Mike Litman
Edited by Humans
A TALK BY MIKE LITMAN

Edited by Humans

At Google, 75% of new code is AI-written. At Anthropic and OpenAI, 100%. The job moved from writer to editor.

Edited by Humans · 2026

The humans stopped writing first.

It happened faster than anyone said it would.

THE NUMBER
75%

Share of new code at Google now written by AI. Stated by Sundar Pichai at Cloud Next, April 2026.

PICHAI · CLOUD NEXT '26 SOURCE: ALPHABET IR
18 MONTHS, 50 POINTS

How we got from a quarter to three-quarters

100%
75%
50%
25%
0
25%
30%
50%
50%
75%
Pichai · Q3 '24
Q1 '25
Q3 '25
Q4 '25
Today
Q3 '24Q1 '25Q3 '25Q4 '25Cloud Next '26

Source · Alphabet IR · blog.google

01
THE JUMP

50 to 75 happened in a single quarter.

From Q3 2024 to Q4 2025, AI-written code at Google climbed from 25% to 50%. That's 18 months for 25 points.

From February 4 to April 22 of 2026 — eleven weeks — it climbed another 25 points. The last jump was as big as the first four combined.

The curve isn't slowing down. It's accelerating.

+25 POINTS IN 11 WEEKS FEB 4 → APR 22, 2026 ACCELERATING
02
THE MECHANISM

The shift wasn't the model. It was the agent.

Until late 2025, AI wrote code as autocomplete. A sentence here, a function there. Useful, but still the human at the keyboard doing most of the typing.

Then coding agents went mainstream. Claude Code. Cursor. OpenAI Codex. Copilot Workspace. The contract changed: write a ticket, get a pull request. The human stopped writing lines and started writing tasks.

Cursor hit $2B ARR by February 2026. GitHub Copilot crossed 20 million users. The agents didn't arrive. They overtook. That's how 50% becomes 75% in under three months. Not a better model. A different mode.

AUTOCOMPLETE → AGENT CURSOR: $2B ARR COPILOT: 20M USERS CLAUDE CODE CODEX
GOOGLE ISN'T ALONE

Every frontier company has crossed the line.

Anthropic
~100% · "pretty much 100% of code is AI-generated"
CHERNY · JAN 2026
OpenAI
100% · "I don't write code anymore"
ROON · JAN 2026
Google
75% of new code
PICHAI · APR 2026
Meta
50% target within a year; some orgs 50–80% today
ZUCKERBERG · APR 2025
Microsoft
20–30% overall; some repos past 50%
NADELLA · APR 2025
Amazon
79% of auto-generated code shipped without edits
JASSY · 2024

Methodology varies by company · Directionally consistent

03
THE PATTERN

The closer you are to building AI, the closer to 100%.

1
Frontier labs
The companies training the models. Anthropic. OpenAI.
~100%
2
Users of AI at scale
Heavy internal adoption, not the ones training the models. Google.
75%
3
Sellers of AI tooling
Shipping the coding assistants, catching up on their own adoption. Microsoft. Meta.
30–50%
4
Everyone else
Not publicly measuring yet. The rest of the software industry, and every adjacent creative field.
THE LAG

The builders are 12 to 18 months ahead of the users.

Which means the rest of the industry is already on the curve. They just haven't looked down yet.

04
THE FLIP

Humans used to write and review. Now they only review.

The old contract: an engineer writes the first draft, another reviews it. Both are authors. Both leave their fingerprints.

The new contract: the model writes the first draft, the human accepts, rejects, or redirects. The model is the author. The human is the editor.

The labour moved one step up the stack. The writer's chair became the editor's desk.

WRITE → REVIEW DRAFT → DIRECT ONE STEP UP THE STACK
THIS HAS HAPPENED BEFORE

When the camera arrived, illustrators moved up the stack.
When the model arrived, engineers moved up the stack.

