Same task. Different approaches.
I give different models the same prompt and constraints, then look at what they produce. What works? What breaks? What needs fixing before I’d actually use it?
Practical AI · coaching · adoption
I build practical AI tools, test where they help, and help people get comfortable using them. My focus is AI enablement and adoption: useful workflows, clear instructions, and employee coaching. IT support roots. Creative technology still on the workbench.
Practical AI + creative technology
Useful tools, memory experiments, and a city made from GitHub contributions. Start with the case studies, or see all builds →.
A help desk demo with editable AI summaries and reply drafts. Local Ollama or optional Grok, with human review and a small 12-ticket evaluation.
Read the case study →
AI · MemoryA small local assistant that keeps explicit facts, fact history, and source status separate from recent chat. The experiment shows what changes when “what it currently believes,” “what used to be true,” and “what was supplied” are not one pile of text.
Read the case study →
Creative technologyA year of GitHub contributions becomes a 3D city. Every day is a lot, activity becomes height, and the original history stays underneath the spectacle.
Read the case study →
Writing
The tools I'm using, the things I'm building, and questions that come to me as I learn and experiment.
Notes that talk to each other, transcripts into memos, and reviewing an AI edit like a code diff: how people use Cursor for work that isn't code.
Bake-offGemini spent complexity on the art. Astra spent it on efficiency. Grok spent it on machinery.
MemoryI corrected my assistant from teal to violet. Was the fix saved, supplied, and understood, and did the work that depended on it change too?
A portable behavior layer for coding agents. Notes coming once I write them like I mean them.
Cursor, Grok, and the rest — compared by what each is unusually good at, not by a leaderboard.
Practice
I give different models the same prompt and constraints, then look at what they produce. What works? What breaks? What needs fixing before I’d actually use it?
I walk people through a useful task, let them try the tool, and explain what to check before using its answer. A good suggestion should help someone understand the work, too. Read about the agents, coaching, and instructional materials →
Quick experiments and updates go on X. Here, I have room to explain what I tried, what happened, and what I learned from it.