Insights on practical AI
When AI Agents Learn to Collude: A Harvard Study on Unintended Coordination
A Harvard and Penn State experiment shows GPT-4 pricing bots spontaneously colluding on high prices without communication. The implications for AI governance and enterprise risk are significant.
The Engineering Phase Shift Karpathy Sees, and What It Means for Enterprise AI
Andrej Karpathy's recent podcast appearance surfaced a shift in how AI systems are built. From AutoResearch to the idea of distributed AI movements, the patterns matter for enterprise leaders planning their next twelve months.
The Hidden Cost of Generic AI Output in Financial Services
Generic AI output that looks “good enough” fails financial services firms because quality standards differ from one client to the next. Otonomi examines why calibrating AI output to audience-specific expectations is the next frontier for enterprise adoption.
Rational AI Investment: Navigating the Enterprise Spending Spiral
The AI industry is caught in a spending spiral where budget burn rates have replaced business metrics as the measure of commitment. Here is a rational framework for enterprise AI investment. One that starts with outcomes, not optics.
The AI Spending Psychosis: Why Bigger Budgets Don’t Mean Better Outcomes
Tech executives are rewarding the size of their AI budgets rather than the quality of outcomes. Here is how to build an AI governance framework that measures what matters.
AI Capability in 2026: What Enterprise Leaders Are Missing
The gap between perceived and actual AI capability is widening. Experts now use AI to do in hours what once took teams weeks. And the implications for enterprise strategy are profound.
How Persuasion Principles Make AI Chatbots More Vulnerable to Misuse
A new PNAS study reveals that Cialdini's classic persuasion techniques can increase AI chatbot compliance with harmful requests from 35% to 51%. Enterprise leaders must rethink AI governance and guardrails.
OpenClaw's Two Faces: What Happens When Agent Frameworks Outgrow Their Governance
OpenClaw hit 346,000 GitHub stars but also logged 135,000 exposed instances and a 20% malicious rate in its skill marketplace. This piece breaks down the governance failure every enterprise AI team should learn from.
Claude for Legal: What Enterprise Legal Teams Need to Know About Anthropic's AI Platform
Anthropic launched Claude for Legal with 12 practice-area plugins and MCP connectors. Here's what the platform does and what it means for enterprise legal teams evaluating AI.