How Speakeasy Builds Its Product with Claude Code | Sagar Batchu, Speakeasy CEO

How does a dev tools startup use AI to build its own product? Sagar Batchu, co-founder and CEO of Speakeasy, opens up Gram, Speakeasy's open-source MCP platform, and shows the setup that lets coding agents work in a real codebase. Then Claude Code builds a new feature live. In this session: The CLAUDE.md that guides Claude Code through Gram's codebase: project structure, coding guidelines and data-model rules Boyd's Law of Iteration: small, atomic steps the agent can verify one at a time mise tasks for lint, build and SDK generation, so the agent can check its own work A live build: Claude Code adds a tool-call counter across the backend, the SDK and the dashboard, then debugs from a screenshot (it works by the end of the session) How Speakeasy uses internal MCP servers so anyone on the team can look up customers, workspaces and SDKs from chat How the team handles model changes: evals, a strong test harness and code review Why Speakeasy open-sourced Gram Gram demo: from an OpenAPI file to a hosted MCP server, with curated toolsets and custom tools that chain API calls Recorded live as a free Maven Lightning Lesson on October 20, 2025. Links from the session: Gram on GitHub: https://github.com/speakeasy-api/gram Gram's CLAUDE.md: https://github.com/speakeasy-api/gram... Speakeasy AI Control Plane (getgram.ai now lands here): https://ai.speakeasy.com/ mise: https://mise.jdx.dev Model Context Protocol: https://modelcontextprotocol.io Official MCP Registry: https://registry.modelcontextprotocol.io Smithery: https://smithery.ai MCPJam: https://www.mcpjam.com/ Goose: https://goose-docs.ai OpenAPI Initiative: https://www.openapis.org/ Learn to build what agents want - better APIs, MCP servers, CLIs, Skills, and plugins for ChatGPT and Claude - in my live course, Build What Agents Want. Code SPEAKEASY: https://maven.com/emmanuel/build-ai-r... More free lessons:    • Building APIs quickly with AI tools   Connect on LinkedIn: Sagar Batchu:   / sagarbatchu   Emmanuel Paraskakis:   / emmanuelparaskakis   Chapters 0:00 Welcome: how Speakeasy builds its product with AI 0:50 Sagar Batchu on starting Speakeasy 3:19 Speakeasy's challenge: build fast and lead on AI 5:19 The open-source Gram repo 6:36 CLAUDE.md: a guide for your coding agent 7:20 Boyd's Law: small steps the agent can verify 9:10 mise tasks the agent can run and check 13:10 Live demo: Claude Code adds a feature to Gram 18:01 One command to bring up the whole platform 19:26 Q&A: AI for non-AI products 20:18 Internal MCP servers for company operations 25:14 Q&A: team roles and picking frameworks 27:33 Running the feature and debugging with screenshots 31:50 Q&A: handling model changes (evals, tests, code review) 33:26 Why Speakeasy open-sourced Gram 34:36 From an OpenAPI file to an MCP server 36:37 When an API is (and isn't) a good MCP server 38:49 Toolsets: curating the tools you expose 40:05 Custom tools that chain API calls 43:06 Resources from the session 44:48 More free lessons, and the course 46:30 Q&A: the demo works after all #ClaudeCode #MCP #AIagents