Astro Teller: How X Kills 1,000 Ideas to Find One Waymo

What if the secret to building world-changing companies was getting really good at killing ideas? Astro Teller, CEO of Moonshots at X, has done exactly that — and the results include self-driving cars, drone delivery, and the AI lab that shaped Google.

Key Facts

• Ideas tested per decade: 1,000–1,500, with only a handful graduating (Astro Teller)
• Ideas tracked per year: 100–200 receive a code name; far more are considered (Astro Teller)
• Key metric: Learning per dollar — not progress or revenue (Astro Teller)
• Graduates: Waymo, Wing, Google Brain — all originated at X (Astro Teller)
• Kill criteria timeline: Written at the start of full-time work, typically reviewed at 6–12 months (Astro Teller)

What Is X's Moonshot Factory and How Does It Work?

X is Alphabet's internal innovation lab designed to find and test radical ideas at scale. It operates like a '21st century Bell Labs,' testing 1,000–1,500 ideas per decade, killing most of them early, and graduating the survivors as independent companies with founding equity for the teams.

Most founders go all-in on one idea. X is built to do the opposite — deliberately and efficiently. As Astro Teller describes it, X was created to explore 'over the horizon' opportunities that, if they worked, could become 'enduring businesses that are really good for the world.' The challenge: most won't work, and the system has to be honest about that from day one.

Out of this process, some projects — like Google Brain — become so important to Google they're absorbed back in. Others, like Waymo (self-driving cars) and Wing (drone delivery), become standalone Alphabet subsidiaries. Increasingly, mature projects graduate as fully independent companies where Alphabet holds a minority stake and the team gets founders' shares. The whole machine runs on one uncomfortable truth: celebrate the killers, not the survivors.

"Our job is to try one thousand, maybe fifteen hundred ideas per decade. And if we get a couple Google Brains or Waymos or Wings per decade, especially if we're really efficient at throwing most of those ideas away early — then this is really good for our investor Alphabet and really good for the world." — Astro Teller

Learning Per Dollar: The Metric That Replaces Progress

X doesn't measure progress — it measures learning per dollar. In early-stage moonshot work, the goal is to answer the scariest questions as cheaply as possible, not to show activity or momentum. This reframe prevents founders from rowing hard in the wrong direction.

The biggest mindset shift Astro Teller demands from Xers is abandoning progress as the primary metric. Progress, he argues, is a smokescreen — it makes teams feel good without actually reducing risk. Instead, X's internal quarterly reviews are called 'quarterly learning reviews,' explicitly to stop teams from reporting activity rather than insight.

Teller uses a vivid analogy: 'If we're lost in a rowboat, I'm going to spend ninety percent of my time with the sextant and ten percent of my time rowing.' The corollary for founders: if you can see your goal, run fast. But if your goal is over the horizon — or might not exist — learning is the only thing that matters. Every dollar spent should answer a question, not just build something.

"People have learned that progress is the ultimate metric instead of learning being the ultimate metric. But especially while we're at X, because we're not yet company building — information gain, the learning per dollar, is what we're actually trying to maximize." — Astro Teller

• Quarterly reviews at X are called 'learning reviews' — not progress reports
• The goal: maximize information gain per dollar spent
• Progress without learning is 'rowing frantically in the ocean'
• X graduates companies at seed or Series A stage — years of learning, not revenue

Train the Monkey First: Tackle Risk Before You Build

Always attack the riskiest part of a project first. If you can't solve the hard problem, nothing else matters. X calls this 'training the monkey' — a framework that forces teams to sequence work around the biggest unknown, not the easiest task.

The 'train the monkey' framework came from a live conference moment where Teller struggled to explain why teams should work on the riskiest thing first. His analogy: if your project is to get a monkey to recite Shakespeare on a 10-foot pedestal, should you build the pedestal or train the monkey? Building the pedestal first looks like progress — but it's zero percent done on the real risk.

