Stripe Ships 1,300 PRs a Week With AI. Your Sales Team Can't Send 10 Emails?

Stripe's engineering team now ships about 1,300 pull requests per week using AI agents. Not as drafts for humans to rewrite. As completed PRs that go through code review and merge.

Steve Kaliski, a Stripe engineer, explained the system on Lenny's Newsletter. The agents are called "minions." An engineer adds an emoji reaction to a Slack message describing the work. The agent picks it up, writes the code, opens a PR. A human reviews and merges.

The part worth paying attention to isn't the 1,300 number. It's this:

Activation energy matters more than execution capability.

That sentence reframes the entire AI adoption conversation. And it applies to sales teams just as much as it applies to Stripe's engineers.

---

What Stripe actually solved

Stripe didn't build a better code-generation model. They used existing tools — their open-source agent harness Goose, cloud development environments, git worktrees. The AI itself is table stakes.

What they built was a trigger mechanism so frictionless that non-engineers started using it to ship code. A Slack emoji. That's it. No separate app, no context switching, no configuration screen. See a task in Slack, react with an emoji, get a PR.

The result: adoption spread beyond the engineering team. People who had never written code were shipping changes through minions. Not because the AI got smarter. Because starting got easier.

Activation energy in sales

Most sales teams evaluating AI tools focus on capability. Can it write good emails? Can it research companies? Can it personalize at scale? These are reasonable questions. They're also the wrong starting point.

The better question: what happens at 8:47 AM on a Tuesday when your rep sits down with coffee and 40 accounts to work?

If using the AI tool means:

1. Opening a separate application
2. Pasting in account context
3. Configuring the message type
4. Reviewing and editing the output
5. Copying it back to the outreach tool

...then you've added five steps before a single email goes out. The rep will open LinkedIn instead. Not because LinkedIn is better. Because opening LinkedIn takes one click and produces dopamine. Your AI tool takes five clicks and produces a draft that needs editing.

That's activation energy. The gap between "I should do this" and "I'm doing this."

Why reps revert to old habits

Sales managers watch this pattern constantly. A new tool gets rolled out. Training happens. Week one, usage spikes. Week three, it's halved. Month two, three reps use it regularly and the rest have quietly abandoned it.

The usual diagnosis: "The team needs more training." Or: "The tool isn't good enough."

The actual problem, almost always: the tool adds friction to a workflow that was already frictionful. Reps don't abandon tools because the output is bad. They abandon tools because starting is hard.

Stripe figured this out for engineering. The equivalent for sales would be a system where the rep's morning looks like this: open the tool, and five companies with context and draft outreach are already waiting. No setup. No configuration. No research phase. Just review, adjust, send.

The distance between "sit down" and "first email sent" determines adoption. Not the quality of the tenth email.

Three lessons sales teams should take from Stripe

1. Build on surfaces your team already uses.

Stripe didn't create a new app for their agents. They put the trigger in Slack, where engineers already spend their day. For sales, the trigger should live inside the CRM, email client, or Slack — wherever reps already are at 8:47 AM. A standalone AI dashboard that requires a separate login is dead on arrival.

2. Make the default state "ready to act."

Stripe's agents don't wait for a detailed prompt. An emoji on a Slack message is enough context. For sales, the system should present ready-to-go outreach when the rep shows up. Not a blank canvas where the rep has to tell the AI what to do. The AI should already know: here are five companies showing buying signals this week, here's why each one matters, here's a draft for each.

3. Measure time-to-first-action, not features.

Stripe's metric wasn't "how good is the code." It was "how fast does work start." For sales, track the time between login and first outreach sent. If that number is 45 minutes (research, list building, drafting), and a tool brings it to 5 minutes, adoption will stick. If the tool is technically superior but the time-to-first-action stays at 30 minutes, reps will churn off it.

The machine-to-machine future

One more detail from Kaliski's talk worth noting. He demoed AI agents autonomously spending money through Stripe's machine payment protocol — agents transacting with third-party services to accomplish tasks.

This points somewhere interesting for sales. If AI agents can research, draft, and eventually send outreach autonomously, the role of the rep shifts. Less "do the work" and more "review the work and make judgment calls." The system handles everything before the close. The human handles the close.

We're not there yet. But the trajectory is clear. And the teams that figure out activation energy now will be positioned to adopt the next wave of autonomy when it arrives.