The Made vs Shipped Scorecard

01 Why counting output fools you

Figma's chief design officer told TechCrunch there's 17 times more code being produced than before, and only 1.3 times more releases. The same two numbers show up in a study of GitHub developers published through the National Bureau of Economic Research. Files touched rose almost 4 times, changes sent for review 2.5 times, releases 1.3 times. Every step closer to the customer, the gain shrinks.

It isn't only code. In a marketing team it looks like this:

AI writes forty versions of an ad. One person approves ads. A week of blog drafts. One person reads them for legal. Most of a team's work isn't making. Atlassian, the company behind Jira, estimates only about a fifth is. The rest is reviewing, approving, aligning, deciding and waiting.

So the extra output mostly waits. In engineering data from Faros, as teams moved to high AI use, the median time a change spent in review went up about 5 times. You feel the making. You don't feel the waiting, because it happens on someone else's desk.

Stop counting what AI makes. Count what ships.

02 The five numbers

Put these on the wall and fill them in every Monday. Count the same way each week so the trend means something.

Made. Everything the team and its AI produced: drafts, ads, pages, code. Count finished drafts, not ideas.

Shipped. What actually reached a customer: an ad that went live, a page that's published, a feature people can use.

Waiting. The age, in days, of the oldest thing sitting in someone's queue for a yes.

Survived. Of what shipped a month ago, what's still live and wasn't redone. Work that comes back to be redone was never really done.

Reviewer hours. How much time the people who check things spent checking.

Number Marketing team Product team

Made Ads made Features built

Shipped Ads live Features released

Waiting Days in approval Days in review

Survived Ads still running after thirty days Features still live, not redone

03 Your ratio, with a worked example

Divide made by shipped. That's your ratio. If it climbs every month while shipped stays flat, AI isn't speeding you up. It's filling a queue.

A worked example (made-up numbers)

Your team made 120 things last month and shipped 30. That's a ratio of 4. This month AI helps you make 200, and you ship thirty-two. Your ratio is now above 6. Everyone feels busier and more productive. The customer got two more things.

Monday scorecard

WEEK OF: [DATE]

Made: [NUMBER]
Shipped: [NUMBER]
Ratio (made / shipped): [NUMBER] Last week: [NUMBER]
Waiting (oldest item, days): [NUMBER] Whose desk: [NAME]
Survived (of what shipped a month ago, still live, not redone): [NUMBER] of [NUMBER]
Reviewer hours: [HOURS]

Ratio up and shipped flat? [YES / NO]
This week's move: [SMALLER BATCHES / MAKER CHECKS FIRST / FEWER STARTS]

04 Three moves when the ratio climbs

Smaller batches. Faros found changes got bigger as AI use went up, and it tells teams to check whether the big ones take longer to review, and to cap their size if they do. Ten small things clear a gate faster than one huge one. The maker checks first. If you used AI to make it, you read it before anyone else has to. AI work tends to look right, so the problems sit under the surface, and the people who can find them are your most experienced ones. Don't start more than your reviewers can clear this week. A full gate doesn't need more input. It needs fewer starts, or a faster yes.

Point some AI at the gate. Have it check the work against your brief before a person ever opens it, so reviewer hours go to the calls only a person can make.

Watch for skipped reviews. In the Faros data, 31% more changes went in with no review at all, and bugs per developer went up 54%. When the pile gets big enough, people stop checking, so a faster yes only counts if the work still gets read.

Sources

TechCrunch: Figma announces new features to define how agents work on its platform NBER Working Paper w35275: Writing Code vs. Shipping Code Atlassian: The AI efficiency paradox Faros: AI Engineering Report 2026, The Acceleration Whiplash

Mathew's LibraryLibrary card

Free access

Keep reading free with your library card

One email opens this template and all 86 resources in Mathew's library. No account, no password.