Cursor just deleted the CMS running their website. And it wasn’t a mistake.
This was the work of Lee Robinson, a dev educator at Cursor.
He saw the way a CMS slowed the team down, and just decided to get rid of it. Three days, over 300 agent pull requests, $260 in tokens, and suddenly Cursor’s agents could ship code straight to the website from slack.
Was this the money? Not really. Sure, the CMS had cost Cursor nearly $57,000 so far this year.
But the real cost wasn’t the invoice—the real cost was cutting agents out of the loop. And that’s what I want to talk about this week.
At Cursor, it is unacceptable to click through menus instead of delegating to agents. Those menus aren’t innocent. They’re built to make content easier but in 2025 they’ve become a wall between the agents and the work.
Last week I argued that most enterprise agents fail because they don’t remember. This week is the uncomfortable follow-on: even if you solve for memory, most companies will run into a wall of software that blocks those agents from acting.
The work itself is stuck—hidden behind click paths, scattered across draft modes, encoded in tribal knowledge. You buy agents; you get a conversation layer on top of the same bottlenecks.
Cursor’s story shows what it looks like to break through. Not by getting better agents, but by asking which abstractions are still earning their keep. I’m extending that story here to get at the real heart of the problem—what it takes to get to Cursor’s level of fluency with agents when you don’t start with AI-native talent.
This briefing covers:
The abstraction tax: Why convenience layers that helped humans for decades are now blocking agents—and how to see where you’re paying it
Two kinds of primitives: What makes agents reliable (last week) versus what makes work shippable (this week). You need both.
The cultural pattern: Why Cursor’s marketing team commits code, and what “code wins” actually means for non-engineers
Six concepts to teach your non-tech teams besides prompting: State, artifacts, change records, checks, rollback, traceability
Three prompts to make this real:
The Abstraction Tax Audit: Find where your current tools are blocking agent leverage
The Primitive Fluency Gap Analysis: Design training around work primitives for non-technical teams
The “What Would We Delete?” Exercise: Identify sacred workflows that might be candidates for simplification
The winners won’t be the companies with the most agents. They’ll be the ones where enough people understand the primitives to delete sacred workflows and let agents actually operate.
I think this story is one of the most interesting of the year, and it illustrates one of the big change themes we’re going to see in 2026 across companies. Jump in and get a head start here!
This is an Executive Circle briefing, a Sunday newsletter exclusively for Founding Tier Members. You can learn more via this 60 second video explaining what’s in each tier, and you can change your plan here.
