Barrels and Ammunition
Output rises with the people who can carry work the whole way, not with headcount.
At a glance
The problem it solves
Most teams grow on the assumption that more people means more shipped, and for a while it does, because the first few hires absorb work that was already queued. Past a certain size it tends to stop, and the usual explanations are communication overhead, onboarding, or a process that hasn't kept up with the headcount.
Those are real and they don't account for the shape of it. What generally happens in a growing organisation is that work gets divided into pieces small enough to hand out, and each piece then needs somebody to define it, somebody to agree it and somebody to accept it back, so the number of boundaries grows faster than the number of people does. None of it is anybody's fault and it doesn't show up in any individual's output, because everyone is busy and most of them are doing good work.
The thing that actually moves is different. It's how many people can carry a piece of work from a vague idea through to something customers are using, without the relay. Organisations tend to have far fewer of those people than they think, and they usually can't name them, partly because the trait doesn't appear in a performance review and partly because it looks like insubordination when it does appear.
The idea
Keith Rabois drew the distinction in a 2014 lecture on operating a company, and the two words are ammunition and barrels. Ammunition is work product, so a design, an analysis, a document, a quarter of engineering effort, all of it real and none of it self-directing. A barrel is somebody who can take an idea, decide what it ought to be, get the organisation to agree, and ship it.
One piece of work drawn as a left-to-right run of five stages, from an idea through framing, deciding, building and agreement, to shipped. Below it are two lanes covering the same distance. The first, labelled ammunition, is three separate bars with dashed breaks between them marked hand-off. The second, labelled a barrel, is one continuous bar with no breaks in it.
The claim underneath the vocabulary is closer to arithmetic than to character. If shipping something requires the whole run to be covered, and most people can cover one stage of it well, then output is bounded by the people who can cover the run rather than by the total headcount. Ammunition added to a team that has a barrel raises what that barrel can ship. Ammunition added to a team that doesn't have one accumulates as inventory, which is why a group can produce more documents every quarter and less software.
It's worth being careful about what the words describe. A barrel is a thing that happened to a piece of work, so somebody carried it, and the shorthand attaches that event to a person, which is convenient in a lecture and misleading in a performance review. The useful version of the idea is a question about the run and where it stalls, and the version that gets repeated is a question about who is good.
How to use it
Start from the shipped work rather than from the people, because a list of names produced from impressions tends to reproduce whoever is most visible. Take the last five things that reached customers and, for each one, write down who made the call when the requirements were ambiguous, who went and got the decision that was blocking it, and who was still on it in the final fortnight.
Then look at what those people are doing now. The common finding is that they've been promoted into coordination, so they're running a team, chairing the review and carrying nothing themselves, which turns the scarce capability into more of the abundant one.
Count the hand-offs on one piece of work while you're there, because the answer is usually higher than anyone expects and it's the cheapest thing to change. A run that crosses six boundaries needs six people to be available and agreeable in sequence, and removing two approval gates often does more for throughput than another hire would.
Where somebody looks close, the missing ingredient is more often authority than ability, so hand over a real decision and watch what happens to the timeline. That experiment tells you something about the organisation as well as about the person, which is the more useful half of the result.
Where it breaks down
It's a claim about people and it gets used as a caste system. Ammunition is a demeaning word for a colleague, and once the two labels exist in an organisation they get applied, quietly, in calibration meetings, by people working from impressions. Most of the shipped work anywhere is done by people who cover one stage extremely well, and a vocabulary that sorts them into the lesser category is going to cost more in retention than it returns in insight.
Star performance travels badly, which undercuts the whole premise. Groysberg tracked more than a thousand star equity analysts across moves between banks and found their performance dropped on arrival and stayed down. If the capability were a property of the person it would move with them. What it looks like instead is a property of the person plus the relationships, the resources and the permission that particular organisation gave them, which means a barrel is partly something you make and not only something you find.
The evidence is one operator's experience of a few unusually successful companies. Rabois is describing what worked at PayPal, LinkedIn and Square, and the companies where the same approach was tried and the outcome was bad aren't in the lecture, because nobody gives that lecture. Advice drawn from survivors overstates how much of the outcome the advice explains.
It explains everything after the fact and predicts nothing before it. If something shipped, whoever pushed it through was the barrel. If it didn't, there wasn't one. Stated that way the framing can't be wrong, which is a warning sign rather than a strength, and it's the same objection that applies to most frameworks built from an operator's pattern-matching.
Who gets seen this way is subject to the usual distortions. Carrying work end to end means deciding without asking, cutting scope unilaterally and occasionally being wrong in public, and how that behaviour is read depends heavily on who's doing it. The people most likely to be given the benefit of the doubt are the people who already have standing, so the framing will tend to confirm an existing hierarchy while sounding like a fresh look at it.
It says nothing about making the run carryable. The interesting question is usually why the run has six approval gates on it, and the framing points at people instead, which is the more flattering answer for whoever designed the process.
In one line
Throughput follows the number of people who can carry a piece of work the whole way, so find out who they are from what actually shipped, and then avoid the usual reflex of putting them in charge of meetings.
References
- Primary
How to Operate (opens in a new tab)
Where the vocabulary comes from, and worth watching rather than reading a summary of, because the idea takes about four minutes of a lecture on operating a company and reads differently in that setting. It's an operator describing what he does, offered with no claim to evidence.
- Further
The Mythical Man-Month: Essays on Software Engineering
The older and better-argued version of the same observation, from 1975, which is that adding people to late work makes it later because the communication paths between them grow much faster than the people do. Brooks is about coordination cost; this concept is about who can absorb it.
- Critique
Chasing Stars: The Myth of Talent and the Portability of Performance
More than a thousand star equity analysts tracked across moves between banks, with performance dropping on arrival and staying down for years. The strongest available argument that what looks like a property of a person is substantially a property of the place they were doing it.