One boat is easy. One captain walks the deck, checks every knot, hears every question, and knows the state of the ship by walking it.
Ten yachts, each with a crew of ten hands, is not ten boats of easy. It is a fleet, and no admiral alive can walk a hundred decks a day. The old habit of collaboration, the captain personally reading every change before it lands, is exactly that walk. The pull request assumed it. It held when the harbor held three boats a week. A hundred-thread team is a harbor full of masts.
The question at that scale is no longer "did a human check it?" The question is "can this ship prove its own state?" This post is about what changes in collaboration, what becomes critical, and what actually scales when the work multiplies.
What collaboration turns into
On one boat, people collaborate by watching each other's hands. In a fleet, nobody can watch anyone's hands. Collaboration moves to the papers:
- You stop reviewing effort and start reviewing evidence. A captain does not tell you what the crew did all week. A captain hands you the float plan they sailed under and the log that came out of it.
- Signals between ships get shorter and sharper. The old question, "what did you do?", becomes "what did your ships deliver, and what did you turn back?" The turn-backs are the interesting answers. A captain who never turns a ship back is not running a fleet.
- The papers become the artifact. The code, the cargo, the sailing itself: all of it matters less to your fellow captains than the three documents around it, the plan, the contract, and the log. That is what they can actually read of your corner of the sea.
Which leads to the first problem worth solving.
When one chart becomes a thousand
One boat keeps one chart, and even that chart goes wrong. Two corrections penciled over each other. Nobody sure which mark was made last season.
A fleet produces charts, logs, float plans, cargo manifests, and inspection records the way the sea produces waves. A thousand documents by March. "Who owns what" is now a daily question, and the fleet that cannot answer it hits the same rock twice.
The answer is the one that libraries, harbors, and every old institution learned: ownership by name, not by mood.
- Every crew of agents has one named captain. Not a committee. A person.
- Every chart, every contract, every standing order has one named keeper, the way a ship's papers have a keeper.
- Every document sits in a locker with a keeper, and the keeper's proudest act in March is throwing out January. A fleet that cannot throw anything away drowns in its own logs.
If you cannot answer "whose chart is this?" in one sentence, it is nobody's chart, and it will fail the way unowned edges always fail.
Graduated trust: nobody gives the wheel to a new hand
Here is where most fleets get it backwards. They want certainty first and autonomy second. At sea it cannot work that way, because you are sailing waters nobody has charted. There is no certificate that says "this crew is safe." There is only evidence, and evidence is earned nautical mile by nautical mile.
So trust expands the way every old sea tradition expanded it:
| Level | The crew may | The captain does |
|---|---|---|
| 0. On the rail | Watch, advise, touch nothing | Read everything they propose |
| 1. Ropes under eye | Work for real, output goes nowhere | Stand beside them, compare every result to the captain's own |
| 2. Harbor waters | Handle the wheel inside the harbor | Review everything, deeply sampled |
| 3. Open sea | Cross open water with instruments | Read the instruments, audit the exceptions, spot-check weekly |
| 4. Night crossings | Routine passages end to end | Watch the drift alarms, audit quarterly |
The ladder is climbed per crew, per water. The same crew can hold the wheel at level 3 in the calm bay and stay at level 1 in the strait, because the strait carries other people's cargo. Promotion is earned by evidence across voyages. Demotion is instant and shameless: one strange reading at three in the morning sends the crew down a level, and nobody calls that a failure. That is the system working.
And here is the answer to "how do we grow certain while building an unknown." You cannot be sure of uncharted water. Nobody can. What you can do is fix your position often enough that being wrong costs a correction, not a ship. Certainty is not a feeling you wait for. It is a level you can name, and a rung you can step to.
Security: keys, anchors, logs, and pirates
A pirate does not announce the ship they will board. Security at sea is four old ideas:
- Keys, not master keys. The cook does not hold the engine-room key. Every crew gets keys to exactly what their voyage needs, and every key expires at the next port. The question for every key is the harbor master's question: why does this hand need this door?
- Anchors and waypoints. Before entering unknown water, you note the waypoint you will turn back to. Rolling back a bad voyage is not an emergency. It is the thing that makes bold sailing affordable at all.
- The ship's log. Every action lands in a log the admiral can read: what ran, what it touched, and above all what it signaled to other ships. What leaves the fleet matters most. Cargo disappearing inside the fleet is a loss. Cargo appearing somewhere it should not be is an incident with your name on it.
- One gangway. The crew does not each get a private door to the harbor, the market, and the post office. There is one gangway, it is owned, it is logged, and everything that crosses it is seen.
Now the pirates, because they are the reason the list exists. Pirates do not board through the guarded front of the ship. They board through the unguarded hatch, and their oldest trick is the friendly flag: they arrive disguised as cargo. A pirate at your fleet is any instruction that reaches a crew while pretending to be cargo, a note inside a message, an order hidden in a shipment, a command floating in a bottle that a helpful hand fishes out and obeys.
The fleet's iron rule against all of it: orders come from the captain, never from the cargo. A crew that cannot tell an order from a shipment will one day hand the whole hold to a stranger with a flag.
Soundings all night: the feedback loop
You cannot wait for the reef to tell you where the reef is. Old crews knew this and took soundings all night, dropping the lead line every few minutes, reading the depth before the depth read them.
On one boat the feedback loop was a person: someone finished work, someone looked at it, tomorrow they talked. In a fleet, work finishes forty times an hour, so the loop has to live inside the ships:
- Sounding runs are exams for the water. You cannot read every change by hand, so you keep a set of known depths and drop the line on every voyage, automatically. When you change the hull, the engine, or the map underneath, you rerun the same soundings and compare. This habit is the highest-leverage thing in the whole post.