At X, every team member knows to ask: 'What's the monkey?' This means identifying the single assumption that, if wrong, kills everything — and running at it immediately. For Wing's drone delivery project, early monkey-work focused on safety. The team eventually realized that using many cheap hobbyist motors (instead of perfecting complex control systems) made the drone inherently fault-tolerant. Safety dropped as a risk, and the team moved on. For a water-from-air project, the monkey was techno-economics: could the team tell a specific story about reaching a penny per liter within one year? If not, there was no reason to continue.

This logic also determines which ideas never even get started. If the dominant risk is marketing, PR, or government relations — areas X isn't world-class in — those aren't X moonshots. The monkey has to be technical, operational, or scientific.

"If you built the pedestal, you're half done but you're really zero percent done at burning down the risk. Everyone at X will just say as shorthand to each other, 'what's the monkey?' — meaning, how is the thing that you're doing now attempting to burn down the risk on the riskiest part of the problem?" — Astro Teller

Kill Criteria: A Pre-Written Message to Your Future Self

Kill criteria are specific, pre-agreed conditions written at the start of a project that define when it should be stopped. Written when optimism is high, they protect against the sunk-cost bias and zombie startups by giving teams an honest benchmark to return to months later.

The moment a team goes full-time on a project at X, they must write their kill criteria — before emotion and investment cloud judgment. Teller frames it as 'an honesty message from you to your future self.' The criteria should name the specific monkeys and set minimum thresholds: what must be true in 6–12 months for the project to continue?

Critically, Teller insists teams write these criteria themselves. If they don't own the criteria, they won't respect them. And if a team is more committed to keeping their project alive than to finding the truth, Teller sees that as a team alignment problem — not just a project problem. The default at review time: kill the project unless there's a compelling reason not to. That's a hard cultural standard to maintain, but X reinforces it by celebrating the people who stop things, not just the ones who graduate companies.

For founders outside X, the lesson is structural: zombie startups persist because the incentive system never asks founders to honestly evaluate whether to stop. Writing kill criteria early — before the sunk-cost sets in — is one of the few tools that can counteract that bias.

"Write down your kill criteria. This is like an honesty message from you to your future self. At some point, call it a year in the future, the following things should be true... In order for us to be taking this project seriously. And of course, at that moment, I'm going to be feeling so cocky and everything's going to be great." — Astro Teller

"The default is we're going to kill the project. So then I say to you, look, I know you're excited about it, Pablo. But we pre-agreed that this was kind of the monkey. You don't have a really good answer to this. Why would we continue it in the absence of that really good answer?" — Astro Teller

Moonshot Market Fit vs. Product Market Fit

Product market fit — people paying for something — is too narrow for moonshots. Teller calls the deeper goal 'moonshot market fit': understanding the acupressure points of an entire system, building trust with gatekeepers, and designing for industry-level change, not just a sellable product.

Teller doesn't dismiss product market fit — he extends it. For a conventional startup, classic PMF signals matter: are the right people paying? Does revenue exceed cost? Is it pull or push? But for a moonshot trying to change an entire industry, those signals can be misleading. A company that sells to incumbents who don't want the industry to change may build a profitable business while inadvertently protecting the status quo.

His grid example is instructive. X spent years building rapport with grid operators — doing work that looked like product development but wasn't really PMF — just to earn the trust required to eventually transform how electricity grids operate. That's moonshot market fit: fitting with the system you're trying to change, not just the customer willing to write a check today.

The warning for founders: 'If I ask you to get to the moon and you show me that you're a foot closer to the moon every day, not interesting. There's probably a ladder to the moon.' PMF can be a ladder — locally coherent, globally insufficient — if it doesn't connect to the systemic change you're actually trying to drive.

"We want moonshot market fit, which is much more holistic. Often we're trying to change a system out in the world. So finding some little thing that people will buy — which in some sense is a product market fit — actually doesn't have a very good chance of changing the system." — Astro Teller

Traditional Startup vs. X Moonshot Factory Approach