LA Unified's AI chatbot went dark three months after its launch

The classroom note

On 20 March 2024, at an event at the Edward R. Roybal Learning Center, Los Angeles Unified and its vendor AllHere unveiled Ed, described as an AI-powered educational co-pilot that would analyse a student's grades, attendance and district data and recommend resources in dozens of languages. Board members praised its promise at launch, and the district's own release called it a first-of-its-kind platform for the nation's second-largest school system. Three months later there was no platform left running.

What the evidence says

Reporting by The 74, a nonprofit education newsroom, states that AllHere's chief executive left the company and it furloughed most workers amid financial problems in June 2024, and that district officials pulled Ed offline immediately because no AllHere staff remained to supervise it. A later investigation by the same outlet, drawing on public-records requests, reports the board had approved a $6.2 million contract with AllHere, the district had paid roughly $3 million of it by the collapse, and AllHere filed for bankruptcy in September 2024. Officials are reported as saying the district was not financially harmed and had received the services it paid for; that is an official's account relayed by reporters, not an audited finding.

The implementation question

The mechanism worth naming is dependency, not the AI model. A learning-analytics platform reading grades, attendance and district data needs continuous vendor operation to run safely: someone must supervise its outputs, patch it, and answer for what it tells a family. When the vendor lost the staff to do that, the district's only safe option was to take the tool down, regardless of how much of the $6.2 million contract had already been spent. That is a single point of failure a procurement decision can create even while the underlying technology keeps working.

What holds and what fails

A pilot with a small, reversible commitment and a named exit plan can hold even if a vendor later struggles, because the district's exposure was bounded from the start. A large, district-wide contract with a single early-stage vendor for a service touching every student's records does not hold under the same stress: reliance on Ed's continued operation, and the recommendations it had already given families, needed no fault on the district's part to become a liability once AllHere failed. Editorially, the launch's own language, describing itself as first to systematically leverage AI, is the kind of claim that should have been paired publicly with a contingency plan, not only a task force.

• What happens to the tool, and to guidance it already gave families, if the vendor cannot continue operating?
• Is the contract's size matched to how established and financially stable the vendor is, or to how the launch will look?
• Who at the district can pull a live AI tool offline quickly, and has that authority been exercised before?

A launch event proves a platform can be switched on; it says nothing about who keeps it running once the company behind it cannot.