Turn Your Company Context Into Skills, Sub-agents, and Agent Teams

Enroll

Every new tool adds to a backlog you never actually build anything with. The fix is not another tool. It is turning the work you already repeat into assets: skills an agent can invoke, sub-agents that specialize, and teams that run multi-step projects. This is the construction course. You define skills, wire up sub-agents, and orchestrate them into teams, working in Claude Code and the OpenAI stack, which are the tools that transfer to the job. One durable pattern beats chasing every new framework.

The outcome

How does this prepare you for a job?

Agent and platform teams are hiring for exactly this construction work, and interviews test whether you can apply the pattern to a problem you have not seen rather than recite it.

Designed and shipped reusable agent skills with trigger descriptions, instructions, attached scripts and eval checks, which cut the time spent on repeat tasks across a team.
Built specialized sub-agents for research, coding and QA, each with scoped tools and its own system prompt, in Claude Code.
Orchestrated a lead and worker agent team across a multi-step workflow, using parallel fan-out and human review gates.
Implemented a skill-resolver pattern that routes an incoming task to the right skill or agent automatically.
Instrumented agent runs with structured logging and evals, to catch regressions in how a skill behaves.
Documented a shared skills library and onboarded teammates onto it.

Tools

AI tools learnt

The construction stack, which is the same set the job market is standardizing on.

Hands-on skills

Write a skill, covering its trigger description, its instructions, the scripts it attaches, and the edge cases that make it reliable
Decide between a skill, a script, a sub-agent and an agent team, using a rubric based on edge cases, number of steps and frequency
Scope a sub-agent by setting its role, toolset and permissions, and writing the system prompt that keeps it narrow
Orchestrate lead and worker teams, covering fan-out, merge, retries and stop conditions
Route tasks with a resolver, so the right skill fires without being hand-held
Version and eval skills, so that a tweak to a prompt cannot silently break a workflow
Debug a failing chain by working out whether the fault is the prompt, the context, the tool or the orchestration

Syllabus

Modules

Level: Basic

Your first skill

Find the work you repeat, and write its trigger and instructions
Attach a script, then walk the edge cases until it stops surprising you
Test the triggering, covering when it should fire and when it must not

Your first sub-agent

Role, toolset and boundaries, plus the system prompt that enforces them
Scoped permissions, covering what it may read, write and run
Why a narrow agent outperforms a general one

Chain them

A two-step pipeline where a skill's output feeds a sub-agent
Add the human gate at the point where it matters
Run one real task end to end, then review the trace

Level: Advanced

Agent teams that scale

Lead and worker orchestration, parallel fan-out, and handling merges and conflicts
Cost and latency budgets per run, which is the constraint platform teams enforce
Stop conditions, and recovery for when a worker goes sideways

The skills library as a product

A resolver, versioning, evals and documentation, run the way an internal platform team would run it
Adoption mechanics, meaning what makes teammates actually use the library
Deprecating a skill without breaking the things that depend on it

FAANG-scale jobs for agent teams

Repo-wide migrations that must not silently break the build
Code-review triage at a volume of thousands of diffs
Incident-summarization pipelines that correlate signals for whoever is on call

Ship one

Package a team for one real workflow, end to end
Measure the leverage by comparing cycle time before and after
Write the design doc you would defend in a systems interview

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.