From Co-pilot to Fleet Commander

Fleet Commander for Executives

For: Leaders who own a P&L, a function, or a team, and who already use an AI assistant for real work. No technical background required.

You leave with: A one-page fleet plan, a one-page agent policy, a measurement sheet, and a working daily brief that runs in your own tools.

1. The co-pilot ceiling. Why one assistant plateaus, and what a fleet does instead.
2. The personal operating system. Build the morning brief, meeting prep, and follow-up tracker that keep you organized.
3. Writing the brief. Delegation as a skill: goal, inputs, done-criteria, constraints.
4. Operating model and org design. Who owns the fleet, which roles appear, and where the human gates go.
5. Governance on one page. What agents may read, write, spend, and send, and what happens when one gets it wrong.
6. What to delegate, what to keep. A decision grid for where fleets win first.
7. Metrics: four numbers, not a vibe. Hours returned, cycle time, review catch rate, cost per completed task.
8. Running the fleet weekly. The standing review that keeps the fleet, and you, organized.

Capstone: Present your fleet plan to peers and Matthew, who challenge it the way a skeptical board would.

Format: Live cohort, in person or remote, with labs in your own tools. Duration, cohort size, price, and dates: details to be announced.

Fleet Commander for Developers

For: Engineers comfortable in a language and a terminal, who have called a model API and have a real workflow that an agent should run.

You leave with: A shipped harness around your own workflow: tool contracts, a permission model, an eval suite in CI, tracing, and a cost dashboard.

1. What a harness is, and why your demo is not one. The anatomy of a system that runs unattended.
2. Tool contracts. Tight schemas, side-effect flags, idempotency, errors the model can act on.
3. Permissions and the blast radius. Scope every tool; gate every irreversible action.
4. Evals for behaviour. Golden sets, rubric grading, adversarial cases, regression in CI.
5. Context and memory. What goes in the prompt, what persists, and what it costs to forget.
6. Orchestration: one agent to many. Planner and worker patterns, fan-out, hand-offs, failure handling.
7. Cost and observability. Trace every run; put cost per completed task on a dashboard.
8. Shipping and operating. Version prompts with tools, canary agent changes, roll back on a failing eval.

Capstone: Demo your harness running unattended on live-shaped input, present the eval results, and walk through one failure you found and fixed.

Format: Live cohort, hands-on, bring your own repo. Office hours between sessions. Duration, cohort size, price, and dates: details to be announced.

Fleet Commander for Security

For: Security engineering, platform security, and GRC practitioners who are comfortable with identity, secrets, and cloud infrastructure, and who need a posture before the agents arrive.

You leave with: A security posture for one real agent workflow: threat model, compute and GPU requirements, identity and secrets design, audit spec, data-flow map, and incident runbook.

1. The agent threat model. Prompt injection through tool output, over-permissioned tools, memory poisoning, exfiltration by “helpful” action.
2. Trusted execution and confidential compute. What enclaves and confidential VMs protect, what they leave exposed, and when to require them.
3. Secure GPUs. GPU isolation, confidential GPU modes, and attesting the full stack from CPU to accelerator.
4. Secrets and the agent’s keys. Short-lived, scoped, brokered credentials that never land in a transcript.
5. Agent identity and least privilege. An identity per agent, delegation chains, revocation.
6. Audit trails that reconstruct intent. Log the prompt, the calls, the results, and the decision, with integrity.
7. Data residency and the model boundary. Where prompts, outputs, and embeddings actually travel.
8. Detection, response, and governance for fleets. Kill switches, quarantine, and a runbook that includes “stop the fleet”.

Capstone: Defend your posture in a tabletop exercise where Matthew plays the adversary.

Format: Live cohort, remote or in person, with labs in an instructor-provided environment and homework against your own. Duration, cohort size, price, and dates: details to be announced.

This is for you if:

• You are paying for AI seats and not seeing the lift.
• You have shipped a chatbot and now need agents that call tools without breaking things.
• You have been told “the agents are coming” and need a security posture before they do.
• You want to leave with artifacts you built, not a deck you watched.

This is not for you if:

• You want a prompt cheat sheet.
• You want a vendor bake-off before choosing a model provider.
• You want an AI strategy with no homework. Every track has labs.