natural language processing; Wang is interrupting his doctorate program at the University of Illinois Urbana-Champaign, where he did research on interactive language agents powered by foundation models.
“None of us were surprised to see the Cognition demo in terms of technology,” Brennan said. “We all knew that this was there, but seeing it all come together into a user experience really just excited us to start pushing on building that in the open.”
Brennan also noted that while tools like Copilot are very helpful for developers, they aren’t (yet) focused on the entire “agentic loop of writing code” akin to a self-driving car. That’s what All Hands AI is aiming for, even if this still remains somewhat aspirational. It’s not like you can give the agent access to a company’s entire JIRA backlog and let it loose and accomplishing every task in that. Indeed, Brennan — like most people in the industry today — thinks that there will be a need for human developers in the loop for a very long time.
There are also still unsolved questions around what the user/developer experience for such a system should look actually like. All Hands AI does have a designer on staff, though, and it’s good to see it looking into these questions early on. Right now, the experience is also still somewhat decoupled from the development environment, but the team plans to build integrations with VS Code and other editors soon.
As with many open source startups, All Hands AI expects to monetize its service by offering paid, closed-source enterprise features. “We think that there’s a bunch of software that we can build that complements the open source that’s really delivering value to big enterprises where we can feel good about building that in a closed source way to help make sure that we have a sustainable open source project that’s getting financial contribution back from the larger enterprises that are using it,” Brennan said.
With this first funding round, though, the team plans to build out its technology stack before it delves deeper into monetizing the service. In addition to Menlo, which led this round, Pillar VC, Betaworks and Rebellion also participated. The company also brought on a number of angels, including Hugging Face co-founder Thom Wolf; Cloudera co-founder Jeff Hammerbacher; and PyTorch creator and Meta VP Soumith Chintala.
Comment