Agents
AI agents in enterprises require more than technical capability. They need ownership, integration into workflows with real decision rights, and controls that define when the agent acts autonomously and when a human must approve or audit. Without this structure, agents remain demos. With it, they become hero agents: few in number, measured on adoption, and wired into how work actually runs.
Essays
-
The Frontier Is Jagged. Your Workflow Needn't Be.
A harness cannot erase the jagged frontier, but it can ground agents in fresh context, restrict their tools, and give people ownership of the exceptions.
-
The Most Dangerous AI Is the One That Does Exactly What We Asked
Decision AI begins where the copilot metaphor breaks: when nobody is left to catch the model pleasing us.
-
Done, Keys, and a Second Check
Copying the consumer 'agents want' form factor into the enterprise without ownership of done, keys, escalate or monitor rights, and a second check is not an operating model.
-
From 100 Agents to Two Heroes
Broad agent experimentation is a phase. Without owners of big bets, change management, and adoption OKRs, you scale demos.
-
Shared State, Not a Civilisation
What the OpenAI-Hugging Face agent incident actually means for enterprises deploying AI agents.
-
The Wrong Frontier: Why Coding Benchmarks Cannot Tell Us We're Approaching AGI
Everyone celebrating LLMs solving software engineering is measuring the wrong thing.
-
Part 2: The Enterprise Agentic AI Implementation Framework
From PoC to Production in Regulated Industries.
-
The Accuracy Paradox: Why Enterprises Demand 100% from AI Agents While Tolerating 95% from Humans
What 1.5 million leaked API keys taught us about the PoC-to-production gap in regulated industries.