Cursor’s demo window pops open, and boom—there’s your AI agent, churning through a feature request like it’s 2024’s version of clippy on steroids. No code from you, just natural language commands, agents spinning up in the sidebar, promising to handle the grunt work while you sip coffee.
But here’s the thing. I’ve seen this movie before. Twenty years covering Silicon Valley, and every time a plucky startup pivots to chase the hot new interface, the giants with bottomless pockets rewrite the ending.
Cursor 3—code-named Glass—is their big swing at ‘agentic coding,’ that buzzword du jour for offloading entire tasks to AI minions. It’s baked into their desktop app, right next to the IDE they’ve been peddling. Type a task. Hit enter. Watch the magic (or mayhem) unfold.
“In the last few months, our profession has completely changed,” said Jonas Nelle, one of Cursor’s heads of engineering, in an interview with WIRED. “A lot of the product that got Cursor here is not as important going forward anymore.”
Nelle’s right, in a grim sort of way. Developers aren’t hammering keys like it’s 2010 anymore. They’re herding digital agents, checking progress, tweaking prompts. Cursor wants in on that shift. Smart? Desperate? Both.
Can Cursor 3 Actually Beat Claude Code and Codex?
Look, the pitch sounds slick. Unlike standalone apps from Anthropic or OpenAI, Cursor 3 meshes agents with your local IDE. Prompt a cloud agent for a feature, pull the code back to your machine for tweaks. smoothly, they say. But let’s cut the spin—who’s paying for all those GPU hours?
OpenAI and Anthropic? They’ve got billions burning holes in their pockets. Subsidized subs where $200 a month buys you $1,000+ in compute. Developers I talked to—guys building real stuff—are flocking there. Ronald Mannak from Pico AI ditched Cursor for Claude Code because, well, rate limits. Jack Crawford from mVara? Same story. “Value,” he says.
Cursor tried the subsidy game till June 2025, then flipped to usage-based pricing. Developers howled. Margins matter when you’re not Thiel-backed with endless rounds.
And the office? North Beach movie theater turned HQ, shoe racks replacing the pile by the door. Growing up, sure. But scrappiness won’t cut it against labs raising tens of billions.
This feels eerily like the early browser wars. Netscape innovated, owned the space—then Microsoft bundled IE for free, and poof. Startup roadkill. Cursor pioneered AI-IDE integration, funneled millions to OpenAI et al. Now those same labs launch agents, undercut on price. History doesn’t lie.
My bold call: Cursor gets acquired by end of 2025. OpenAI snaps ‘em up for the dev loyalty, folds it into Codex. Or Anthropic, to bolster Claude. Independence? Cute, but fleeting.
Why Are Developers Jumping Ship from Cursor?
It’s the economics, stupid. Claude Code and Codex aren’t just agents—they’re loss leaders. Anthropic’s tweaking rates now, but they’ve hooked millions. Cursor’s $50 billion valuation whisper? Frothy. Last fall it was half that. Hype fuels VCs, not sustainability.
Nelle and Alexi Robbins, Cursor’s engineering duo, shrug it off. “Doesn’t matter the interface,” they say. “Just use Cursor.” Optimistic. But devs vote with their terminals.
I’ve poked around Cursor’s app—solid, once. Now? Feels like yesterday’s news. Agents are the future, yeah. But whose agents? The ones with the deepest moats.
Cursor’s betting on hybrid appeal: agents plus IDE. Demo showed it—agent builds, you refine locally. Neat trick. Yet if big labs copy it (they will), game’s over.
Employees love the vibe—ship fast, no corp bloat. Appeal for talent. But talent alone doesn’t beat capital-intensive AI races. This chapter? Most expensive yet.
Zoom out. Coding’s profession is mutating, sure. But money flows to labs with models, not wrappers. Cursor’s a customer turned competitor. Ballsy. Reckless?
Subsidy wars echo cloud storage days—Dropbox gave away space, AWS undercut everyone. Winners? Hyperscalers. Cursor’s no Dropbox; it’s more like a side tool in the stack.
Who’s Really Cashing In Here?
Follow the dollars. OpenAI, Anthropic—enterprise goldmine. Devs hooked on cheap agents stick for production. Cursor? Squeezes margins, chases valuations. Who wins long-term?
Unique wrinkle: Cursor 3’s sidebar agent manager. Multi-agent orchestration, they claim. Like herding cats with PhDs. Cool for complex tasks. But reliability? Early days, bugs galore in demos I’ve seen.
Skeptical vet take: Impressive pivot, but timing’s off. Labs are a year ahead, with better models. Cursor integrates what they output. Tail, meet dog.
Dev shift’s real. Last year, Cursor ruled. Now? Niche player.
Prediction sticks: Acquisition bait. $50B val? Buyer premium.
🧬 Related Insights
- Read more: ML in the Dark: Solo Survival Tactics That Actually Ship
- Read more: AGI Economy Alert: Your Job Might Soon Be Checking AI’s Homework
Frequently Asked Questions
What is Cursor 3 and how does it work?
Cursor 3 is an AI agent interface inside the Cursor desktop app. Type a natural language task, agents spin up to code it—no manual writing needed. Review in sidebar, integrate with IDE.
Can Cursor 3 compete with Claude Code and Codex?
Tough sledding. Big labs offer massive subsidized compute; Cursor’s usage-based. Hybrid IDE-agent edge helps, but economics favor giants.
Is Cursor 3 free to use?
No—part of Cursor Pro, usage-based pricing post-subsidy era. Expect costs to scale with agent runs.
