Fleet Series 11 — The Two Loops You Still Drive

Working notes for a planned episode — an outline, not an article. Details will shift when it gets scripted. Series index: Building the Fleet.

The closing zoom-out: after ten episodes on the machine, this one is about the humans the machine runs inside. It promotes the "three-loop hierarchy" from a closing beat in episode 01 into its own episode — the human-facing counterpart to "the evaluator is the product."

The hook

The question every agentic-fleet demo eventually gets, usually from someone slightly worried: "so what do you do all day?" If the fleet plans, writes, tests, and merges — what's left for you? The answer is the whole episode: your coding loop is the fastest, smallest one, and it runs inside two bigger loops you still drive.

What the viewer walks away with

• The coding loop (minutes) is the innermost loop, not the whole game — it nests inside a feedback loop (hours) and an outside loop (days–weeks)
• Automating the inner loop doesn't remove your job; it moves it up a level — from writing code to writing intent and owning outcomes
• The better the fleet gets, the more the bottleneck moves outward — speed inside raises the stakes outside
• "The evaluator is the product," seen from the human's side: intent + accountability are what you do all day

The three loops (the whole episode is this one diagram, drawn inside-out)

• Coding loop — minutes — the fleet runs it. Generate → evaluate → keep/reject on a single leaf. The ~5 lines everyone demos. You don't sit in this loop anymore; you set its rules.
• Feedback loop — hours — you run it. You review what landed and update the things that steer the inner loop: the spec, the evals/gates, the lessons/rules, the task tree. Every correction here is a decision that travels inward — the next wave inherits it. This is where a human actually spends the day.
• Outside loop — days to weeks — reality runs it. Real users, real needs, changing priorities decide what's even worth building. No amount of inner-loop speed helps if the outside loop is pointed at the wrong thing.

The anchor line: "Your loop is the fastest, smallest one; it runs inside two others you still drive."

The human's remaining job (name it plainly)

• Intent — specify what you want clearly enough that a machine can build it and an evaluator can check it. Specs, evals, taste. An eval is intent, made executable (this is the gates episode, from the other side).
• Accountability — own what ships. The machine has no skin in the game; you do. The merge line the agent doesn't cross is where accountability lives.

The counterintuitive turn (the beat that makes it land)

• Speeding up the inner loop has diminishing returns — it's already fast, and it's the part the machine does best. The leverage is in the two loops you still drive.
• So the better your fleet, the more your day tilts toward intent and judgment, not less. Automation doesn't shrink the human job; it concentrates it into the parts that don't automate: deciding what's worth doing, and standing behind it.
• The failure mode to show: a fast, well-run inner loop building the wrong thing, beautifully. A fleet with no one tending the outer loops is a very efficient way to be wrong.

Beats

• Cold open on the question — "so what do you do all day?" — pose it, promise to answer it by the end
• Draw the coding loop (callback to 01/02), then zoom out once, then once more, adding a loop each time
• Land the anchor line, then name the two human jobs: intent + accountability
• The transmission belt: how a decision made in the feedback loop (a new eval, a lesson, a spec edit) actually reaches the inner loop — spec + evals are the wires
• The counterintuitive turn: optimizing the loop the machine already runs is the low-leverage move
• Close by paying off the hook: what you do all day is the two outer loops — and that's more, not less, than before

Demo ideas

• A real feedback-loop turn on camera: a wave lands, you spot a class of mistake, you write it into the lessons/eval, and the next wave doesn't repeat it — the human loop steering the machine loop
• Split screen: the inner loop churning leaves (minutes) beside a human-loop artifact changing (a spec/eval diff) — different clocks, same system

Source material

• Fleet Series 01 — where this began as the closing "three-loop hierarchy" beat (now promoted here; 01 keeps a short teaser pointing to this episode)
• Fleet Series 05 — Gates — "the human gate is the feedback loop wrapping the coding loop"; this episode is 05's thesis from the human's side
• Panaversity's Loop Engineering Crash Course — origin of the three-loop framing; credit if quoting the "fastest, smallest loop" phrasing
• Society of Mind — a Lens on the Fleet and the pipeline-as-a-model notes — adjacent closing/philosophical material that pairs with this episode

Open questions

• Placement: ships as the closing episode (11) after the deep dives, or as an early companion to 01? Leaning closing — it's the payoff, and it lands harder once the viewer has seen the machine in full
• Fold in the other closing/philosophical material (Society of Mind, pipeline-as-a-model) into one outro episode, or stay tight on the two-loops idea? Leaning tight
• Title: "The Two Loops You Still Drive" vs "So What Do You Do All Day?" — the question is grabbier for a thumbnail