Every Team Has a Barrel Problem. AI Just Exposed It.

TL;DR

• A barrel is someone who can independently drive an initiative from inception to success. Your capacity for parallel initiatives equals your barrel count, not your headcount.
• AI is ammunition: scalable, cheap, rapidly improving execution support. Adding AI tools doesn't add barrels. It gives existing barrels more firepower.
• Most stalled initiatives are barrel problems, not execution problems. AI just made that impossible to ignore.

---

Your AI investment isn't stalling because of the tools. It's stalling because of your barrel count. AI just made that impossible to ignore.

Most teams see a productivity boost and then plateau. Engineers write code faster. Analysts produce summaries in minutes. Some backlog items ship ahead. But the number of meaningful initiatives running in parallel doesn't change. Strategic work still bottlenecks. Important things still wait.

The barrels vs. ammunition framework

Keith Rabois introduced the barrels vs. ammunition distinction in a 2014 YC lecture. The definition is precise: a barrel is someone who can independently drive an initiative from inception to success. Give them a goal and you can fire and forget. They accumulate resources when needed, motivate people when necessary, surface problems proactively with enough lead time to intervene. Ammunition is everyone else: competent, valuable, essential for execution, but operating in support of a barrel's direction rather than generating direction independently.

Your capacity for parallel initiatives equals your barrel count, not total headcount.

At PayPal, 254 people at acquisition, considered one of the densest talent networks in tech history. Barrel count: 12 to 17. Rabois asked Sam Altman at a conference how many barrels OpenAI had when it was smaller. The answer was two. Two. The number of things a company can genuinely pursue in parallel at any given moment is the number of people who can own an outcome end-to-end and deliver it without constant direction.

Hiring more ammunition behind the same barrels doesn't increase your parallel capacity. It increases coordination overhead. More people, same velocity on what matters, plus the drag of managing people who are waiting for direction.

Why AI is ammunition, not barrels

AI is ammunition. Excellent ammunition. Infinitely scalable, rapidly improving, cheap per token. A barrel with AI access has far more execution capacity than a barrel without. Code gets written. Briefs get drafted. Research gets synthesised. Data gets structured. Tasks that used to require briefing and managing a junior person now happen in the same session.

The engineering director Rabois described at Ramp is the sharpest example: personally shipping as much code as he did as an individual contributor, while managing a team of 20, because he uses AI as a second team. One barrel, abundant AI ammunition.

What doesn't change: the number of meaningful initiatives that can run in parallel.

Three barrels with AI tools can pursue three initiatives, faster and with more output per initiative. They cannot suddenly pursue six. The ceiling on strategic breadth is still the barrel count. AI raises the floor (each initiative moves faster) but the ceiling stays put.

Why AI makes the barrel problem visible

Before AI tools became widely available, the bottleneck was less obvious. An initiative might stall because the barrel didn't have enough execution support. You could plausibly argue that more engineers would fix it. Sometimes that was even true.

AI removes execution capacity as a legitimate explanation for stalls. When a barrel with good AI tooling can produce in an afternoon what previously required a week of team coordination, the stalled initiative can no longer be attributed to insufficient ammunition. The stall is the barrel. Either there isn't one assigned to the initiative, or the person assigned isn't actually a barrel.

This is uncomfortable. It surfaces a truth most organisations have been able to obscure: many of the people running important initiatives are not barrels. They're senior ammunition. Smart and experienced, but they need meaningful direction to advance. They don't generate it independently.

What a barrel actually does

Rabois is specific about the threshold: "Get us over that hill. One way or another, this is going to happen." The barrel accepts that the outcome is theirs to own and treats constraints as problems to solve rather than reasons to escalate. They come back proactively with what they've tried, what they've diagnosed, and what help they need with enough lead time to act on it.

The ammunition vs. barrel distinction isn't about seniority or skills. It's about agency.

I've seen this play out building OpenChair and OpenTradie solo: one barrel (me), AI as ammunition. Every initiative passed through a single decision-making point. The AI could produce an enormous volume of output. It could not decide what output mattered. It could not hold direction across a six-week build and adapt when the initial approach wasn't working. That was the barrel's job.

And it's a scarce skill.

The hiring implication

Most organisations hire ammunition and hope it develops into barrels over time. The senior engineer who's been around long enough. The product manager with ten years of experience. Sometimes the transition happens. More often it doesn't, and the organisation ends up with expensive, credentialed ammunition.

Rabois draws a useful distinction: are you hiring for value creation or value preservation? Value preservation (protecting what exists, managing quality, reducing risk) tends to benefit from experience and domain knowledge. Value creation (building something from intent to outcome) requires a different quality, and years of experience doesn't reliably predict it.

In an AI-native team, value creation capacity is the scarce resource. AI handles the execution that made large ammunition pools necessary. What the team needs is people who can generate and hold direction on an initiative until it delivers. More barrels.

The constraint-driven team design that AI makes viable only works if the constrained team includes genuine barrels. One barrel plus AI tools outperforms five expensive ammunition roles with the same tools, because the barrel generates the direction that makes the tools useful. The team of one model is only viable when that one person is a barrel.

Before asking how to get more out of your AI investment, count your barrels. That number sets your ceiling.

---

Frequently Asked Questions

How do you identify a barrel before hiring them?

References are the most reliable mechanism, and Rabois recommends 20 of them for senior hires. The right question isn't "was this person a good employee?" It's "can they own an outcome from zero to done without being driven from above?" Practically: find people who've seen them operate in ambiguous situations and ask for specific examples of when they picked up something undefined and delivered it without significant direction. Work history patterns are useful too. Multiple promoted-from-within roles, a track record of projects that actually shipped, minimal attribution of stalls to external factors.

Can ammunition develop into barrels?

Sometimes. The transition requires a specific kind of exposure: taking ownership of something ambiguous, getting stuck, figuring out how to unstick without escalating, and delivering. Organisations that develop barrels deliberately give promising people assignments where the goal is clear and the path is not, then observe what happens. The ones who figure it out are the barrels. The ones who surface every obstacle as an escalation are ammunition. AI-era assignments are good forcing functions for this, since the execution ceiling is high and the judgment ceiling is what actually limits the outcome.

Does every team need barrels, or just fast-moving startups?

Every team that wants to run meaningful initiatives in parallel needs barrels. Large organisations often solve this through process rather than people, which works until the environment changes fast enough to outrun the process. In conditions where the capability landscape changes month to month, process-heavy execution is structurally slow. The organisations that will adapt quickest to AI-era changes are the ones with enough barrels to redirect and pursue new initiatives without waiting for organisational consensus to catch up.