Field notes
Artificial Intelligence
Field notes from shipping agents, automations, and custom software. One useful idea, written for operators.
Showing 1–10 of 17 field notes
-
Gemini 4 Argon: frontier coding and cyber defense, phased for a reason
Google’s Gemini 4 Argon leads on long-horizon coding and cyber defense, but Fairwind and gov pre-release still gate access. Operators keep the judgment gate.
-
OpenAI DevDay 2026: Dots, GPT-6.1 Sol, and Why Humans Still Own the Judgment Gate
OpenAI shipped always-on Dots, GPT-6.1 Sol, and a Decisions API. Viking Labs’ take: route by job, keep humans on judgment, automate inside bounds.
-
Your company needs an OpenClaw strategy. Start with one workflow.
OpenClaw turns models into scoped, always-on agents on hardware you control. Viking Labs builds SaaS and agent systems. Start with one bounded workflow.
-
GPT-6 Sol and Luna: Cheaper High-Volume Models Still Need Bounded Ops
OpenAI launched GPT-6 Sol and GPT-6 Luna on Sep 22, 2026, about 50% cheaper than GPT-5.6 peers. Why cheaper models help throughput but don’t replace bounded workflows, agents, and human judgment.
-
What Grok Bot Galaxy showed about AI teammates that actually work
SpaceXAI’s three-day Grok Bot Galaxy (Sept 15–17) put live demos and role deep-dives on stage in San Francisco. Here’s what mattered for teams shipping real workflows.
-
Claude Opus 5.5 is a cheaper frontier engine. It still needs an owner.
Anthropic’s Claude Opus 5.5 claims Fable-level work at about 40% less than Opus 5. Useful engine. Still needs an owner, harness, success number, and exception paths.
-
Grok 4.7 is a better long-horizon agent. It still needs an owner.
xAI released Grok 4.7 today: a bigger step on coding agents, long professional jobs, and self-checking work. The useful question for a business is the same one we asked for Sol and Astra. Where does this remove a real constraint, and where does it just raise the bill? A better long-horizon agent is useful. A…
-
What we’re stealing from Grok Bot Galaxy (and what we’re not)
Three people. Three days. A company built live on stream with Grok Bot. That was the headline of Grok Bot Galaxy. We watched it the way operators should: not for the spectacle of “AI builds a company alone,” but for the operating rules that survived contact with a real build. The useful lesson isn’t unbounded…
-
Give version one a success number
Most AI pilots stall because nobody defined what success looks like. Here is how to scope a first release around one job and one metric you will actually check.
-
How to Build a Business Dashboard That Doesn’t Get Abandoned
Most internal dashboards fail not because of bad code, but because they were scoped around data instead of decisions. Here’s how to build one that lasts.