Emil @emilsnotes

x.com/i/article/2099…

Super excited to be working with @chainguard_dev, the $3.5bn security rocketship. With a deeply technical product, getting users up and running means helping them work through a lot of technical questions. @kapa_ai helps Chainguard users onboard faster and find answers themselves, freeing up the team’s engineers from answering the same questions over and over. Awesome to be building together!

Excited to be partnering with @harnessio! Harness has a deeply technical platform, and we're working together to make that knowledge easier to access for both developers and AI agents. It starts with accurate answers grounded in Harness's documentation, helping developers work through technical questions and giving agents the context they need to be useful (MCP is a gamechanger here). Excited to keep building together!

We'll be serving delicious cookies and great coffee tomorrow and Thursday morning 8.30-10AM! there may or may not be merch also luma.com/legora-jpfg

@yoheinakajima @lawsonelliott awesome! thanks

@yoheinakajima @lawsonelliott awesome! out of curiosity, how did you build data pipelines from all the different sources? (e.g., to pull data and convert to agent friendly formats)

@yoheinakajima @lawsonelliott do you then sync data from other sources into a repo (slack, notion etc) + keep them up to date?

a useful hack for understanding where your agents struggle: ask them. we added a simple feedback tool to ours: > agent encounters something broken or missing > decides whether it’s a meaningful issue > reports it with the context from the task > team gets structured feedback instead of digging through traces it's way more high signal than looking at agent traces.

Aaron Levie @levie

3 weeks ago

Systems of record have never been more important than in a world where you have AI agents that will do 100X more work on these platforms than people ever did. Agents will be querying the data in these systems, processing tasks, executing workflows, collaborating with human and

117 81 912 158K 751

Agentic commerce isn't starting with sneakers. It's starting with semiconductors. Telink Semiconductor, 2 billion+ chips shipped, powering IoT devices for Logitech, Panasonic and Harman, just went live with the first agentic product selector in the industry (built on @kapa_ai). Before: → Read datasheets across 80+ SKUs, dozens of parameters each → Wrestle a 15-dropdown filter table hoping you guessed the right specs → Or email a sales engineer and wait After: → Type "I'm building a Matter smart plug" and get hard spec filters applied over the live catalog → Compare parts side by side in rendered cards, straight to the product page → No exact match? It says so and recommends the closest tradeoffs. Why complex B2B first? Consumer agentic checkout keeps stalling because people don't trust agents to spend their money. Here nobody's asking the agent to buy, just to narrow 80 SKUs to 3 against a spec sheet. Exactly what these models are great at. Agentic commerce doesn't start where trust is hardest. It starts where retrieval is hardest. Excited to partner with the Telink team on this 🚀

is there appetite for a tutorial on how we built our own internal ai sre with @LangChain deep agents that got instant internal PMF overnight?

@jeffreyhuber this is awesome! huge congrats man

83% of @stripe now uses their internal AI agent weekly. Not because they picked the right model, but because of one unglamorous layer buried near the end of their write-up: retrieval. Stripe's Kai is connected to 1,000+ internal skills and tools. The hard problem wasn't the agent loop, it was deciding what the agent gets to see for a given question. Coding agents rarely face this. The repo already organizes everything for them. Knowledge work has no such structure. So Stripe built a retrieval layer that picks the right context before the agent does anything, and by their own account that's what makes answers correct by default. We see the same across 200+ companies at @kapa_ai. Agents rarely fail because the model isn't smart enough. They fail because the wrong context got pulled in and the agent ran with it. Everyone debates models and harnesses. Retrieval quietly decides whether your agent is right. x.com/emilygsands/st…

Emily Sands @emilygsands

2 months ago

Coding tools changed what’s possible for engineers. We took that same model and asked what could give more Stripes—sales, finance, ops—that same boost. Here’s how we built Stripe’s Knowledge AI Platform. x.com/i/article/2082…

62 49 544 296K 1K

We just wrapped up the latest @kapa_ai quarterly offsite in Copenhagen. Every 90 days we fly the whole team in (from US, Europe, Asia) and zoom out on the entire company: what worked, what didn't, and what the next 90 days needs to look like. We still operate on YC time. 90 day sprints, set ambitious goals, ship, take stock, repeat. It's kept us fast even as the team has grown a lot in the last year. Plus the usual hackathons and a lot of Copenhagen evenings along the way. 🇩🇰 See everyone in 90 days. 👋

keyboard_arrow_left Previous keyboard_arrow_right Next

"Most vendors disappear into the background after the sale." A customer said this on a quarterly review call last week. Luckily, about others they'd worked with. It's the comment I'm proudest of all year. In 2026 where software is becoming cheaper and demos easier to build, the real work starts after the sale: → Serving customers every quarter and growing with them → Scaling beyond the initial use case into ones nobody scoped at signing → Surrounding them with a rockstar team that cares as much as they do That last one is the whole game. The hardest thing to build in 2026 is a team that keeps showing up for customers. Get that right and everything else compounds.

@liam_at_shopify @klaviyo 100%

Excited to share that @klaviyo is a kapa.ai customer. Klaviyo has a deeply technical platform, and their AI agent now lets developers self-serve answers to complex API and integration questions instantly, 24/7. Under the hood, kapa acts as the retrieval layer across their entire knowledge surface - help center, developer docs, API specs, academy - continuously kept up to date. Excited to keep building together!

@mattwensing save us from the slopfest that is 2026

@aparnadhinak Awesome! Congrats!!!!

6 users asked our AI agent the same question last week. It couldn't answer any of them. The question: can the Slack bot be limited to specific source groups? It worked for the widget, the SDK, the API... everything except Slack. Every time our agent can't answer something, it flags a knowledge gap, and we get these grouped into weekly reports ( @kapa_ai Coverage Gaps, screenshot below ). Then you guessed it: feed to Claude Code and open a PR. Here's what I find neat: nobody would have ever opened a support ticket for this. It's not a bug, nothing is broken. It's pure ergonomics. But 6 users hitting the same wall in a week means it matters. This is the feedback that used to just evaporate. Now it becomes a PR. Your users are writing your roadmap every day. The only question is whether you're reading it.

Excited to share that @alteryx is now a @kapa_ai customer. Alteryx is a deeply technical product, and in a matter of weeks they launched an AI agent that lets customers self-serve answers to complex technical questions, 24/7. Super excited to keep building together!