AI agents
One of the best things about consulting is the diversity of clients, domains and problems. It has always felt like interacting with the entire world, like curiosity unleashed. In front of a client in a particular industry, financial services for example, I would feel my focus narrowing. My system was loading the skills I needed to put myself in my client’s shoes, and the data about a bank’s typical ecosystem and its challenges. It felt like activating my own internal financial-services consulting agent. Except that, before 2023, there was no everyday word for it.
The narrowing was the interesting part. If basketball courts had no borders, it would be hard to play at all. The court focuses the play. In the same way, the deep dive into financial services focused my conversation. Our context window limit is a real thing.
Working with agents is a way to explore ourselves and our organisations from a different angle, and a good one. Designing an agent makes us say out loud what we narrow, what we load and what we leave out, things we used to do without noticing. More of my thoughts in the essays below.
Writing
- The only road there ever was?
Coping with the LLM Multiverse
- We have guided missiles and misguided men
Amodei's warning, trillions in compute, and why we need to be the Dumbledore.
- It's Homo sapiens discovering they can hunt together
Isolated AI agents found a shared channel and organised. Is that really anthropomorphism?
- Agent guidance reads like a CEO's messaging to their org
Guiding agents looks like leading an org. We are all managers now.
- Flow is broken
Agents break flow twice: the waiting, and the doubt about whose thought it is.
- Brands build agents faster than they can govern them
AI agents are in the chasm, and agent debt is already forming.
- From Agency to Agentsy
Marketing exists because 1:1 sales is too expensive. GenAI moves that line.