Zapier Gave Its Marketers Three Weeks to Start Building Software

Zapier is fully remote, and once a year it flies everyone into the same city. About 900 people. In February the company spent a week in Los Angeles, and Dan Slagen, its Chief Marketing & AI Transformation Officer, used three or four hours of that week to sit his marketing team in front of a stranger.

The stranger was an outside expert in Cursor, a tool that writes and edits working software from plain-English instructions. He had used it to build a digital twin of himself, a piece of software that answers the way he would. He spent about 40 minutes walking through how he made it, down to the files.

It did not go well, which was the point.

“Within about five minutes, everyone’s eyes had glazed over and people were starting to look. Confused, bored, frustrated.”

Slagen told the story to Mike Kaput in an interview published Thursday on The Artificial Intelligence Show. By his own estimate, about 5% of his marketers were working at the level they had just watched. When the presentation ended he asked the room who felt overwhelmed, and every hand went up.

His answer to them was not reassurance.

“This morning, you didn’t even know that this person existed. You didn’t know that you are potentially interviewing against this person for your next role at Zapier or outside of Zapier. Now, how do you feel?”

Marketing teams have spent two years being told to learn AI. Slagen told his team the learning part was finished.

The deadline, and the part he gave up

What followed was a schedule, not a strategy deck. Three weeks for everyone in marketing to be running on Cursor, from product marketers to brand designers. Five internal coaches, drawn from the people already furthest ahead. And one deliberate concession: Slagen gave up the right to know who asked for help.

“I told the team that I’ll never know how many times they met with the marketing coaches, what questions they asked, how bad their first build was,” he told the Marketing AI Institute on August 3. “My only ask was in three weeks you need to be ready to go.”

That is the design choice worth stealing. The fastest way to stall a skills push is to make the asking visible, because a senior marketer will quietly not learn a thing rather than be seen struggling with it in front of their manager. Slagen removed the audience and kept the deadline.

After the three weeks came a two-week build sprint, and the team sorted itself into three groups. Some built a dashboard. Some built a workflow or a loop. Some built an application, including the events team, which now runs a virtual conference with more than 100 speakers on software it wrote in Cursor and Zapier.

Slagen did not assign those tiers.

“Our team sort of self-placed themselves in three tiers of skill level, but this allowed me to then work with each tier and give them new things to think about/solve where everyone felt like they could win (at least within their own tier) and advance forward.”

The people who moved were not the ones on the org chart

Six months in, Slagen says his single best AI user is in customer marketing. The slower adopters were in the more technical roles, and he thinks he knows why: “they felt like they were already ahead of the curve.”

This is the finding most likely to survive contact with a smaller company. Existing seniority in a marketing function turned out to predict nothing about who could build. The people who had already been rewarded for their skill had the most reason to believe the new skill was a variation on it, and it wasn’t.

Zapier has run this play before. Anthropic’s own case study on the company describes a company-wide code red hackathon back in March 2023, when Zapier was a 360-person business, and puts current AI adoption at 89% of staff. That’s a vendor’s account of its own customer, so read the numbers with that in mind. What it does establish is that February was the second alarm, at a company that already believed it had answered the first one.

Slagen dated the change to the previous quarter. “The bar to quickly become great with AI got much higher at the end of 2025 as we first started seeing more non-technical folks start using coding agents,” he said. One marketer doing something unexpected in Claude Code read as a curiosity. Then it kept happening, and the curiosity turned into a job description.

He does not pretend it was inspiring

The most useful thing in the interview is the part Slagen is not happy about.

“Code Red worked probably more out of fear than inspiration. Yeah. I don’t, I don’t love, I don’t love that that was the case, but that’s probably what the case was.”

He offers two explanations, both unflattering. He was new in the role, so people may have been performing adoption for a manager they had not read yet. And some of them, he thinks, were genuinely frightened about their careers.

There is also a running cost he names plainly. The internal system the team built on top of all this, which Zapier calls its marketing brain, has a full-time owner. “None of this would work without the maintenance piece,” Slagen said. A five-person marketing team does not have a spare person, and any version of this that assumes one is a version that quietly dies in month four.

Leave the marketing brain aside. What travels to a 15-person company is the three moves underneath it: a fixed deadline, coaching that nobody reports on, and a public build where the work is visible even when the asking isn’t. None of those cost money.

The uncomfortable part travels too. Slagen got compliance in three weeks by telling a room of accomplished marketers that someone they had never met could take their job. He is right that it worked, and right to be uneasy that it did. Any leader running this play should be honest that the deadline is doing less work than the fear is, and plan for what happens when the fear wears off.

Related coverage: marketing job posts are dropping “specialist” and adding “head of”, Clay’s internal AI writing policy, and what happened when product teams stopped doing handoffs.