Source: LangChain — 2026-08-03
Summary
LangChain published a case study on how Stripe built "Kai," a company-wide knowledge AI platform, in roughly one week using LangChain's Deep Agents harness. Kai grew from 296 users to more than 5,000 in four weeks, now spans 1,000+ skills across 100+ teams, and reports 83% weekly-active usage across 60,000+ sessions — a real production-adoption story rather than a pilot.
Key Takeaways
- The build timeline — about a week from start to internal launch — is the headline claim, and it's specifically credited to building on an existing agent harness (Deep Agents) rather than a custom framework from scratch.
- User growth from 296 to 5,000+ in four weeks, with 83% weekly-active usage, is a real adoption curve, not a vanity pilot metric — most internal AI tools never clear the "people actually keep using it" bar.
- Kai spans 1,000+ skills across 100+ teams, meaning it's structured as a broad internal capability layer rather than a single-purpose chatbot for one department.
- 60,000+ sessions is enough volume to treat the reported usage numbers as a real production signal rather than an early-access anecdote.
Reel Script
Hook (18s)
Most companies spend months building an internal AI tool that a handful of people actually use. Stripe built one in about a week — and four weeks later, five thousand employees were using it every week.
Core Concept (100s)
The interesting part of this story isn't "Stripe built an AI chatbot" — every company is doing that. It's the build time and the harness choice. Stripe built Kai on top of LangChain's Deep Agents framework, which is a pre-built scaffolding for agents that can plan multi-step work, call tools, and manage longer-running tasks — instead of Stripe's engineers writing that orchestration logic themselves from scratch, which is normally where internal AI projects burn months before they even get to the point of being useful. Think of a harness like Deep Agents as pre-built plumbing: the pipes, the joints, and the valves are already there, so the team building on top of it is doing the equivalent of wiring fixtures into a house instead of laying pipe under the foundation. That's what let a team go from zero to an internally-launched knowledge platform in about a week, instead of the usual multi-month internal-tools timeline.
Hands-On (105s)
The adoption numbers are the receipts here, and they're the part worth putting on screen: 296 users at launch, climbing to over 5,000 within four weeks — that's roughly a 17x jump in a month, not a slow trickle. On top of that, 83% weekly-active usage across more than 60,000 sessions, and the platform now covers 1,000-plus distinct skills spread across more than 100 teams inside the company. The reason those last two numbers matter together — skills and teams — is that it shows Kai isn't a single-purpose tool bolted onto one department; it's structured as a broad internal capability layer that different teams are extending with their own skills, which is a very different shape than a typical one-off internal chatbot pilot.
Takeaway (26s)
If your team is staring down a multi-month internal-AI-platform roadmap, the lesson here is blunt: building on an existing agent harness instead of custom orchestration is what got Stripe from zero to production adoption in about a week. Look at what Deep Agents — or an equivalent harness — actually handles for you before your team writes that plumbing themselves.