What this rung really is
Level 4 is where you stop being the bottleneck. Until now one agent did the job one step at a time, and you waited. Here you hand a big job to several agents at once — each takes a slice, works it in its own space, and reports back — so an afternoon of grinding finishes in the time of one. Then you put a gate in front of the result: a single check the work has to pass before you accept it.
The same move shares your foundation. The rules, skills, and brand you built in Levels 1–3 stop being yours alone — your team runs on the exact same setup instead of everyone reinventing it. Clear this rung and you’ve gone from “I run an agent” to “I direct several, and nothing ships unchecked.”
Direct several agents — with a gate on the way out
The move at this rung is orchestration: split one big job across parallel agents, each in its own space, then put a single check in front of the merged result so nothing ships unreviewed.
This rung sits on everything you wired in rungs 1–3. You add orchestration on top — you don't start over.
Do this rung, for real
Everything you need is here — no tabs to chase. First, the jargon this rung throws at you, in plain words. Then the steps, with the exact things to paste or say.
A new client hands you a 60-page site to audit before kickoff. One agent reading it page by page is a full day. Instead you split it: five agents each take a section — technical, content, links, schema, speed — and audit their slice at the same time, in their own workspace, while your main conversation stays clean. Twenty minutes later you have five drafts. Before any of it reaches the client, one gate runs: a second agent re-checks the findings against the live site. A day of solo grinding becomes a directed sprint with a check on the way out. That's L4.
Splitting the work (parallel agents)
Instead of one agent doing everything in a single conversation, you spin off helper agents — each gets a slice of the job, works it in its own space, and reports back. You direct; they run at once.
Why it matters here — You stop waiting on one agent to grind a long task step by step. A research sweep, a big cleanup, a multi-section audit finishes far faster when five agents take a fifth each — and your main thread stays clean because the messy work happens in theirs.
The gate
One check the combined result has to pass before you accept it — a second model reviewing the output, or real browser evidence that it actually works, not just looks done.
Why it matters here — Parallel work is fast, and fast unchecked is dangerous. The gate is the one habit that keeps speed honest: nothing reaches the client until something other than the agent that made it has confirmed it.
Briefing each agent (context engineering)
Each helper agent only knows what you tell it — so before it runs, you hand it a tight, self-contained brief: the slice it owns, what “done” looks like, and what to leave alone. You keep the big picture in your main thread and give each agent just its piece.
Why it matters here — An agent with a vague brief burns half its effort rediscovering the job, and five vague briefs give you five shallow drafts. A tight brief per agent is what turns “five agents at once” from noise into leverage — and it's why your main conversation stays clean instead of clogged with everyone's raw work.
Split one real job across agents
Take a job that's too big for one pass — a multi-section audit, a batch of pages, a repo-wide cleanup — and hand each slice to its own agent. Each works in its own space and reports back, so your main conversation stays clean and the pieces run at the same time. The split only pays off if each agent gets a clear brief — the exact slice it owns and what a good result looks like — so it works its piece instead of guessing at the whole.
“Split this site audit into five — technical, content, internal links, schema, and speed. Give each one its own section, run them in parallel, and bring back a draft for each.”
Put a gate in front of the result
Before you accept the combined work, run one check it has to pass. The simplest gate you can add today: route the output to a different model for a skeptical second opinion. For visual or interactive work, the gate is real browser evidence — a screenshot or a click that proves it works.
“Now send the merged findings to a second model and have it challenge anything that isn't backed by the live site. Flag what doesn't hold up before I see it.”
Share the foundation with your team
The rules, skills, and brand you built in Levels 1–3 don't have to stay yours. Share that setup so your team runs on the same foundation you do — one agent's worth of standards, multiplied across everyone, instead of each person reinventing it.
“Package my rules, skills, and brand kit into one repo and give me the one command a teammate runs to get the exact setup I have.”
How you know it's working
Before L4: one agent, one job, one step at a time, and you're the one waiting. Big tasks take as long as they take, and the only check on the output is you, reading it at the end.
- A real job was split across several agents working at once — you can point to each agent's piece, not one long transcript.
- A gate ran before you accepted the result — a second model's review or real browser evidence, not just your own read.
- Your main conversation stayed clean while the heavy work happened in the helpers' spaces.
- Your foundation is a repo a teammate clones with one command — an update is a pull, not a re-send — and it lands the same on any machine you own.
Make it stick. Pick the one recurring job that's too big for a single pass — the monthly audit, the bulk page build, the cleanup you keep putting off. Run the next one split across agents with a gate on the output, and orchestration becomes how you work, not a one-off experiment.
Our skills for this rung
Linked items are founding-circle skills — clone the repo and run ./install.sh from the skills folder. Unlinked items are practices you build by doing.
Orchestration running — parallel agents with ≥1 gate. Share the config.
Clear this and you've genuinely cleared the rung — not read about it. Keep the proof; it's how you place yourself on the ladder.