Moneyball for AI agents: four lessons for engineering leaders
In 2020, Moneyball taught us to trust the right numbers. In 2026, the same idea decides which AI coding agents are worth paying for.
Originally published January 13, 2020 · Updated November 2, 2026
While the winter holidays are always the perfect time to disconnect from work and enjoy time with family and friends, they usually come at the price of a somewhat painful return to our work lives. To help reduce the stress, we decided to start a new tradition and kick off the year with a good movie or two. Since it had to be something both entertaining and inspiring, we chose a classic: Moneyball, the 2011 film starring Brad Pitt.
Based on true events, it tells the story of Billy Beane, General Manager of the Oakland Athletics baseball team, who started using sabermetrics, an empirical analysis of baseball that uses statistics to improve team performance. With this new approach, Beane took the Oakland Athletics from a low-performing team to a success that shocked the sports world, including a run of 20 consecutive wins.
Six years after we first wrote this, the lessons matter more than ever, because engineering teams now have a new kind of player on the roster: AI agents. Here is what we learned, then and now.
Learn to embrace change (especially when it’s the only option)
The movie reminded us that change is never easy, and it often comes as the last resort: an option that is unpopular, unwanted and fraught with danger. The story does not portray Beane as a visionary who chose a new path because he was bored. He did it because the team was sinking, and it had become clear that the old strategy would not get them anywhere.
Whether you are managing a sports team, a marketing team or an engineering team, there will come a time when a big change is what helps you reach daring goals. So, when was the last time you took a risk with your team: switching to a new tool, changing the way you hold meetings, or looking at your team’s performance in a different light?
There’s magic in numbers
“People are overlooked for a variety of biased reasons and perceived flaws. Age, appearance, personality. […] Mathematics cuts straight through that.”Peter Brand, Moneyball
The movie shows how statistics helped the Oakland Athletics build a winning team, and the same is true in business. No manager can afford to ignore the numbers. Whether you are looking at sales reports, financial projections or a new team structure, the best decisions come from thorough analysis, not gut feeling.
In engineering, we have learned one important refinement since 2020: measure teams and systems, not individual people. The most useful numbers show where work gets stuck, how delivery flows and what it costs, so leaders can fix the system rather than rank the people in it.
Others will follow (but probably not when you need them to)
There is a powerful moment near the end of the movie where the owner of the Boston Red Sox, realizing that sabermetrics will be the future of the game, tells Beane he was “the first guy through the wall,” and that it always gets bloody. Change feels threatening to everyone, not only to the people making it, so expect resistance from all sides. Others follow only after they see that it works.
Two years after Beane turned down the Red Sox’s offer to become their general manager, a reported $12.5 million deal, the Red Sox won the World Series using the strategy he helped pioneer.
Your AI agents need their own Moneyball.
Most engineering organizations now pay for several AI agents at once: an assistant in the IDE, a coding agent, a review agent, and sometimes agents built in-house. And most measure them the way baseball scouts once measured players, by what is easy to see: how much code each agent writes, and how many of its suggestions are accepted.
That is the equivalent of judging a hitter by how hard he swings.
AI spend per developer per quarter, in one year
share of time on new features, barely moving
Source: DX, State of AI Impact in Engineering: Q3 2026 Report.
Billy Beane’s insight was simple: pay for the statistics that actually produce wins, not the popular ones. For AI agents, a “win” is code that reaches production and stays there, at a reasonable cost. So we translated sabermetrics into metrics for agents:
| Moneyball | For AI agents | What it measures |
|---|---|---|
| On-base percentage | Survival rate | Agent code merged and not reverted within 30 days |
| Slugging | Weighted impact | How large and complex the surviving changes are |
| Cost per win | Cost per surviving PR | Token spend divided by PRs that stay in production |
| WAR | Agent WAR | Value above a reference agent, on the same kind of work |
survival = surviving_prs / merged_prs
value = impact – rework
cost = token_spend / surviving_prs
agent_war = (value – reference_value) / 100 PRs
Put together, these give every AI agent a single score: Agent WAR, the value it delivers above a reference agent, compared only on the same kind of work. Put an agent only on easy tasks and it won’t look better than it is.
When teams look at their agents this way, the results can be surprising:
Agent WAR · Q3 illustrative
+4.1
+2.6
+0.4
−1.2
The most-used agent ranks third on value, and each PR it gets into production costs 3× more than the top agent’s.
One rule above all: Agent WAR ranks AI agents, never people.
Engineers aren’t players to be traded on a scoreboard. The point is to decide which agents to pay for, and which kind of work to give each of them. The building blocks are already in Waydev, and Agent WAR is the next step we’re working on: turning them into one clear ranking of your agents.
The lesson hasn’t changed. The players have.
In 2002, the teams that trusted the right numbers won with smaller budgets. In 2026, the same will be true for AI. The companies that measure what their agents actually deliver, not how busy they look, will get more from every dollar they spend on AI. And, as Lesson 3 reminds us, everyone else will follow, just a little later.
See how your AI agents really perform.
Alex Circei
CEO and Co-Founder, Waydev