The pitch was irresistible, which should have been the first warning. Every one of the roughly half a million students in the Los Angeles Unified School District would get a personal AI assistant. It would track their grades. It would nudge them when attendance slipped. It would offer mental health resources at two in the morning to a fourteen year old who had nobody else to ask. The district named it Ed and unveiled it in February 2024 with the kind of press event usually reserved for a new stadium.
Three months later, Ed was gone. The vendor was in bankruptcy. Its founder had been indicted. And on June 21, 2026, Alberto Carvalho, the superintendent who stood on that stage and promised the future, resigned from the second largest school district in the United States, four months after federal agents searched his home and his office.
What Ed Was Supposed To Be
AllHere was an education technology startup founded by Joanna Smith-Griffin. Its product was a chatbot that would sit between a district and its families, handling the enormous, unglamorous communication load that no school system has ever had enough staff to carry. Reminders. Check-ins. Answers to the same forty questions a thousand times a day. On paper this is one of the genuinely defensible uses of a language model, the kind of deployment that does not require anyone to believe the machine is intelligent, only that it is tireless.
Carvalho believed in it publicly and enthusiastically. He described Ed as a personal assistant for students and parents, capable of tracking academic progress and surfacing educational and mental health support. He was not quietly selling a gimmick to a school board that did not understand it. He was the most visible superintendent in American public education, and he attached his name to the product.
The Collapse Took Ninety Days
By the summer of 2024 the majority of AllHere's staff had been furloughed. The chatbot was no longer available to students. The district, having paid three million dollars, walked away, and the company fell into bankruptcy.
Then came the part that turned an ordinary procurement failure into a federal matter. In November 2024 the Justice Department indicted Smith-Griffin, charging that she defrauded investors by inflating the company's financials to secure millions of dollars under false pretenses. Prosecutors alleged the fraudulently obtained funds were spent on personal expenses. That case remains ongoing and separate from anything involving the district.
A whistleblower raised a second alarm, one that has nothing to do with money. The bot, they said, put students' personally identifiable information at risk of getting hacked. That is the quiet horror underneath the accounting scandal. Whatever else Ed failed to do, it was ingesting the records of children.
The Verdict
A school district bought an AI product from a company that could not build it, from a founder now charged with lying about the company's finances, and the only thing that scaled was the damage. The chatbot lasted twelve weeks. The fallout has now lasted two years and cost a superintendent his job.
February 25: The Search
On February 25, 2026, FBI agents searched Carvalho's home and the district's headquarters. The school board placed him on paid administrative leave. He stayed there for nearly four months. Reporting has tied the federal interest to the failed multimillion dollar AI contract and to questions about conflicts of interest, though no charges against Carvalho have been filed and none may ever be.
He resigned effective June 21, 2026, saying he wanted the district to refocus on students. Andres E. Chait took over as acting superintendent. The most recognizable name in urban school leadership in the country left the job over a chatbot that ran for roughly one academic quarter.
The Part Nobody Wants To Say Out Loud
It is tempting to file this under AI hype and move on. That reading is too easy, and it lets the wrong people off the hook. Ed did not hallucinate a citation. It did not defame anyone. There is no evidence in this story that the underlying language model behaved unusually badly at all. What failed here was procurement.
A district with a nine figure technology budget handed three million dollars to a startup on the strength of a demo, a founder's confidence, and a superintendent's enthusiasm, in a category where nobody involved had a reliable way to evaluate whether the thing being sold could actually be built. That is the oldest problem in institutional purchasing, wearing a very fashionable new coat.
Here is the honest counterpoint, and it matters. The need Ed was built for is completely real. Los Angeles Unified serves families across dozens of languages with a counselor to student ratio that would be considered a scandal in any private school in the county. A tireless multilingual system that answers routine questions at midnight would genuinely help somebody. The failure of this contract does not prove the idea was stupid. It proves that a district buying into an unproven category needs to be able to survive the vendor going bankrupt, and this one could not.
What A District Should Have Demanded
Three things, none of them exotic, all of them standard in other kinds of public contracting. Proof the software already ran at scale before signing, not a pilot and a promise. An escrow or continuity clause so a vendor's bankruptcy does not instantly delete a service that children have been told to rely on. And an independent security review of student data handling before a single record moves, rather than after a whistleblower goes to the press.
None of that requires understanding transformers. It requires treating an AI vendor exactly as skeptically as you would treat a company promising to reroof four hundred buildings.
The Pattern Is Getting Familiar
We keep documenting the same shape in different clothing. An institution adopts an AI product faster than it can evaluate it. The product underdelivers or fails outright. The consequences land on people who never chose the technology, in this case students whose data was exposed and a district that is three million dollars poorer. The executives who championed it usually move on.
Carvalho did not move on. That is the only genuinely unusual detail in this entire story, and it is worth recording, because accountability at this level is rare enough to be news by itself. Keep it filed next to the running timeline of AI failures and the docket of AI litigation. The next district is in a conference room right now, watching a demo, nodding.