In the GenAI world, there's always something new popping up, and this time it's a game-changer. Cognition AI has just rolled out Devin – the first AI officially stepping into a software engineer's shoes. The founders of Devin are the world's best competitive programmers, which people used to see on codeforces. The team got some big names like Peter Thiel and other tech stars backing them up, making this debut a pretty big deal. But here's the kicker: Devin is more than just a new GenAI tool—it's a glimpse into the future of how we think about building software. So, buckle up as we explore the real talk behind Devin, its capabilities, and potential risks.
Is Devin really a game-changer?
Using AI in software development isn't exactly a breakthrough. Tools like GitHub Copilot, StarCoder, Replit, and Codeium have been around for a while. These tools have been great at lending a hand with coding, doing everything from spitting out basic code based on what you type, to summing things up or pulling up code snippets to make the team's life easier.
But Devin has more things to offer. It can complete entire coding projects while still in the testing phase, showing incredible coding skills. Cognition hasn't detailed the specifics of how they did this, including whether they're using their own technology or outside tech. However, they do share that these capabilities come from their improvements in long-term thinking and planning. And from what they hint at on their website, coding is just the start—there's much more to look forward to.
What can Devin do?
Do they mean that Devin's not here to take software engineers' jobs? Let's see what it's capable of.
1. Devin successfully answered interview questions from top AI companies. Performed shockingly well on SWE-Bench coding benchmarks, and completed Upwork tasks.
2. It has an all-in-one interface – code editor, terminal, and web browser. It can build apps end to end. Here’s how Devin created the “Game of life” website.
3. It flawlessly completed a job posted on Upwork without much effort.
4. Devin's an impressive learner – it masters new skills and technology in a breeze.
5. It fixes errors—even its own—through debugging. Here's how it fixes a bug with algorithm calculations in the Sympy Python algebra system.
Funny enough, just like the common practice, Devin also relies on print statements for debugging!
6. It showed around 3x the performance of Claude 2 (former best performer) for resolving GitHub issues in real-world problems.
7. Devin can train and fine-tune its own AI models.
8. It's incredibly autonomous.
Getting access to Devin
Right now, Cognition's gearing up, making Devin available to just a select group of people at first. They're hinting that if you want to see what Devin can do for your projects, you should drop them an email. Wider access is on the horizon, just not yet.
What top AI figures say about Devin
Andrey Karpathy has an interesting take on Devin, comparing the growth of automating software engineering to how self-driving tech is evolving. It's all about moving from manual work to letting AI take over bit by bit. Initially, it's humans doing all the work, from driving a car to writing code. Then, AI starts to help out with the simpler stuff, like keeping a car straight or finishing off a line of code. As things progress, AI takes on more, like dealing with traffic or writing big chunks of code.
Karpathy mentions that it's not just about AI doing its thing. We've got to rethink how we interact with it—how we guide it, spot and fix problems, or steer it in a new direction. This might mean big changes in how we use our coding tools.
Here's what François Chollet thinks about Devin:
Risks with Devin
Devin's launch has definitely got people talking. There's a lot of excitement about how it might free up engineers to work on the really cool, complex stuff. But, there are also some worries about jobs and whether AI can really capture the creative spark humans bring to coding. The Cognition team pictures it more like bringing a super capable new team member on board. They're all about enhancing teamwork and boosting efficiency.
Their ambitions don't stop with coding. They're curious about crafting AI helpers that could outdo the current tools in many different areas. Scott Wu, the CEO of Cognition, mentions this is just starting, with lots more challenges and exciting possibilities on the horizon.
Cognition's busy making Devin even better, aiming for a future where software blends the best of human creativity with AI's knack for getting things done. It's still a bit early to say how Devin will fit into day-to-day work for developers or what new kinds of AI teamwork might pop up. But it's clear that AI's role in software development is on an exciting path, and Devin is leading the way.
At the end of the day, one question remains:
Wrapping up
As we come to the end of our journey exploring Devin, the first-ever AI software engineer, it feels like we're standing on the edge of a huge change in how we make software. With its unparalleled ability to take on entire projects, Devin is not just another tool but a signal of what's next in tech. Even though it's just getting started, the buzz and wonder around what Devin can do are really pointing towards a world where AI and our own creativity work together closer than ever.
