AI agent companies face a critical challenge: adoption. Despite their impressive capabilities, most people struggle to make agents part of their daily workflow. The problem isn't the technology — it's the mismatch between how these agents are designed and how we actually work.
Take Devin, the AI software engineer I've been using for months. My first four-hour session felt like magic. I watched it automate tasks, build small projects from scratch, and attempt to contribute to existing codebases. While it excelled at greenfield development, it struggled with legacy code integration — but that wasn't Devin's fault. I needed to provide better prompts and context.
The experience forced a fascinating mindset shift. Using Devin, I wasn't an engineer anymore — I was a technical product manager commanding a team of junior developers. This new workflow felt genuinely exciting.
But then... nothing changed. The next day, I needed to fix urgent bugs in my project. I could've used Devin, but that meant crafting the perfect prompt and monitoring its progress. With the solution already clear in my head, it was faster to open Cursor and just fix it myself. This pattern repeated until I forgot about Devin entirely.
Minion, a general-purpose AI agent, followed the same arc. This iOS app promised to automate web tasks like restaurant bookings and parking tickets. Watching it book a table was entertaining, but I churned immediately — I didn't need it right then. Days later, while booking a flight, Minion crossed my mind as I typed "flights.google.com." But habit won: I was used to doing it manually, and I already knew exactly what I wanted.
Here's the takeaway for AI agent builders: fix the fit between form and function. Agents need to either integrate seamlessly into existing tools or mirror our current workflows.
Cursor gets this right. It's an AI code editor with Composer, their autonomous agent, built in. I use it constantly because it lives where I work. Crucially, Cursor doesn't position itself as an "agent product" — the agent is just a feature. This might be the key to wider adoption.
The market has proven my thesis right. Devin ended up pivoting to be more similar to Cursor and Windsurf — focusing on integrating their agent directly into developers' existing workflows rather than being a standalone product. Meanwhile, Minion has shut down entirely, unable to overcome the adoption challenges of being a disconnected, general-purpose agent. The future of AI agents is clearly in seamless integration, not standalone products.