The workforce has moved beyond text-only work
Why screen recordings, voice notes and smart glasses are becoming normal business inputs, and why AI models need to handle them directly.
Notes on useful agents
Plain-English notes on where agents fit into business processes, how they use existing systems and where people remain in control.
A prompt is not a security boundary
What the OpenAI and Anthropic security incidents teach us about permissions, monitoring and putting agents near real systems.
A capable AI model is not the same as useful work
Why impressive AI demonstrations often fail to become useful everyday work, and what has to exist between the model and the result.
A second agent needs a specific job
Why extra agents help with bounded research and review, but often add cost and noise to the main work.
A chatbot talks. An agent uses tools.
Why access to tools, permissions and approval points matters.
Choosing a sensible first job for an agent
How to identify a bounded process before planning a larger AI programme.
Your ERP does not need an AI makeover
Where an agent can support an existing system, and where it needs to stop.