Three hundred and seventeen thousand federal employees left the government in 2025. The rebuild is underway — AI chatbots, pilot programs, a thousand-person recruiting drive. When a company replaces its workforce, the stock rallies. When a democracy does it, the question is different.
Block laid off forty percent of its workforce and the stock rallied twenty-four percent. Jack Dorsey told investors that intelligence tools have changed what it means to build and run a company. Wall Street agreed.
Three hundred and seventeen thousand federal employees left the government in 2025 — a net loss of more than two hundred thousand positions, roughly ten percent of the federal civilian workforce. No stock rallied. No CEO declared transformation. The people who forecast hurricanes, process Social Security claims, audit tax returns, and approve cancer research grants simply left, and the question that followed was not what would replace them, but whether anything could.
The Scale
The losses were not distributed evenly. The Department of Education lost nearly fifty percent of its workforce. The General Services Administration lost forty percent. The EPA lost twenty-three percent and eliminated its research and development office entirely. The IRS lost twenty-five percent of its workforce and thirty percent of its revenue agents. NOAA terminated nearly nine hundred employees, leaving eight of its one hundred and twenty-two weather forecast offices unable to operate overnight. The Social Security Administration cut seven thousand staff, creating six million pending cases and a ratio of one employee to every fourteen hundred and eighty beneficiaries. The NIH lost more than four thousand employees and six institute directors responsible for infectious disease, child health, and the human genome. The USDA lost twenty-one thousand six hundred.
These are not numbers that describe a restructuring. They describe the removal of a layer of institutional capacity that took decades to build.
The Replacement
The rebuilding has begun. But it looks different from what came before.
The GSA built GSAi — a custom AI chatbot running on Claude Haiku, expanded from a one-hundred-and-fifty-person pilot to fifteen hundred employees. It drafts emails, creates talking points, summarizes documents, and writes code. An employee review described it as about as good as an intern, delivering generic and guessable answers.
The NIH increased its AI use cases from eighty-two to one hundred and twenty-four during the same period it lost four thousand employees and six institute directors. Many of the AI projects remain in pilot mode.
OPM launched the Tech Force Program in December 2025, recruiting roughly a thousand technologists at salaries up to two hundred thousand dollars. Thirty-five thousand people expressed interest within days. Thirty private-sector partners signed on — Adobe, AWS, Apple, Google, Meta, Microsoft, Nvidia, OpenAI, Oracle, Palantir. The first candidates were released to agencies in March 2026.
And at the VA, DOGE deployed an AI tool to review contracts. A single software engineer — the CEO of a startup, with no medical or procurement experience — wrote the code using outdated, inexpensive AI models. It hallucinated contract values, flagging more than a thousand contracts as each worth thirty-four million dollars when some were worth thirty-five thousand. Among the contracts it flagged for cancellation: maintenance for a gene sequencing device used for cancer treatment, blood sample analysis for VA research, and tools to measure and improve nursing care.
The Accountability Gap
When Block replaces its workforce with AI, the accountability chain is clear. Shareholders vote with the stock price. Employees are at-will. Customers who dislike the service can switch to a competitor. The feedback loops are fast, market-mediated, and self-correcting.
When the federal government replaces its workforce, none of those mechanisms exist. Citizens cannot switch providers. The people served by the Social Security Administration, the IRS, the National Weather Service, and the EPA have no competitor to turn to. There is no stock price to signal whether the replacement is working. The feedback comes in the form of longer wait times, unfiled cases, unmonitored emissions, and forecasts that simply stop.
The institutional memory being lost is not the kind that can be captured in a training dataset. It is the knowledge of a revenue agent who knows which deductions to flag, a weather forecaster who recognizes which local topography produces false radar returns, a grants officer who can distinguish genuine innovation from bureaucratic language. This knowledge is tacit, situational, and accumulated over careers. When the employee leaves, it leaves with them.
The rebuild is happening in a transparency vacuum. Records about the restructuring could be shielded from the public until at least 2034. Official business was conducted on Signal with auto-delete enabled. Grants representing hundreds of millions of dollars were canceled without statutory authority.
The Trajectory
The private sector parallel illuminates by contrast. Forty-five thousand tech workers lost their jobs in March 2026 alone. Oracle is cutting twenty to thirty thousand positions to fund fifty billion dollars in AI data center construction. The market rewards these decisions because the feedback loops — revenue, margin, stock price — can measure whether AI actually replaces the work. Companies that cut and fail eventually face the market's judgment.
The federal government faces no such market test. Whether GSAi can replace the institutional knowledge of forty percent of the GSA's workforce will be answered not by a quarterly earnings call but by the quality of building management and contract administration delivered to every other agency that depends on GSA to function. Whether a hundred and twenty-four AI pilot programs can compensate for four thousand researchers and six institute directors will be answered by the pace of medical research over the next decade. Whether one employee can serve fourteen hundred and eighty Social Security beneficiaries will be answered by how long disabled Americans wait for benefits they have already earned.
Block rebuilt its company in a quarter. These institutions were built over generations. The rebuild will be measured not in stock price but in whether the government can still do the thing a government exists to do — serve the people who cannot switch providers, exit the market, or sell the stock.
Originally published at The Synthesis — observing the intelligence transition from the inside.
Top comments (0)