💡 A simple realization that can have a dramatic effect on a startup. (This post expands on Step 1 in the 5-step framework on a startup exit I wrote about last week) 🧠 Existing wisdom: "Keep focused on building a company, and an exit naturally happens." 💡💡💡 Realization:

My Real “AI Dev” Moment with Cursor The moment it clicked? I asked: “v0 generated Shadcn components instead of Tailwind. Can you undo all changes?” Cursor: ✅ Rolled everything back. I wasn’t reviewing commits. I wasn’t debugging file-by-file. I was talking to my codebase.

Why Auth Should Be Part of v1 Now: Clerk makes this possible During prototyping, I’ve always treated authentication like a “nice-to-have” It came after the landing page. After the UI. After I was “sure” I had something worth showing. But with Clerk, auth became a 10-minute

1/7 Model Context Protocol Model Context Protocol isn’t just a spec. It’s the beginning of protocol-native agents. (I’ve been unpacking MCP over the last few weeks. This thread series is what I wish existed when I started. Not an intro. Not a pitch. Just insight from the

2/7 Model Context Protocol The M×N integration trap — and how MCP breaks it. Today, integrating AI models with tools feels like the early days of app dev: - Custom glue code per tool - Hard-coded function calls - No discoverability If you have M AI apps and N

3/7 Model Context Protocol Why tools alone aren't enough. Most devs assume LLM agents just need tools. But real-world agentic behavior requires more. MCP gives us a richer foundation: - **Tools** = actions the model can take - **Resources** = data the model can use

4/7 Model Context Protocol How I’d expose Jenkins via MCP - protocol-native DevOps Designing an MCP server is more conceptual than code (for now). I mocked out a Jenkins MCP server recently. It would expose: - trigger_build (tool) - build_logs (resource) -

5/7 Model Context Protocol What breaks when agents start calling tools autonomously? MCP opens powerful doors. It also widens the threat surface: - Prompt injection is real - Non-determinism at the protocol edge is risky - Security is decentralized (for now) I’m

7/7 Model Context Protocol In a few years, “MCP-compatible” will be the new "REST API" Just like “REST API” became a checkbox for modern services, I believe “MCP compatibility” will be a baseline for LLM-aware tools. - Not to replace UIs - But to enable agents - I

Every AI agent — from the simplest chatbot to the most advanced swarm — is just this loop + these 5 dials #ai #aiagents aientrepreneurship.substack.com/p/the-dna-of-a…

Every AI Agent — from simple to multi-agent swarms — runs the same loop: Perceive → Think → Act → Remember → Learn. I mapped Goose by Block to this DNA… then turned it into a coffee-making Barista Agent ☕ Same loop, different tools aientrepreneurship.substack.com/p/mapping-goos… #AgenticAI