What are the key takeaways from this How I AI episode?

Key insights include: **Project-Based Governance:** Structure enterprise AI around "projects" rather than individual sessions. Projects control model selection, tool access, and cost limits for specific teams or workflows. An HR team project, for example, can block agents from writing sensitive data to public documents while still allowing other tools — removing the all-or-nothing permission problem.; **Three-Layer Data Agent Triage:** Build data agents to query in priority order: first, a blessed analytics layer with pre-validated metrics; second, existing dashboards and their underlying queries; third, the raw data catalog. This hierarchy prevents agents from defaulting to brute-force warehouse queries, protecting infrastructure while improving answer quality and reducing token spend.; **Agent-Ready Infrastructure First:** Pre-AI investments in developer experience, data catalogs, and resilient query layers (Stripe uses Trino) directly determine how well agents perform. Agents amplify existing infrastructure quality and failure modes equally — a poorly indexed warehouse becomes catastrophically slow under agent load, while a well-tiered catalog enables accurate, efficient querying.

What did Sharadh Krishnamurthy discuss on How I AI?

Sharadh Krishnamurthy, engineering manager at Stripe, explains how a 1.5-person team built Kai, Stripe's internal AI agent, in two weeks. Now serving 10,000+ employees with under 10 staff, Kai combines personalized context, project-based governance, a shared skills platform, and agent-ready data infrastructure. Key topics include: **Project-Based Governance:** Structure enterprise AI around "projects" rather than individual sessions. Projects control model selection, tool access, and cost limits for specific teams or workflows. An HR team project, for example, can block agents from writing sensitive data to public documents while still allowing other tools — removing the all-or-nothing permission problem.; **Three-Layer Data Agent Triage:** Build data agents to query in priority order: first, a blessed analytics layer with pre-validated metrics; second, existing dashboards and their underlying queries; third, the raw data catalog. This hierarchy prevents agents from defaulting to brute-force warehouse queries, protecting infrastructure while improving answer quality and reducing token spend..

How long is this episode of How I AI?

This episode is 50 minutes long. SignalCast provides an AI-generated summary so you can get the key insights in about 3 minutes.

How I AI

Build your own company brain: the enterprise AI playbook from Stripe’s engineering team | Sharadh Krishnamurthy

September 7, 2026

50 min episode · 2 min read

Sharadh Krishnamurthy

Episode

50 min

Read time

2 min

Topics

Investing, Leadership, Artificial Intelligence

AI-Generated Summary

Published Sep 7, 2026

Key Takeaways

✓Project-Based Governance: Structure enterprise AI around "projects" rather than individual sessions. Projects control model selection, tool access, and cost limits for specific teams or workflows. An HR team project, for example, can block agents from writing sensitive data to public documents while still allowing other tools — removing the all-or-nothing permission problem.
✓Three-Layer Data Agent Triage: Build data agents to query in priority order: first, a blessed analytics layer with pre-validated metrics; second, existing dashboards and their underlying queries; third, the raw data catalog. This hierarchy prevents agents from defaulting to brute-force warehouse queries, protecting infrastructure while improving answer quality and reducing token spend.
✓Agent-Ready Infrastructure First: Pre-AI investments in developer experience, data catalogs, and resilient query layers (Stripe uses Trino) directly determine how well agents perform. Agents amplify existing infrastructure quality and failure modes equally — a poorly indexed warehouse becomes catastrophically slow under agent load, while a well-tiered catalog enables accurate, efficient querying.
✓Skill Platform Over Skill Builder: Rather than letting employees create isolated automations, build a shared skill library with retrieval, quality evals, and governance. Stripe maintains roughly 50 high-frequency skills used company-wide, plus 100–150 niche skills for specific teams. Automatic platform-generated improvement suggestions keep shared skills accurate as the library scales past 2,000 entries.
✓Human-in-the-Loop via Tool Policies: Assign tool policies at the project level to require human approval before sensitive actions execute — such as calendar writes or external document creation. This targets friction precisely where risk exists rather than applying blanket restrictions, preventing both rogue agent behavior and the user frustration that leads to unsafe workarounds.

What It Covers

Sharadh Krishnamurthy, engineering manager at Stripe, explains how a 1.5-person team built Kai, Stripe's internal AI agent, in two weeks. Now serving 10,000+ employees with under 10 staff, Kai combines personalized context, project-based governance, a shared skills platform, and agent-ready data infrastructure.

