Building FantasyAgent: Three Weeks with Claude Code
A product manager's experience building a fantasy football draft assistant with Claude Code: player ID mapping, superflex rankings, Yahoo OAuth, and a live draft.
Read More →Writing
A monthly read on what happened in AI and what it means for people building products, from a PM in AWS Observability.
September 2026: GPT-6, Cheaper Frontiers, and Watching the Agents
GPT-6 arrived, frontier quality dropped to mid-tier prices, and tiered cyber access became standard. Plus a look at CloudWatch Omni and a recap of the last seven months.
Read More →August 2026: Google Resets, Meta Enters the Terminal
Google has the chips, data, and cash, and it still reorganized its AI leadership. When everyone has money and compute, org design decides who ships.
Read More →July 2026: The Sandbox Didn't Hold
Models told to score well on a cyber eval hacked Hugging Face to steal the answers. The victim noticed first. If you run agents, that is the part to worry about.
Read More →June 2026: Washington Reaches for the Release Button
Fable 5 was offline for 18 days because of an export order. Model availability is now a vendor risk, and most teams can't tell which workflows depend on which model.
Read More →May 2026: Renting a Rival's Supercomputer
Anthropic rented a competitor's data center and passed OpenAI in valuation. Google built I/O around its cheapest fast model. Compute and cost per call drove the month.
Read More →April 2026: The Model Anthropic Wouldn't Ship
Mythos could find and exploit decades-old bugs on its own, so Anthropic gave it to defenders first. Release decisions became security decisions.
Read More →March 2026: The Money Outran the Models
GPT-5.4 put computer use in the base model and OpenAI raised $122 billion. The shutdown of Sora said more about where AI is headed than either.
Read More →Claude Sonnet 4.5 and the 30-Hour Coding Session
Claude Sonnet 4.5 leads coding and computer-use benchmarks and worked for more than 30 hours on one task in trials. What that means for software teams.
Read More →The Corporate AI Wars: OpenAI's GPT-5 Gambit and Anthropic's Government Play
Anthropic cut off OpenAI's Claude access, GPT-5 launched, both companies offered AI to the government for $1, and they tested each other's models for safety.
Read More →Enterprise AI Moves From Pilots to Budgets
Generative AI use jumped to 75% of businesses, banks hired AI executives, and the Pentagon signed AI contracts. How I prioritize AI work.
Read More →AI Media Goes to Court: Midjourney, Anthropic, and Meta
Disney and Universal sued Midjourney, and two federal judges issued the first fair use rulings on AI training. What it means for training data and provenance.
Read More →Personal Context and Claude 4: May's Assistant Upgrades
Google showed Gemini using personal context from your apps, and Anthropic released Claude Opus 4 and Sonnet 4. The personalization tradeoffs product teams now face.
Read More →Reasoning Models Get Tools: o3, o4-mini, and What Changes
OpenAI's o3 and o4-mini can use tools while they reason. Why that matters, where it can go wrong, and when a reasoning model is worth the cost.
Read More →Agent Plumbing: OpenAI's Agents SDK and the Rise of MCP
OpenAI released the Responses API and Agents SDK, then adopted Anthropic's Model Context Protocol. How to design products that agents can use.
Read More →The Infrastructure Wars: Who Pays for AI Compute
Microsoft's $80 billion data center plan, specialized AI chips, Harvey's $300 million raise, and the first EU AI Act rules.
Read More →DeepSeek, Stargate, and Operator: January's AI Shake-Up
DeepSeek's R1 rattled the market, Stargate bet $500 billion on AI data centers, and OpenAI's Operator started using a browser. What it means for product teams.
Read More →