Thoughts on product management, building with AI, and learning in public.

Two transformations are underway in product design. The first is about process: new tools, new workflow, new role. The second is about environment: the products designers build now exist in a world where AI agents can modify, bypass, and route around them. Most coverage of AI and design focuses on the first. This piece covers both.

The products that will ride the AI wave aren't building better chatbots. They're making specific types of context frictionless to collect and frictionless to share. A framework for thinking about where that opportunity is and how it plays out.

SaaS averaged everyone's needs into one product. Personal software goes the other direction. One tool, one person, built right. The interesting part is what happens when you open-source what you built for yourself.

When users start patching your product with coding agents, that is not a threat. It is the clearest signal you will get about what they need, delivered with a working prototype attached.

Most organizations treat AI adoption as a procurement problem. It is a systems problem. Three thinkers explain why implementation usually fails and what success looks like from the inside.

Some products have built themselves around capturing context that AI cannot derive on its own. What they are actually doing, and why portability is the moat.

Every major engineering organization that has published its agent architecture made the same foundational choice: give agents a real development environment. Here is what they built, and how to replicate it.

OpenAI formalized "harness engineering" — designing constraints and feedback loops for AI agents. I've been doing this from the product side for a year. Here's what they're missing.

Everyone says "don''t boil the ocean." That advice assumed the cost of ambition stayed high. It doesn''t anymore.

A designer with no coding background built a personalized book recommendation engine in eight hours. Not for a client. Not for a startup. For herself.

Most teams optimize UI when the problem is system structure.

How explicit weights expose hidden assumptions and make better decisions at scale.

VCs fund demos. Organizations fail because demos dont translate to daily use.

Creative professionals have emotional stakes in their tools. Design for that.

Discovery sessions get repeated. Decisions get unmade. Here is how to fix it.

The tools that make individuals productive with AI can actively harm team collaboration if not redesigned for sharing. Here's a framework for getting it right.

The industry bet on removing humans. The systems that actually work do the opposite.

Multi-dimensional scoring with configurable weights beats gut feel every time. Here's the pattern I used to improve my job search response rate from 5% to 40%.

I built a CLI that turns Claude Code into a job search co-pilot. It qualifies listings, researches companies, and drafts applications—all with human approval at every step.

Control Alt Elite is a small working group I started to make AI exploration practical and social—less "AI inspiration theater," more "I used this on real work and it saved me time.

How I built samzoloth.com using Next.js, Claude Code, and an obsessive attention to interaction design.

How I built a personal AI infrastructure around Claude Code that learns my preferences, runs background automation, and coordinates multiple agents.

The challenges of documenting a live engagement, balancing confidentiality with storytelling, and the technical decisions behind embedded prototypes.

Claudio voice assistant, Pastey clipboard manager, shopping cart prototype, and 57 Claude sessions.

Dave McMahon's storyboard approach to AI pitches, and why boxes-and-arrows diagrams have already lost.

Personal site launch sprint and setting the team up for success during my time off.

Personal site from scratch and translating stakeholder needs into actionable backlog.