Same pattern, different tool. Every time the drafting got mechanised, humans moved to judgement. Not a new crisis. A familiar move.

PHOTOGRAPHY → ILLUSTRATION AI → HAND-CODING NOT THE FIRST TIME
05
THE COUNTER-QUESTION

So what do 5,000 Anthropic engineers do all day?

If AI writes 100% of the code, the fair question is: what's the job?

Spec the system. Review the output. Architect the trade-offs. Test what matters. Debug what slips. Decide what to ship next. None of that scales linearly with output — more code means more reviewing, more directing, more judgement.

Anthropic employs around 5,000 people to do this. The human work didn't get cheaper. It got denser. A team of editors producing the output of an army of writers.

SPEC · REVIEW · ARCHITECT TEST · DEBUG · DECIDE ANTHROPIC: ~5,000 STAFF
THE JOB
Then
Write the code. Get reviewed.
Humans produced the first draft. Review was a check. The engineer's value was in the typing.
Now
Direct the model. Edit the draft.
Models produce the first draft. Review is the work. The engineer's value is in the judgement.
06
NOT JUST CODE

Every creative role has a first-draft task.

The strategist writes the deck. The brand director writes the campaign. The copywriter writes the page. The designer drafts the layout. The lawyer drafts the clause. The analyst drafts the model.

In every one of those roles, the first draft is a cost, not a product. It's what has to exist before the judgement can begin. Whatever can produce that draft faster will. Whatever can produce it for free will win.

Adobe Firefly generated 22 billion creative assets in its first two years. I've shipped 20+ AI-native products in 18 months. I didn't write the first draft of any of them. The design curve is already well up. The strategy curve is right behind.

The ceiling varies. Novel research still hits limits. Boilerplate doesn't. Taste-dependent work lands somewhere in between. But every job has a first-draft task. And every first-draft task is on the curve.

FIREFLY: 22B ASSETS MIKE: 20+ PRODUCTS, 0 FIRST DRAFTS STRATEGY BRAND COPY DESIGN
THE CANARY

Code is the canary because code is countable.

Tokens shipped. Commits merged. Lines written. You can put a number on it. Strategy, brand, editorial — the numbers are fuzzier. The curve is the same.

WHAT THIS MEANS FOR YOU

The new skill is taste.

For strategists & brand directors
You have a head start. Your job was already editorial — reviewing work, directing drafts, spotting what's wrong. That's the skill that matters now. The only thing that's changed is the volume of drafts coming through.
For engineers & writers & designers
The craft hasn't gone away. It moved. The value isn't in the first draft anymore. It's in knowing what "right" looks like and rejecting everything that isn't it.
For CMOs & heads of
The org chart has to flatten. If every IC can ship ten drafts an hour, the bottleneck is the reviewer. Your job is to make more editors, not more writers.
The honest split
The engineering layoffs at Google, Meta, and Microsoft in 2025–2026 are the early signal. Not everyone who survives this shift gets promoted — some get replaced. The people who learn to edit early choose their side.
07
MY VERDICT

The skill is recognition, not specification.

I don't always know exactly what I want up front. But I know when the output isn't it — and that's enough. Reject, redirect, refine. Not because I can describe the right answer in advance, but because I recognise the wrong one the moment it arrives.

Twenty-plus AI-native products in 18 months. I didn't write the first draft of any of them. I knew when each one wasn't ready. That's the whole skill.

RECOGNITION BEATS SPECIFICATION 20+ PRODUCTS · 0 FIRST DRAFTS REJECT → REDIRECT → REFINE
THE NEW PREMIUM

"Edited by Humans" becomes a quality signal.
Like grass-fed. Like hand-stitched.

When the machine writes the first draft of everything, human judgement becomes the premium. The stamp goes on the work worth paying for.

Edited by Humans · 2026
A TALK BY MIKE LITMAN

The first draft belongs to the machine.
The last call still belongs to you.

mikelitman.me · hello@mikelitman.me

Continue from slide ?