04 Notes
Field notes from the practice.
Working notes on intelligence methodology, weak signals I'm tracking, and applied-AI workflows from real engagements. Not opinion pieces, process notes.
- AI for Intelligence · · 5 min readGoogle gave its strongest Gemini model to cyber defenders firstGoogle's Gemini 4 Carbon goes first to vetted cyber defenders, not paying customers. Also: a Cloudflare AI data leak and Anthropic's 40% government discount.
- AI for Intelligence · · 7 min readAxiom AI agents machine-checked the longest proof in mathematicsAxiom's AI agents checked a 15,000-page proof and found 34 problems in it. Also: Anthropic buys a biotech, $63bn for its chips and Italy's AI-at-work rule.
- AI for Intelligence · · 6 min readOpenAI says its model disproved the Exponential Time HypothesisOpenAI says its AI disproved a core assumption of computer science; outside experts will check. Also: a Qwen model's upload attempt and a planted agent file.
- AI for Intelligence · · 6 min readAnthropic files for Nasdaq IPO under the ticker ANTAnthropic filed to list on Nasdaq and reported $16.3bn in half-year revenue. Also: an agent incident in Georgia, Sweden's chatbot block and Meta's AI terms.
- Technology Intelligence · · 4 min readTechnology intelligence starts with a decision, not a databaseMost technology intelligence begins with a tool and a data dump, and dies in a landscape deck nobody reopens. Start from the decision, work back to the questions, then touch the data.
- Competitive Intelligence · · 3 min readWeak signals: what they are and where I look in 2026Three-layer framework for surfacing weak signals, and the source categories most analysts still miss.
- Technology Intelligence · · 3 min readTechnology scouting vs. technology intelligence, the working distinctionScouting is acquisition-oriented; intelligence is decision-oriented. Same data, different outputs. Mixing them up wastes both.
- IP Intelligence · · 3 min readWhy patent landscape work is broken, and what to fix firstMost patent landscape reports are visualization theater. They show what a database can plot but skip the harder question: what should the IP team actually do differently?
- AI for Intelligence · · 3 min readWhat Namířeno taught me about translating expert language with AIBuilding a tool that translates Czech parliamentary legislation into plain language taught me three things about where AI translation actually works, and where it fails predictably.