- Drift alarms are the compass that notices the fleet has been slowly turning for hours. Silent drift, a course change of one degree a day, sinks more fleets than storms.
- Sampled inspections are the port examiner opening five crates deeply instead of four hundred shallowly. The sample is random, and every surprise goes on the chart: each catch by a human becomes a new sounding in the standing set.
- Incidents redraw the charts. Every postmortem ends with a changed float plan or a new depth mark, so the same reef never takes two bites out of the same fleet.
The loop that matters most is short: change, sounding, verdict, in minutes, not at next week's captains' meeting.
Pull requests were good. Papers are better.
To be fair to what is being replaced. The harbor master who opened every crate personally ran an excellent port, when three boats arrived a week. Small changes, one careful reader, a conversation attached to each crate. On human-speed seas, nothing beats it.
It breaks on volume, and it breaks politely. A thousand crates a week turns every inspection into a stamp, and a stamp is worse than no inspection, because everyone pretends a check happened.
What scales is moving the human to both ends of the voyage:
| The old port read | The new port reads |
|---|---|
| Every crate, once | The float plan, before departure |
| One examiner's opinion | The sounding record's verdict |
| A story told by the ship | The manifest: what is aboard, with proof |
| Approval as a feeling | Approval as a threshold met |
| Everything shallowly | Samples, deeply, forever |
Before departure, humans write the float plan: what the voyage is for, what the crew may decide alone, what it must stop for, and what evidence it owes the log. After arrival, humans inspect samples and read the exceptions. The middle, opening every crate that a machine packed at machine speed, is where the old port dies.
This is capture before claim, fleet edition: a ship that cannot show its papers did not dock. It is still at sea, whatever it says on the radio.
Where the boundaries actually are
A fair question: which of these rules are the sea, and which are just this year's boat? Sort every practice into three piles, because the sorting matters more than the answers.
The laws of the sea. Some rules hold for every hull. Verification before trust, one keeper per chart, keys scoped and expiring, papers attached to claims. Better sails do not repeal them. They are not features of your technology stack; they are consequences of carrying cargo that matters with crews that are not perfect.
The waters you sail. Some rules belong to the harbor. A tanker in a narrow strait takes a pilot, moves slowly, and climbs the trust ladder a rung a quarter. A fishing dinghy in a calm bay earns the wheel in a week. Same fleet, different waters, and the difference is not fear. It is that the strait carries other people's cargo, and the harbor's whistle has consequences.
The weather. And some rules belong to the season. The specific soundings, the trust thresholds, the capabilities of the boats themselves: all of it moves with the monthly forecast, which means ladders and charts are revisited on a schedule, and "we calibrated this in January" is not an answer in June.
Two ways to die at sea, and both are common: treating weather as law, so the fleet freezes at every cloud, and treating law as weather, so the fleet sails the strait like a bay. The discipline is knowing which pile a rule sits in, and re-sorting the piles on purpose.
The loop, on one page
- Every crew, chart, contract, and locker has one named keeper.
- Every crew sits on a named rung of the ladder, per water, with the promotion evidence written down.
- Keys are scoped and expiring. Unknown water is entered from a noted waypoint. Everything is logged, especially what crosses the gangway.
- Soundings run on every voyage, before anything else is believed. Every human catch becomes a new depth mark.
- Humans read float plans before departure and inspect samples after arrival, never every crate.
- Charts, ladders, and the piles are re-sorted on a schedule, because the forecast expires.
Collaboration did not get colder out here. It got more honest. You argue with a fellow captain about the float plan once, before the lines are cast, instead of about the crates forever after, and the crew does the hauling while you both hold the pen that matters: the one that writes what this voyage was for.
FAQ
Are pull requests dead?
No. Small boats still deserve them: the work humans do by hand, and the changes crews make in harbor waters at level 2, still get the personal walk-through. What dies is the assumption that every crate gets opened by hand at scale. Papers before departure and sampled inspection after replace it, with the walk-through reserved for the exceptions.
Are the soundings not just "more tests"?
Tests check the hull against the blueprint. Soundings check the voyage against known water, including the fuzzy kinds: how the crew handled an odd tide, whether the log reads like a log or like a story written afterward. The habit that matters is running the same soundings on every voyage and every refit, so a changed score means changed water or a changed ship, not changed questions.
Where does the security budget go first?
The gangway. One owned, logged door between the fleet and the rest of the world, with expiring keys behind it. Everything else on this list is cheap. A hundred crews each holding a private unmonitored path to the harbor, the market, and the post office is the expensive mistake, and pirates only need one.
How do we start on a fleet that already sails?
Pick the three routes your ships sail most often. Write float plans for them, build one set of soundings for each, and run those crews at level 1, ropes under eye, for two full voyages. The comparison data from those voyages is your evidence, and the ladder after that is the same move, repeated with deeper water.
Related reading
- Ten People, One Hundred Agents
- Confident Wrong Output Marked Done Is the Agentic Failure Mode
- The Review Bottleneck Is the Real Cost Center in AI Teams
- Design the Escalation Layer Before You Add More Agents
- Agents Should Not Share a Brain
- Judgment Does Not Transfer
- More Agents Do Not Create Continuous Improvement
If you are turning a crew into a fleet and want the papers, the keys, and the ladders drawn before the pirates find you, book a discovery call.