The Number That Matters
Stripe's internal AI platform, Kai, hit its quarterly adoption target in one week and scaled to 5,000+ users in roughly four weeks. Today, 83% of Stripe uses it weekly. Marketing leads at 95% adoption. GTM teams sit at 87% — ahead of engineering.
These aren't vanity metrics. They're a signal about where enterprise AI actually lands when it's designed correctly. The full breakdown is in Lenny Rachitsky's interview with Sharadh Krishnamurthy.
Why GTM Teams Led Adoption
Most enterprise AI rollouts fail the same way: the tool gets built for technical users, distributed to everyone, and then quietly shelved by anyone who isn't a developer. Stripe inverted this.
Kai was designed explicitly for non-engineers — sellers, finance, ops, marketing. The interface required zero configuration. It lived inside Slack and Google Suite, where those teams already spent their days. Sessions were self-contained, meaning a seller doing deal prep didn't need to understand prompt engineering or manage a persistent context window. They just asked questions and got answers grounded in Stripe's internal data.
That design choice — meet people where they already work instead of pulling them toward developer tooling — is why GTM outpaced engineering. The use cases (deal prep, cross-functional data synthesis, competitive analysis) mapped directly to work those teams were already doing manually.
What Non-AI Companies Should Build From This
Stripe is a technology company, but Kai's architecture is not a technology company's privilege. The underlying logic is replicable for any company with internal documentation, CRM data, and a Slack instance. Here's what the blueprint actually requires:
1. Start with the data layer, not the interface. Kai works because it's grounded in Stripe's internal knowledge — not generic LLM output. Before you build anything, audit what structured knowledge actually exists in your org and where it lives.
2. Design for the lowest-friction user, not the most technical one. If your rollout requires training sessions to explain how to use the tool, the adoption curve will be slow and fragile. The right test: can a seller on their third week use it without asking IT for help?
3. Anchor use cases in existing high-frequency workflows. Deal prep, pipeline reviews, and competitive briefings happen every week regardless of whether AI exists. Dropping Kai into those workflows means adoption doesn't require behavior change — it just accelerates what's already happening.
The actionable move this week: Identify one high-frequency workflow in your GTM or ops function where employees are currently synthesizing information manually — pulling from Slack, docs, and CRM simultaneously. That's your Kai v1 use case. Define the data sources it would need to be useful, and scope the interface around that single job-to-be-done before expanding.