I Shipped a Self-Improving Agent in an Afternoon - Tyler Hogarth, Prisma

Tyler Hogarth, head of engineering at Prisma, shows how he built Gremlin, a self-improving coding agent his team now depends on, in one afternoon. Gremlin is a cloud agent that lives in Slack. It has access to Prisma's GitHub repos, secrets and knowledge base, runs on Prisma Compute and Prisma Postgres, and handles everything from small bug fixes to full features. Support, finance and product people kick off work through it too. Running coding agents on engineers' laptops sounds fine. In practice, it breaks. Local machines can slow to a crawl because every plan, side project, and CVE investigation spawns its own subagents. Small fixes can pile up. And the context for what to build sits in Slack threads, one step away from an agent that could build it. You'll see how he tackled this by putting a Mastra agent between Slack and an OpenCode sandbox. It pushes back on vague briefs, limits what the sandbox can reach, and reviews the results before anything gets merged. You'll also see how Gremlin improves itself, what it cost to build, and why open-weight models deserve a closer look. Recorded at the TypeScript AI Conference in London, July 23, 2026, presented by Mastra. 🔗 MENTIONED IN THIS TALK Prisma: https://www.prisma.io Agentic Engineering at Prisma: https://www.prisma.io/blog/agentic-engineering-at-prisma OpenCode: https://opencode.ai Sentry Seer: https://sentry.io/product/seer/ 📚 MASTRA RESOURCES https://mastra.ai https://x.com/mastra_ai https://mastra.ai/community/discord https://github.com/mastra-ai Learn Mastra: https://mastra.ai/learn https://mastra.ai/books/principles-of-building-ai-agents https://mastra.ai/books/patterns-of-building-ai-agents WHAT IS MASTRA Mastra is an open-source TypeScript framework designed for building and shipping AI-powered applications and agents with minimal friction. It supports the full lifecycle of agent development — from prototype to production. CHAPTERS 00:00 Intro 00:38 Three problems worth solving 02:41 Model agnostic, nothing rebuilt 03:31 What Gremlin is 04:00 The first architecture 04:37 Putting a Mastra agent in the middle 05:21 Security and review 06:16 Prisma Compute and Prisma Postgres 06:40 Three demos, three merged PRs 07:34 Gremlin at work 10:16 Self-improving, with safeguards 11:07 What it cost 12:12 Lessons learned 14:37 Your job is the harness