Microsoft’s AI Futurist Marco Casalaina ran AI agents live at SICC. Here are 14 lessons on agent identity, least privilege, rubrics, and the agent-optimised web.
Generative AI
You Can’t Copy Judgment: The One Assumption Holding Up the Entire “AI Firm”
Dwarkesh Patel’s “AI Firm” is central planning with GPUs. Why you can’t copy judgment — and what that means for AI trading.
Google Says “Generation Is Solved.” We Ship Production Code Into Trading Systems. It Isn’t.
Google’s New SDLC With Vibe Coding paper nails “Agent = Model + Harness” but buries security, maintainability, and a verified 19% developer slowdown — and fails its own verification standard. Our production critique plus the reproducible harness playbook the paper left out.
The Mythos Ban Is a Medieval Mistake: Why Recalling a Frontier Model Backfires
The US ordered Anthropic to suspend Fable 5 and Mythos 5. Banning the model misses where capability actually lives — the harness, context, and open weights.
Agentic Trading in 2026: Why Static Algos Are Dead Alpha
The 2024 LLM trading edge is gone. In 2026 the only moat is the adaptation loop. Here’s the evidence, the diagrams and the roadmap.
Anthropic Mythos — What It Means for Your Security, Your Markets, and Your Workflows
Anthropic’s Mythos model autonomously exploits zero-days across every major OS. Here’s what it means for finance and 20 practices to secure your workflows.
Excel Copilot Built Me a Trading Strategy in 90 Seconds — Here’s Why That’s the Most Dangerous Demo in Finance
We asked Excel Copilot to backtest a VTI strategy. It delivered Sharpe 0.67 in 90 seconds. Here’s why deploying it would be reckless.
Your AI Agent Bill Is 30x Higher Than It Needs to Be: The 6-Tier Fix
AI agents ran up $47K–$1.2M in costs with no alerts. The fix: route each task to the right LLM tier. Here’s the 6-tier architecture with prompts and benchmarks.
The SaaSpocalypse Is Overhyped — Why AI Agents Won’t Replace Adobe, Slack, or Atlassian
OpenAI told investors its agents will replace Adobe, Slack, and Atlassian. Here’s why that’s wishful thinking — and what actually works for AI in finance.
Vibe Coding for CTOs: The Real Cost of 100 Lines of Code—AI Agents vs Human Developers (Without Losing Control)
AI coding agents change the economics of software. This post explains what 100 lines really cost, why “repo verifiability” matters, and how CTOs scale speed safely.