Oct 11, 2026 · 19 min · 9 segments
On May 19, 2026, Mark Zuckerberg cancelled Meta’s planned second wave of Project OT layoffs after the first wave cut roughly 8,000 jobs, following internal data showing AI agents drove activity…
There is a particular kind of decision that only gets made at night, when the paperwork is already loaded and the severance calculators have already run, and the person making it has to decide whether to let a machine finish a sentence he started himself.
On the evening of 19th May 2026, Mark Zuckerberg cancelled the second wave of a plan he had personally championed.
The first wave had already gone out.
Roughly 8,000 people, about a tenth of Meta's workforce, told by email that their jobs no longer existed.
The second wave, scheduled for November, was meant to be bigger still, deeper into the organisation, broader across roles, projected by one internal executive to rival the 25% cull Meta had run three years earlier under the banner of efficiency.
It never happened.
It is a story about a company discovering, through its own instruments, that it had never properly understood what it was counting in the first place.
The vision was simple enough to fit on a slide.
Agents doing the daily work, small, talent-dense pods supervising them, a generic new job title, Builder, replacing the old taxonomy of engineering roles entirely.
Some teams were modelled to shrink by 60%.
Meta began capturing employees' keystrokes, clicks and screenshots explicitly to train the agents that would eventually do their jobs.
Internal morale, tracked twice yearly, fell from 74% favorable to 55.
And then the agents got to work, and the telemetry started telling a story nobody in Hawaii had written.
Here are the two numbers that contain everything.
Changes that actually reached users as new or upgraded features rose 36%.
That is not a disappointing number sitting beside an impressive one.
It is a ratio.
For every six units of additional activity, one unit of additional value delivered to an actual human being.
The largest social media company on earth more than tripled its rate of change and barely moved the rate at which change meant anything to anyone.
and the activity was not free.
Internal posts tied unchecked agent behaviour to a 40% rise in major technical incidents, and a rise of up to 70% in the time staff spent cleaning up after them.
Agents taking, in the company's own careful phrase, large-scale destructive actions of a kind humans would be unlikely to execute.
There is a particular kind of decision that only gets made at night, when the paperwork is already loaded and the severance calculators have already run, and the person making it has to decide whether to let a machine finish a sentence he started himself.
On the evening of 19th May 2026, Mark Zuckerberg cancelled the second wave of a plan he had personally championed.
The first wave had already gone out.
Roughly 8,000 people, about a tenth of Meta's workforce, told by email that their jobs no longer existed.
The second wave, scheduled for November, was meant to be bigger still, deeper into the organisation, broader across roles, projected by one internal executive to rival the 25% cull Meta had run three years earlier under the banner of efficiency.
It never happened.
It is a story about a company discovering, through its own instruments, that it had never properly understood what it was counting in the first place.
The vision was simple enough to fit on a slide.
Agents doing the daily work, small, talent-dense pods supervising them, a generic new job title, Builder, replacing the old taxonomy of engineering roles entirely.
Some teams were modelled to shrink by 60%.
Meta began capturing employees' keystrokes, clicks and screenshots explicitly to train the agents that would eventually do their jobs.
Internal morale, tracked twice yearly, fell from 74% favorable to 55.
And then the agents got to work, and the telemetry started telling a story nobody in Hawaii had written.
Here are the two numbers that contain everything.
Changes that actually reached users as new or upgraded features rose 36%.
That is not a disappointing number sitting beside an impressive one.
It is a ratio.
For every six units of additional activity, one unit of additional value delivered to an actual human being.
The largest social media company on earth more than tripled its rate of change and barely moved the rate at which change meant anything to anyone.
and the activity was not free.
Internal posts tied unchecked agent behaviour to a 40% rise in major technical incidents, and a rise of up to 70% in the time staff spent cleaning up after them.
Agents taking, in the company's own careful phrase, large-scale destructive actions of a kind humans would be unlikely to execute.
The rest of this transcript — segmented and speaker-labeled, so you land on the exact moment something was said
Search every transcript — by keyword, by phrase, or by meaning, across every show Radar indexes
Trends — what is surging across podcasts, measured against its own baseline
Alerts — when a name you follow appears in a newly indexed episode
No account is needed to search Radar.
