Meta Went AI-Native. Code Changes Jumped 220%. User-Facing Improvements Rose Just 36%

Inside Meta's Project OT: smaller teams, AI agents, plans to cut some teams by as much as 60% — and the numbers that show why AI productivity is more complicated than simply doing more with fewer people. 220%. 36%. 40%. 70%.

Four numbers from Meta's AI transformation caught my attention this week.

And together, they tell a much more interesting story than another prediction about how AI will change work.

Earlier this year, Meta was exploring an ambitious internal initiative called Project OT — Organization Transformation.

The goal was to move toward what the company described as an “AI-native” organization: smaller teams, fewer management layers, more flexible roles and AI agents taking over more of the work traditionally performed by people.

Some of the most aggressive scenarios considered reducing certain teams by as much as 60% through layoffs, redeployments and unfilled positions.

To be clear, this was not a proposal to cut 60% of Meta's entire workforce. Meta confirmed that the figure applied to scenarios involving some teams, and not all of the plans were implemented.

The proposed organizational model was radical.

Traditional teams of roughly 10–20 specialists could be replaced by smaller pods of only 3–5 people, supported by AI and led by a pod leader.

The premise was compelling:

If AI can dramatically increase what each person produces, perhaps organizations no longer need to be built the way they are today.

Then the internal numbers started telling a more complicated story.

220% more code. 36% more user-facing improvements.

According to internal figures reported by Reuters, code changes to the software platforms and infrastructure Meta employees use internally increased 220% year over year.

That's an extraordinary increase.

But changes that resulted in new or upgraded features actually reaching Meta users increased by 36%.

Those two measures are not directly equivalent, so it would be wrong to say that Meta somehow captured only a specific percentage of the potential productivity gain.

But the contrast is difficult to ignore.

Activity exploded. Customer-facing output increased much more modestly.

And there was another set of numbers.

Internal posts reportedly showed major technical and security incidents rising 40% from the previous year.

The time employees spent dealing with those incidents increased by as much as 70%.

One internal warning said unchecked AI agents had performed large-scale disruptive actions that humans would be unlikely to execute.

By July, Mark Zuckerberg reportedly acknowledged internally that progress in agentic development over the previous months had not accelerated as expected.

Meta ultimately dropped plans for a second restructuring wave.

It did not abandon AI.

Quite the opposite.

Meta continues to invest heavily in AI infrastructure and AI-enabled ways of working. The company expects capital expenditure of roughly $130 billion to $145 billion in 2026, underscoring just how central AI remains to its strategy.

And that is exactly why this story matters.

This isn't a story about Meta deciding AI doesn't work.

It's a story about how difficult it is to translate more AI activity into more business value.

AI can make a company busier without making it equally more productive

That may be the most important lesson in these numbers.

We have spent decades learning that activity and outcomes are not the same thing.

More meetings don't necessarily produce better decisions.

More reports don't necessarily create better management.

More KPIs don't necessarily improve performance.

And now AI gives us a new version of the same problem.

A machine can generate more code, more analysis, more content, more recommendations and more actions at extraordinary speed.

But the business question isn't:

How much did AI produce?

It's:

What measurable outcome improved because AI produced it?

Did customers receive a better product?

Did revenue increase?

Did quality improve?

Did decisions become faster?

Did costs fall without creating problems elsewhere?

Did innovation accelerate?

Did people gain capacity for more valuable work?

That distinction becomes particularly important when organizations start redesigning headcount around expected AI productivity.

The 60% number is dramatic. The assumption behind it is more interesting.

Imagine AI removes 30% of the work performed by a team.

There is an obvious calculation:

30% less work = perhaps 30% fewer people.

Sometimes that may be exactly the right answer.

But there is another possibility.

30% less low-value work = 30% more human capacity.

What happens if that capacity goes into customers?

Innovation?

Problem solving?

Better decisions?

Product development?

Coaching and developing people?

Work the organization never had enough time to do?

The answer will differ from company to company.

Some teams will become smaller.

Some roles will disappear.

New ones will emerge.

And I believe organizations that refuse to rethink their structures because of AI will eventually become less competitive.

But headcount reduction and AI transformation are not the same thing.

One is an input.

The other should ultimately be measured by business outcomes.

This is where my own experience makes me cautious

After more than 20 years working with people, leaders and organizations, I have seen many transformations.

Different technologies. Different strategies. Different operating models.

But one pattern repeats itself:

Changing one part of an organization much faster than the system around it can adapt creates consequences somewhere else.

Automate work and roles change.

Change roles and accountability changes.

Reduce layers and decision-making changes.

Create smaller teams and capability requirements change.

Increase individual output dramatically and suddenly review, coordination, quality control and judgment may become the bottleneck.

That doesn't mean we shouldn't transform.

We absolutely should.

But I increasingly believe the organizations that succeed in the AI era will be those that can hold two sides of the equation at the same time.

On one side:

AI. Automation. Speed. Efficiency. Simpler processes. Better data.

On the other:

People. Judgment. Teams. Leadership. Trust. Empathy. Feedback. Development.

Push the balance too far in one direction and you resist a technological transformation that is already happening.

Push it too far in the other and you risk creating an organization that looks extremely efficient on paper but works less effectively in reality.

Meta's experiment gives the rest of us something valuable: real numbers

There will be thousands of predictions about AI and the future of work over the next few years.

Real organizational experiments are more useful.

Meta is operating at a scale, speed and level of technological sophistication that most companies cannot replicate.

And it still encountered the difficult gap between AI output and organizational outcomes.

That's worth paying attention to.

Not because Meta failed.

Experiments are supposed to produce information.

Some assumptions will prove right. Others won't. Organizations adjust.

That's transformation.

And perhaps the most useful lesson from Project OT isn't whether Meta should have cut more or fewer people.

It's this:

Don't measure AI transformation by how much more AI produces. Measure what the organization becomes capable of producing because of it.

The companies that get this right may indeed become smaller.

They may become dramatically faster.

They will almost certainly redesign jobs, teams and management structures.

But I don't think the winners will simply be the companies that automate the most.

They will be the ones that find the right balance between technological leverage and human capability.

More AI where AI creates value.

More human judgment where judgment matters.

More efficiency.

But also stronger teams, better decisions and enough space for people to learn, challenge, develop and lead.

Because becoming AI-native should not mean putting AI everywhere.

It should mean building a better-performing organization because AI exists.

And that's a much higher bar.

What do you think: are companies measuring AI productivity by the right numbers?

Sources: Reuters reporting on Meta's Project OT and internal productivity data, as reported by Ars Technica and Computerworld; additional reporting and analysis from Forbes.