Blog
Notes on building AI products.
Short writing on taking AI ideas into production: product definition, architecture, and the reliability work that sits between a demo and a customer.
Let’s talk about your productFeatured posts
Featured · 5 September 2026
AI Can Build Your Product. It Can't Own It.
AI is at its best on greenfield. Maintenance runs on context, and nobody has context for code they never wrote. Why the handoff from AI-built product to human-owned system is the problem worth solving.
- AI agents
- software architecture
- startup engineering
Featured · 4 August 2026
Don't Build Another AI Assistant Into Your SaaS. Make It Agent-Ready.
For SaaS products, the winning move isn't a built-in chatbot. It's MCP: let customers connect the agents they already use. Why that's stickier, what's hard about it, and where new founders have a real shot.
- MCP
- AI agents
- SaaS
Published posts
28 September 2026
AI Made Building Faster. It Didn't Make Thinking Faster.
Shipping with AI can happen overnight. Good product decisions still need time to cook. Why founders should not let tooling outrun judgment.
- AI product development
- product judgment
- startup engineering
- AI coding
- AI MVP
24 September 2026
SWE-bench Won't Tell You Which Model Belongs on Your Codebase
Public coding benchmarks miss what matters in production: codebase fit, planning, debugging, and how much human cleanup the model leaves behind.
- LLM coding benchmarks
- SWE-bench
- AI coding agents
- production codebase
- agent evaluation
18 September 2026
Nobody Believes Your SaaS Pitch. They're Usually Right.
Early SaaS founders face a credibility tax. Most pitches deserve skepticism. Here's how the rare 1% earn belief without shouting louder.
- saas
- startups
- founders
- credibility
16 September 2026
AI Agents Don't Fund Open Source. Status Does.
Open source is unpaid social infrastructure funded by status, craft, learning, and career signaling. If the maintainer class shrinks, agent imports do not replace the incentive model.
- AI
- open source
- agents
- software engineering
- incentives
15 September 2026
We Built a Support Agent. The Safe Version Barely Needed an LLM.
Once a support agent is safe enough for refunds and legal answers, the LLM often shrinks to intent classification and slot filling. Ask whether that is product value or theatre.
- AI
- LLM
- customer support
- agents
- architecture
11 September 2026
AI Won't Cause Mass Unemployment. It Might Make Knowledge Scarce Again.
The internet made knowledge cheap and applied skill scarce. AI is flipping that. The interesting question for founders isn't mass unemployment — it's what becomes valuable when skill gets automated.
- AI
- job market
- startups