Agentic IDEs Take Over: Cursor Surpasses $2B Revenue and Launches 'Automations' for AI Agents

Host: I am joined by Marcus Thorne. Marcus is a Principal Engineer and has spent the last decade building high-scale distributed systems, and more recently, he’s been one of the early power users pushing Cursor to its absolute limits. Marcus, welcome to Allur! Guest: Thanks, Alex! It’s great to be here. Honestly, I’ve been obsessed with this specific topic lately. It feels like we’re finally seeing the "bridge" being built between just chatting with an LLM and actually having a digital teammate. Host: Right? And that $2 billion revenue figure for Anysphere—the team behind Cursor—is just... I mean, wow. It’s a watershed moment. Why do you think Cursor is suddenly pulling away from the "legacy" players like GitHub Copilot or JetBrains? Guest: You know, I think it comes down to one fundamental design choice. Most of the big players treated AI as a feature. It was a sidebar or a popup. But Cursor—they basically said, "What if the LLM was the core of the IDE?" When you use it, the AI isn't just looking at the file you have open; it has this deep, indexed context of your entire project. It knows your folder structure, your types, your weird legacy utility classes... everything. Host: It’s the "context" piece, isn't it? I remember when I first tried it, I was shocked that it actually understood how my Laravel routes connected to my controllers without me having to copy-paste the code into a prompt. Guest: Exactly! And that’s what led to this revenue surge. Developers—especially in fast-moving startups—are realizing that the "plugin" model is too slow. But the big news this week is "Automations." And Alex, this is where it gets really interesting. We’re moving from reactive AI to proactive AI. Host: Okay, let’s dive into that. Because "Automations" sounds like a broad term. In the Cursor context, what does it actually *do*? Is it just running scripts? Guest: It’s much deeper. Think of it as an event-driven workforce. Currently, if I want the AI to fix a bug, I have to open Cursor, hit Cmd+K, and tell it what to do. With Automations, the AI responds to *triggers* outside of the editor. So, imagine you’re in Slack, and a developer says, "Hey, the login endpoint is throwing a 500 error in the staging environment." A Cursor agent can be set to monitor that Slack channel, see the bug report, go into the codebase, analyze the logs, and by the time you actually open your laptop, there’s already a Pull Request waiting for you with a proposed fix. Host: Wait, actually? It can trigger off a Slack message? That sounds like... well, it sounds like I might be out of a job soon! But seriously, how does it handle the complexity? Because a bug fix in a complex Go microservice isn't just a one-line change. Guest: Well, I don't think we're out of jobs yet, but our roles are changing. And you hit on a great point: complexity. Cursor handles this through what they call "Orchestration vs. Execution." It’s not just one single LLM trying to do everything. It’s a hierarchy. You have a "Planner" agent that looks at the high-level goal, and then it spawns "Worker" agents to execute specific tasks in parallel. One worker might be refactoring the database schema, another is updating the API docs, and a third is writing the unit tests to make sure the fix didn't break anything. Host: Oh, interesting! So it’s like a project manager and a team of juniors all living inside your repository? Guest: Precisely. I actually had a "eureka" moment with this last week. I had a massive refactor—moving a bunch of PHP logic into a more service-oriented architecture. Normally, that’s a two-day job of manual "find and replace" and fixing type hints. I set an automation to handle the migration, and it just... worked through the files one by one, keeping the global context in mind. It was spooky how well it handled the edge cases. Host: That is wild. But okay, let's talk about the "janitorial" side of coding. You mentioned documentation. We all hate writing docs. Every developer says they’ll update the README, and then nobody does. Guest: Oh, tell me about it. It’s the first thing to rot. But with agentic IDEs, documentation becomes "living." Because the agent is the one performing the refactor, it knows exactly what changed. Part of the automation chain is: "If the API signature changes, update the Swagger docs and the README." It closes that loop. We’re moving toward "self-healing" and "self-documenting" codebases. Host: I love the idea of self-healing code. So, like, if a CI test fails, the agent just... wakes up and tries to fix it? Guest: Exactly. You can configure a trigger: `on: event: ci_test_failure`. The agent looks at the stack trace, identifies the regression, and drafts the patch. It’s a huge shift for DevOps. Instead of a developer getting a PagerDuty alert at 3 AM for a minor regression, the agent might be able to resolve it—or at least provide a very high-quality post-mortem for the human to review in the morning. Host: Okay, Marcus, I have to play devil's advocate for a second. If we have agents autonomously pushing code and fixing bugs... what about security? I’m thinking about "prompt injection" or an agent accidentally introducing a vulnerability or, I don't know, hardcoding a secret because it was "easier" to make the test pass. Guest: That is the million-dollar question. And it’s why Cursor is talking about "Agentic Firewalls." You can't just let an AI have free rein over your production branch. There has to be a governance layer. These are basically automated checks—using both static analysis and other "security-specialized" LLMs—that scan every AI-generated PR. It looks for hardcoded credentials, insecure dependencies, or logic that violates your team's security policy. We’re basically building AI to watch the AI. Host: It’s like a digital "trust but verify" system. Guest: Exactly. And honestly, I think we’re going to see a shift in what makes a "senior" developer. It’s not going to be about who can type the fastest or who knows the most obscure Go syntax by heart. It’s going to be about who can best *orchestrate* their AI workforce. Who can write the best "automation recipes" and who has the architectural vision to guide these agents. Host: That’s a powerful way to put it. It feels like we’re becoming architects rather than just bricklayers. Guest: Exactly. And I think that’s why Cursor is winning. They aren't trying to help you lay bricks faster; they’re giving you a construction crew. Host: I love that analogy. Marcus, this has been such an eye-opener. Before we wrap up, for the developers listening who are maybe still on VS Code or IntelliJ and feeling a bit overwhelmed by this—where should they start if they want to dip their toes into this "agentic" world? Guest: Honestly? Just download Cursor and try the "Composer" feature first. It’s their multi-file editing mode. Once you see it refactor three files at once and keep them all in sync, you’ll have that "lightbulb" moment. And then look into their new Automations beta. It’s going to change how you think about your "idle" time as a developer. Host: "Idle time"... man, the idea that my code could be improving while I’m at lunch is both amazing and a little bit haunting. Marcus Thorne, thank you so much for joining us on Allur today! Guest: My pleasure, Alex. Thanks for having me!