For institutions
The control plane for AI agents
A financial institution is about to let an agent settle a claim, release a payment, or merge a change into production. Before it acts, somebody has to have decided who may delegate that act to a machine, under what mandate, with what check before the act, and what record after it. The talk goes through that control plane from the board's appetite down to one act, with a map a firm can hold its own vendors against.
- Suits
- Boards and risk committees, chief risk and operating officers of banks and insurers, supervisors.
The mandate as the unit of risk
Firms set risk appetite on the portfolio and write delegated authorities for their staff. An agent gets entitlements. The mandate is the join: a versioned authority over one class of acts, which says what the agent may do, what evidence it needs, when the firm may refuse, and who answers when it does. The talk takes appetite from the portfolio down to the act and shows what a mandate looks like on paper.
- Suits
- Risk and compliance functions, model risk management, internal audit.
The value of what the model knows
A token has a price. What the model is permitted to rely on has none, so it is treated as free, and the cost appears later as a decision nobody can defend. The talk puts governed knowledge into the economics of an AI task as a variable of its own, and shows what changes in a business case once it is there.
- Suits
- Chief financial officers, heads of AI, programme sponsors.
- Behind it
- The Value of What the Model Knows
Dynamics blindness
A model answers correctly at the point where it is asked and breaks a rule three systems away. The gap is architectural, and no amount of prompting closes it. The talk takes one permission change through a bank's systems to the breach, and then to what has to be underneath the agent for the breach to be seen before the act.
- Suits
- Architects, chief information officers, heads of AI engineering.
The Knowledge Layer
Institutions already have controls for their data and systems. The talk examines how those controls need to extend to the interpretations and assumptions an AI system relies on when it acts, and presents the layer that would hold them as a reference architecture: the claim, its standing, the check before the act, the record.
- Suits
- Architecture boards, engineering leadership, the technology teams of regulators.
- Behind it
- The Knowledge Layer · Admissibility
For a wider audience
The right size of intelligence
An ant colony has no ant in charge, and a brain has no neuron that knows what the brain knows. The intelligence is in the interactions between the units, and it has a size: units too large or too clever make the whole dumber. The talk takes that from colonies and brains to the multi-agent projects that failed last year, and asks how big an AI agent should be.
- Suits
- General and technical audiences, universities, festivals.
- Behind it
- The Right Size of Intelligence
Intelligence was never in the machine
Plato got there first, and a developmental biologist is now giving the oldest idea in philosophy physical force: intelligence is a property of organisation, and the machine only ever held a part of it. The talk follows that idea from cells to companies to the models being built now, and ends on why a cell, a child, and a company all stop trying when nothing comes back.
- Suits
- General audiences, science and humanities audiences.
What binds in the AI age
Cotton gins, ledgers, licensed dealers: the history of holding a person is a history of the instruments that made it pay. AI changes the value of holding, binding, or predicting a person. The talk asks how binding of that kind has come apart before, and what the stranding of people looks like now.
- Suits
- General audiences, policy and ethics audiences, universities.
- Behind it
- Slavery in the AI Age · The Quiet Stranding
Guided emergence
Some things can be designed, and some can only be grown. A city, a language, and a market were grown, with the rules written into the places where people meet. The talk is about the line between the two, and what it means for building systems in which many agents, human and machine, act at once.
- Suits
- Design and architecture audiences, founders, general audiences.
- Behind it
- Guided Emergence · The Emergence Layer
Contact for an invitation.