How to operate a Daycare of Agents

Enroll

If you can delegate, trust but verify, and review a report's work, you already hold the core skills for running a team of AI agents. Leverage stopped coming from writing more code or hiring more people. The old model that equates capability with lines of code is obsolete, and upgrading your tools while keeping the old instinct only gets you to the wrong place faster. You do not have to be the most technical person in the room to run agents, because the instincts that make you a good manager transfer directly. This course gives you a repeatable structure for running several agents at once without babysitting them.

The outcome

How does this prepare you for a job?

Leads are increasingly asked to show they can get output through AI and not only through people. Skill demonstrates itself, and a rehearsed interview script does not.

Operated a lead and worker agent team covering research, drafting, code review and QA, which delivered a multi-step workflow end to end with human-in-the-loop approval gates.
Defined the delegation and escalation policy for agent workflows, covering task ownership, verification thresholds and stop conditions.
Implemented guardrails, meaning scoped permissions, tool allowlists and audit logging, which reduced agent error rates before delivery.
Built context-engineering practices covering memory as against context, compaction and retrieval, which kept long-running agents on task.
Onboarded entry-level engineers to supervise and review agent output.
Instrumented agent runs to detect drift in the output and to triage failures.

Tools

AI tools learnt

These tools transfer to the job. Automation builders such as n8n and Zapier do not.

Hands-on skills

Decompose a project into tasks an agent can own, each with explicit inputs, outputs and conditions for being done
Write the trust-versus-verify rule for each task, covering what an agent ships unreviewed, what a human signs off, and what halts the run
Stand up a lead and worker agent team, and route work between the sub-agents
Engineer context deliberately, deciding what lives in memory, what gets retrieved, and what gets compacted away
Set tool allowlists, scoped permissions and audit logging, so that an unsupervised agent cannot take an irreversible action
Detect drift in output using spot-checks and evals, and diagnose whether the fault lies in the prompt, the context or the orchestration
Coach a junior teammate to review agent output, covering what to spot-check, what to reject and what to escalate

Syllabus

Modules

Level: Basic

The management-to-agents transfer

Which people-management instincts map onto agent operations: delegation, trust but verify, and escalation
Where the analogy breaks, because an agent does not learn from a 1:1, it learns from context
Write the job description for your first agent, covering its role, tools and boundaries

Your first delegated agent

Stand up a single agent with a scoped toolset in Claude Code
Define its trust-versus-verify rule and add a human-in-the-loop gate
Run one real multi-step task end to end, and review the output

Reviewing the work

Spot-check the way a manager does, sampling the output rather than reading every line
Catch slop before it ships, and write down the criteria for rejecting it
Keep a run log you can audit later

Level: Advanced

Orchestrating a team at scale

Lead and worker patterns, parallel fan-out, and strategies for merging the results
Context engineering across long-running agents, covering memory as against context, and compaction
Cost and latency budgets per run, which is the constraint every big-company platform team enforces

Guardrails that pass an enterprise review

Scoped permissions, tool allowlists and audit trails
The failure modes of prompt injection and data exfiltration
The approval gate, meaning what an agent may never do without a human

FAANG-scale problems agent teams get pointed at

Code review at Google-scale volume, triaging thousands of diffs without drowning the reviewers
On-call and incident triage, with agents that summarize, correlate signals and escalate
Fleet-wide migrations, where the agent must not silently break the build

Leading humans who lead agents

Onboard entry-level engineers to supervise agents
Set team norms and review SLAs
Make the case in a promotion packet, based on leverage delivered rather than headcount managed

Students can take the full course (both levels) or either level on its own.

Register

Waitlist

Join the course waitlist

No cohorts are scheduled yet. Join the waitlist and we’ll email you when registration opens.