Designing the operating model for a human-led, agent-enabled company.

The agency model is moving beyond recurring deliverables. Its next advantage will come from designing how human judgement, AI agents, software and evidence work together.

The question is not whether AI can do the work. It is where human judgement creates value, where AI should assist, and which repeatable actions can be delegated safely.

This is not a plan to remove people. It is a way to make accountability explicit. A useful AI operating model gives each layer a clear role: a human manager sets the objective, agents execute within defined constraints, software and APIs connect the records, and people handle exceptions, relationships and material decisions.

How I map a process

For example, take a new inbound lead. The input is a form submission and the CRM record; the output is a reviewed qualification note and a proposed next action. An agent may deduplicate and prepare context, but a person approves any material promise, pricing or sensitive response. The first test checks accuracy, handoff time, exceptions and the audit trail before expanding permissions. This is a method illustration, not a published client result.

• Human onlyJudgement, trust, negotiation, accountability and high-impact decisions remain with people.
• AI assistedAI prepares research, drafts, analysis and recommendations while a person reviews the result.
• Agent executableRepeatable, permissioned actions may run within defined rules, logs, escalation paths and approval controls.

Current research agenda

• Multi-agent coordinationHow specialised agents divide work, transfer context and recover from failure.
• Operational memoryHow demonstrated processes can become reusable skills without turning errors into permanent behaviour.
• System connectionHow CRM data, advertising, content, reporting and customer operations connect through software and APIs.
• Governance and evidenceHow permissions, approvals, logs and measurable outcomes keep automation accountable.