AI Throughput Gap
AI made the typing faster. This works out how much faster your team actually ships, the most faster coding could ever buy you, and which stage now decides when work goes out. Put in your own days. Nothing you enter leaves your browser.
Calculator
Your team ships about 1.5x faster: a change that took 10 days now takes about 6.7 days.
Even if code wrote itself, the most your team could get is 2.0x, because the other stages don’t move.
Deciding and reviewing are now your longest stages. Halving the time spent deciding gets you 1.8x. Doubling the AI speed-up on building again gets you 1.7x. The next gain is in deciding, not in faster coding.
These are your estimates, not measurements. The Engineering Deep Dive measures them from your own delivery data.
Why there's a ceiling
A change doesn't ship when the code is written. It ships when someone has decided what to build, built it, reviewed it, released it and checked it worked. AI coding tools speed up one of those. The rest stay exactly as slow as they were.
Take the example above: 10 days end to end, 5 of them building. Make building 10 times faster and the change takes 5.5 days, so the team is about 1.8x faster, not 10x. Make building instant and it's still 5 days. That's the ceiling, and it's set by the stages AI doesn't touch.
So once building stops being the longest stage, more AI seats are the wrong next spend. The gain is in whichever stage the calculator names: usually review, release or the time it takes to decide.
Join the Drift Digest
One short email every couple of weeks: practical fractional CTO and founder-AI notes, new AI tells as models change, and a quarterly nudge to re-measure your voice.
Something went wrong. Please try again, or email web@ctoondemand.co.uk.
Common questions
Why does AI make engineers faster but not the team?
A change goes through several stages: deciding what to build, building it, reviewing it, releasing it and finding out whether it worked. AI coding tools speed up the building. The other stages stay as slow as they were, so the time saved on building shrinks as a share of the whole, and the team-level gain is capped by how much of lead time building was in the first place.
What is the most faster coding can buy a team?
Divide your total lead time by the time left once building is taken out. If building is half of lead time, an infinitely fast AI doubles delivery at most. If building is a quarter, the most it can buy is about 1.3x. This is Amdahl’s law applied to delivery, and it holds whatever the tools get better at, because it is about the stages AI does not touch.
How do I find out how long each stage really takes?
Most teams estimate, and the estimates are usually wrong in the flattering direction. The real numbers come from your pull request history, your ticket tracker and your deploy pipeline. The Engineering Deep Dive in the AI Discovery Report measures them, stage by stage and team by team.
Want the real numbers, not estimates?
The Engineering Deep Dive measures every stage from your pull requests, tickets and deploys, team by team, and tells you which one to fix first. If faster coding is still your best lever, it'll say so.