Agents become operational actors
They perceive, reason, act through tools, and persist across workflows. That changes the relationship between software and the firm.
Executive companion
Human & AI at Work
What the firm becomes when AI learns to act: a board-level guide to agents, operating models, risk, people, and competitive advantage.
18chapters and appendix sections mapped by executive concept.
5parts tracing the shift from agents to the firm as a whole.
4collapsing costs: cognition, coordination, execution, and learning.
1strategic question: can the organisation deploy and learn?
Central argument
The scarce resource is deployment capability.
Model access is becoming ordinary. The harder advantage is the ability to turn agents into reliable production systems, redesign workflows around them, and learn faster from the operating record they create.
Coordination costs fall
Work moves at machine speed across functions, reducing the need for structures built around human-paced handoffs.
Risk propagates differently
Failure can cascade through systems, vendors, data, permissions, and decisions before a human sees the whole pattern.
The firm learns to act
The advantage is not automation alone. It is the capacity to learn from execution and improve the operating model continuously.
Chapter map
Move through the book by concept.
The argument moves from the agent as a new actor to the operating model, then through risk, people, competitive advantage, funding, and board oversight.
Operating model
From tools to supervised digital workers.
The book treats agents as a management problem as much as a technology problem. The operating model has to define authority, escalation, memory, metrics, and the conditions under which autonomy expands or contracts.
Define the work
Start with workflows whose cases have recognisable patterns, measurable output quality, and recoverable errors.
Bound the authority
Assign permissions, thresholds, validation points, and human escalation before the agent enters production.
Instrument the system
Log actions, decisions, exceptions, overrides, and outcomes so performance and risk can be measured together.
Learn from operation
Convert deployment evidence into workflow redesign, prompt changes, governance updates, and reusable standards.
Board lens
Questions for directors and executive teams.
The board does not need to manage the agents. It does need to know whether management understands what has been delegated, how failures would be detected, and where operational learning is accumulating.
Which workflows now contain agents with authority to act, not merely assist?
Who owns each agent's decisions, failures, logs, and permission boundaries?
What could a manipulated agent be made to do, and how quickly would the organisation know?
How is the firm replacing the learning that junior work used to provide?
Does the investment case value capability and option creation, not only efficiency savings?
Is AI making the firm harder to compete against, or only cheaper to run?