At its best, programming is a creative endeavor, but in this age of transforming everything else, much of a developer’s day is filled with what All Hands AI co-founder and CEO Robert Brennan calls the “work-oriented key” such as writing unit tests. , dependency management and keeping the documentation up to date. AI, on the other hand, may not be creative, but it is quite good at exactly those routine tasks.
All Hands AI, which announced a $5 million funding round led by Menlo Ventures on Thursday, aims to build model-agnostic open source AI agents that can handle most of this work and allow developers to focus more of their time on doing what they do. better.
A few months ago, Cognition AI demonstrated Devin, an AI agent that could plan and execute complex engineering tasks—and. perhaps most importantly, build and deploy new end-to-end applications.
“The people at Cognition came in with their demo Devin and I — and I think every other software engineer in the world — was amazed by that video,” Brennan said in an interview before Thursday’s announcement . “I think it really catalyzed in our imagination what the future of development will be like, but it also scared us that it had been developed as a closed source and that it was kept in this walled garden that we could not see. contribute and really proper as a developing community.”
This open source project, which started as OpenDevin earlier this year and is now called OpenHands, started with a text file on GitHub and now has more than 30,000 stars and more than 150 contributors.
The idea is that the OpenHands agent will become a proactive pair programmer who works hand-in-hand with the developer and who can handle much of the developer’s day-to-day work. That may involve writing tests and implementation and application, but also recognizing that a change in a file (perhaps the name of a function) could affect how other parts of the application function and asking the developer if must adjust the affected files accordingly.
“AI will completely change how developers work. But it will not change their preference to adopt open source, especially when it comes to technology that affects their daily work,” said Joff Redfern, a partner of Menlo Ventures and former chief product officer at Atlassian. “By building in the open, All Hands helps 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 working in natural language processing and building agents. Brennan first worked on document review 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 interrupted his doctoral program at the University of Illinois Urbana-Champaign, where he conducted research on interactive language agents powered by foundational models.
“None of us were surprised to see the Cognition demo in terms of technology,” Brennan said. “We all knew this was there, but seeing it all come together in one user experience really excited us to start pushing to build that out into the open.”
Brennan also noted that while tools like Copilot are very useful for developers, they are not (yet) focused on the entire “agent cycle of writing code” like an autonomous car. That’s what All Hands AI is aiming for, although it still remains somewhat aspirational. It’s not like you can give the agent access to a company’s entire JIRA backlog and leave it and accomplish every task in it. Indeed, Brennan – like most people in the industry today – thinks there will be a need for human developers in the loop for a very long time.
There are still unresolved questions about what the user/developer experience for such a system should actually be. All Hands AI has a designer on staff, though, and it’s good to see that they’re looking into these questions first. Right now, the experience is 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 features. “We think there’s a lot of software that we can build that complements open source that really gives value to big companies where we can feel good about building that in a closed way to help make sure we have a sustainable open source project. This is getting a financial contribution from the larger companies that use it,” Brennan said.
With this first round of funding, however, the team plans to build its technology stack before delving deeper into monetizing the service. In addition to Menlo, who led this round, Pillar VC, Betaworks and Rebellion also participated. The company also brought in a number of angels, including Hugging Face co-founder Thom Wolf, Cloudera co-founder Jeff Hammerbacher and PyTorch creator and Meta VP Soumith Chintala.