Key Questions Answered

•Project-Based Governance: Structure enterprise AI around "projects" rather than individual sessions. Projects control model selection, tool access, and cost limits for specific teams or workflows. An HR team project, for example, can block agents from writing sensitive data to public documents while still allowing other tools — removing the all-or-nothing permission problem.
•Three-Layer Data Agent Triage: Build data agents to query in priority order: first, a blessed analytics layer with pre-validated metrics; second, existing dashboards and their underlying queries; third, the raw data catalog. This hierarchy prevents agents from defaulting to brute-force warehouse queries, protecting infrastructure while improving answer quality and reducing token spend.
•Agent-Ready Infrastructure First: Pre-AI investments in developer experience, data catalogs, and resilient query layers (Stripe uses Trino) directly determine how well agents perform. Agents amplify existing infrastructure quality and failure modes equally — a poorly indexed warehouse becomes catastrophically slow under agent load, while a well-tiered catalog enables accurate, efficient querying.
•Skill Platform Over Skill Builder: Rather than letting employees create isolated automations, build a shared skill library with retrieval, quality evals, and governance. Stripe maintains roughly 50 high-frequency skills used company-wide, plus 100–150 niche skills for specific teams. Automatic platform-generated improvement suggestions keep shared skills accurate as the library scales past 2,000 entries.
•Human-in-the-Loop via Tool Policies: Assign tool policies at the project level to require human approval before sensitive actions execute — such as calendar writes or external document creation. This targets friction precisely where risk exists rather than applying blanket restrictions, preventing both rogue agent behavior and the user frustration that leads to unsafe workarounds.

Notable Moment

When describing agent failures during Kai's early rollout, Krishnamurthy noted that agents nearly took down core Stripe systems on multiple occasions. The team caught each incident in time, but the pattern revealed that agents don't introduce new failure modes — they dramatically amplify the ones already present in existing infrastructure.

Know someone who'd find this useful?

Episode Transcript

Agents are very creative at bringing your infra down. It turns out that agents just, like, dial up all your failure modes. It just multiplies the amplitude of problems you can get. There were agents that went rogue. There were agents that may have almost taken down core systems. One of the cool things about projects that can be very concrete for people is the idea of tool policies. Let's say you're a person on the HR team who's dealing with a bunch of sensitive information. You really don't want the agent to son of a go rogue and put that sensitive data into some public Google document that all Stripes can access. But you also don't want to tell them, oh, you can't use any tools because your workloads are too sensitive. I love this idea of this, like, three layer triage that a data agent can go through, and that's really smart. Find existing reports, then use the analytics layer to find the right query. And then if, you really have to fall down to the data catalog and write your own query, your data warehouse has to be very resilient to high volume queries because when in doubt, an agent will just brute force it. We could test this out and it's going to kick off what is a human in the loop workflow. Create a calendar invite for me and Wong tomorrow at 11AM Pacific. This isn't the fun part. The fun part is what I showed before. Agents are really good. They're very creative. So we got to put some restrictions on them so they don't go rogue. Welcome back to How I AI. I'm Claire Vaux, product leader and AI obsessive here on a mission to help you build better with these new tools. Today, we have Sharad, an engineering manager at Stripe and part of the team who built Kai, their internal company brain and company agent. He's gonna show us why you might wanna build your own custom agent for your company, what are the governance and control mechanisms of Kai that make it super special, and how to build not just a skill building skill, but a skill building platform for your team to share their automations and workflows with the rest of the company. Let's get to it. This episode is brought to you by DX. In a recent study across more than 500 engineering organizations, DX found that spend on AI tools has grown 28 x over the last year. The share of AI authored code is climbing, but overall innovation has remained flat. As teams generate code faster, new friction in code review and validation is offset those early velocity gains. DX tracks speed, quality, and cost together across the software development lifecycle, giving engineering leaders clear visibility into how AI impacts delivery and whether those investments are translating into real value. Download the full report at getdx.com/howiai.getdx.com/howiai. Shared, it's so nice to have you here, and I …

Get the full transcript (9,293 words) + summary by email — free

One-time email with the complete transcript and AI summary of this episode. No account needed.

One email, no spam. We’ll also show you what SignalCast does.

Browse all How I AI transcripts →