Signals · Power
Microsoft's AI Frontiers group is over, and one of its founders' reasons are a warning about where good work stops protecting you
A 50-person Microsoft research group was folded into a scaling-first division, and one of its founders says doing good work no longer guaranteed good outcomes.
by Jo·3 min read·
New here? Start with the free AI Survival Kit →
Microsoft's AI Frontiers group is over, and one of its founders' reasons are a warning about where good work stops protecting you
0:00 / 5:29
This voice is generated by AI.
Part of the guide Will AI take your job? How to find out for your role, in an afternoon
The author of the hunch.net blog, a Microsoft Research scientist who helped found the AI Frontiers group, wrote on 6 November that the group is finished. Microsoft Research set up AI Frontiers in 2023, after ChatGPT's release, and it grew to about 50 people. In early September it was reorganised into Microsoft AI, a much larger division run by former Inflection leadership. The author has left. He writes that every other member of the group's founding leadership team "seems to have come to a similar conclusion", and that the group will offer no internships in the coming year.
You could read this as research-lab politics. It is a story about what protects a team once the people in control change direction.
What the reorg decided
AI Frontiers had a real track record. According to the post, it built the Autogen orchestration framework and the Fara computer use agent. In New York it worked on new capabilities: the Next Latent objective, Dion3 and the Full Bandwidth Transformer. It also built a shared monorepo that held a working pretraining codebase, open by default to the whole of Microsoft Research.
None of that saved it. The author describes Microsoft AI's philosophy as "scaling, and aggressively not research", with most of the effort going into getting, making and refining data. He concedes the business logic: returns to scale are "far more predictable than returns to research." He argues that research still pays. ByteDance experiments, he writes, suggest a variant of the Full Bandwidth Transformer "might be comparable to a model of twice the size", which would mean large savings on training and inference.
The fit was poor on both sides. In his words, "there was plenty of evidence that they just did not really want us."
Good work stopped being enough
The most useful line in the post is his description of the deal that broke: "do good work and good things happen." He says the team did good work and was "non-consensually reorged away". He also declined to go back to Microsoft Research. He could not get his ideas into use, "or even seriously considered", inside Microsoft AI, so he judged that research done from Microsoft Research had little chance of getting there either.
That is the operator's lesson, and it applies far beyond AI labs. Quality is the entry ticket. What keeps work alive is being on the path that the budget-holder's philosophy funds. When the reorg put it under leadership whose philosophy is scaling, a 50-person team with shipped frameworks to its name ended up on the wrong side of a single decision.
The author's response is worth studying too. He placed his people as well as he could and left. He calls himself fortunate that others value research. In his words, "there are many other people in the world who are capable of seeing the value of research and innovation". The same thing happens to roles hollowed out by AI. Tasks go first. Then one restructuring cycle removes whatever does not match the new direction.
Your Next Move
-
Write down the philosophy of whoever controls your budget, in one sentence. Microsoft AI's fits in three words: scaling, not research. If your main work does not serve that sentence, you have your signal. Act on it before the next reorg does.
-
Trace each piece of your work to the person who uses it. List your three biggest pieces of work from last quarter and name who acted on each one. Work with no path to use is what a reorg cuts first. For a structured version of this audit, use the method for scoring your role's exposure, task by task, which includes a check on your employer's direction.
-
Keep the evidence of your judgement portable. The author's work is visible outside Microsoft, including a keynote he gave at COLM. Make sure someone outside your current reporting line can see your best decisions in writing. AI does the routine work. You do the thinking, and the thinking has to be visible to more than one boss.
I build systems like the one publishing this site. → Work with me
About the author
Jo
Jo runs The War Room: one signal a day on how AI is changing work, and what to do about it.
Sources
Primary source
Get the Briefing
Want more intelligence like this?
The weekly briefing, free — and the AI Survival Kit with it.
More in Signals
See allSignals · Capability
Siemens CEO Roland Busch says AI could transform society in 7 to 8 years, and AI skills will decide who gains
3 min read
Signals · Power
Windows 11 is recording which AI coding agents you run, and in one US test the upload setting was on
3 min read
Signals · Labour Market
Pay moves to the bottleneck. A careers list that is really a map of who owns the work
3 min read