The most eye-opening product lessons sometimes come from listening to engineers and observing tech trends, not just customer research or OKRs. #ViteConf in Amsterdam reminded me how much PMs can learn by stepping into the developer world.

Agentic traces contain perfect information about an agent’s behavior with every plan, action, and retry. But that information gets lost in a sea of JSON. So we built AgentPrism: open source React components that turn traces into visual diagrams for debugging AI agents. You can

Enjoying @ViteConf – great talks, great people! ⚡️ Just caught Matt Kane sharing what’s next for Astro. Love seeing where the Vite ecosystem is heading.

"I couldn't sleep at night, if it all sucked..." @jeffreyhuber of @trychroma talks about how design could be a moat for a devtool company, and why for him this level of attention to detail is a baseline. Which is something he actually looks for in people when recruiting.

Glad Evil Martians could play a part in making Bolt v2 happen. 🤟 I've seen firsthand how much effort and change went into this release. Love how Bolt redefines app building – fantastic update!

📕AgentPrism Storybook is now live! Explore all React components and UI patterns from AgentPrism in an interactive playground. Dive into TraceViewer, TreeView, and various atoms—see exactly how each component handles traces. Try it: https://t.co/C3hoi7TWyC

This is an alpha release of Agent Prism — the APIs may evolve. We appreciate your feedback and contributions. If building with AI agents or distributed tracing, try it and share your thoughts!

Out-of-the-box with Agent Prism 🧩 - TraceViewer: browse traces, spans & details - Custom layouts: TraceList, TreeView, DetailsView - OTLP support for GenAI/LLM traces - Responsive + dev-friendly (React 19, Tailwind, TS) Code 👉

🚀AgentPrism is Here! A React component library for visualizing OpenTelemetry traces from AI agents—covering LLM calls, tool executions, and agent workflows. Instantly explore trace details and span hierarchies with TraceViewer, or customize layouts using modular components.

The takeaway: Early-stage AI startups need performance metrics. Growth-stage AI startups need operational effectiveness metrics. The handoff between the two is where money evaporates or where competitive advantage is born. 👉 What metrics are you tracking most closely right now?

2️⃣ Weird usage isn’t just a bug—it’s a business signal. We’ve identified abuse of a freemium model, anomalies in token spend, token anomalies, loopholes… all insights. Tracking usage vs. spend helped us decide whether to fix leaks or accept them as low-cost noise.

1️⃣ Operational effectiveness isn’t the same across audiences. In our case, solo users generate more $, but teams are more cost-effective. Still, it’s worth deeply analyzing both. For example, we found a tiny code misprint that let teams overspend by 3B tokens. Expensive lesson.

Two lessons from the trenches of one of our clients:

That’s when things get messy: Ready to “lose” 3B tokens? Or 4B? No idea? What if I say $100k… or $300k?

But here’s the trap: the very metrics that help you win PMF can blind you once you start to scale. At the growth stage, the question shifts from “How does this work for users?” to “How does this work for the business?”

Most AI startups I work with are laser-focused on performance metrics in the early days: accuracy, speed, reliability, consistency, task completion rate, and finally, user satisfaction. That’s exactly what you need when proving your solution and chasing PMF.

If you’re building AI-powered tools and existing trace platforms feel like overkill, let’s talk. We’re specifically interested in teams who strive to bring AI reasoning to end users, not just another debugging dashboard. Hit me up.

We’ve spent a decade making complex devtools actually usable (Tines, Bolt, many others), so we know the difference between “functional” and “delightful to use,” and we’re applying that same lens to AI observability.

We're reimagining AI trace observability. A few weeks ago, a former Evil Martians client reached out with an exciting challenge: AI trace visualization. While there are many solutions, there’s no great open-source library that ticks all the boxes. So we took this challenge.

https://t.co/C1rc8uH09e is taking the world by storm. In this article I dive deep into how @stackblitz made their code gen product grow from 0 to $4M ARR in a month, and how that’s affecting @supabase and everyone one of us, frankly https://t.co/fUmTtZZzLB @ericsimons40