John Langford wrote on the blog hunch.net on November 6 that Microsoft’s AI Frontiers research group is over (hunch.net; our piece). A Microsoft service that grew out of contextual-bandit research Langford co-authored reached its end first: Microsoft Learn says that from September 20, 2023, customers could not create new Azure AI Personalizer resources, and that “The Personalizer service is retiring 25 August 2026” (Microsoft Learn).
By our arithmetic, that retirement date came about 10.2 years after the June 2016 paper describing Microsoft’s Decision Service, on which Langford is a co-author. The open-source library the research used, Vowpal Wabbit, published its latest release on September 27, 2026, according to GitHub.
Business Pill · Distribution Advantage
An evergreen Business Engineer explainer on distribution advantage. It teaches the general idea only and does not discuss Microsoft, Personalizer or Vowpal Wabbit.
The key insight:Microsoft Learn says no new Azure AI Personalizer resources could be created from September 20, 2023, and gives the service a retirement date of August 25, 2026; Microsoft Research says Personalizer was released from its Custom Decision Service pilot, and the 2016 Decision Service paper lists John Langford as a co-author.
The paper, “Making Contextual Decisions with Low Technical Debt”, was posted to arXiv on June 13, 2016. Its listed authors include Alekh Agarwal, Sarah Bird, John Langford, Siddhartha Sen and Alex Slivkins (arXiv).
In a 2017 Microsoft Research blog post, Agarwal and Langford wrote: “Clicks on Microsoft’s news website MSN.com increased 26 percent when a machine-learning system based on contextual-bandit algorithms was deployed in January 2016 to personalize news articles for individual users.” The same post said the technology was available as a Microsoft Cognitive Service called the Custom Decision Service, and that the core contextual-bandit algorithms were available in Vowpal Wabbit (Microsoft Research).
A Microsoft Research project page says of the Custom Decision Service: “The research pilot was successful and released as Azure Cognitive Services Personalizer Preview” (Microsoft Research).
From the June 2016 Decision Service paper to Personalizer’s August 25, 2026 retirement date is about 10.2 years, by our arithmetic; Vowpal Wabbit published release 9.11.9 on September 27, 2026. Data: arXiv, Microsoft Research, Azure blog, Microsoft Learn, GitHub.
Personalizer at Build 2019
Microsoft introduced Personalizer at its Build conference, in an Azure blog post dated May 4, 2019, as reinforcement learning for the enterprise. The post said: “Xbox saw a 40 percent lift in engagement by using Personalizer to display content to users that will most likely interest them.” (Azure blog)
The Retirement
The Microsoft Learn overview of Personalizer carries the notice: “Starting 20 September 2023, you can’t create new Personalizer resources. The Personalizer service is retiring 25 August 2026.” It recommends migrating to the open-source microsoft/learning-loop project (Microsoft Learn).
On GitHub, microsoft/learning-loop shows 13 stars, with its most recent push on November 27, 2024, according to the GitHub API on November 7, 2026 (GitHub).
What Is Still Shipping
The Vowpal Wabbit repository was created on GitHub on July 31, 2009, and shows 8,734 stars and 1,919 forks, according to the GitHub API on November 7. Its latest release, 9.11.9, was published on September 27, 2026 (GitHub).
What Langford Wrote
Langford’s post does not mention Personalizer or Vowpal Wabbit. Explaining why Langford did not return to Microsoft Research, the post says: “I am just not comfortable being at a company where the research I do has no path to real use.” The post also says: “So AI Frontiers is over.” Our earlier analysis counted three Big Tech AI research groups reorganized since 2023 (our piece).
Our Numbers
By our arithmetic, June 13, 2016 to August 25, 2026 is about 10.2 years. Personalizer accepted new resources for about 4.4 years, from the May 4, 2019 launch post to September 20, 2023, also our arithmetic.
The Structural Read
As we read the dates, the service had a shorter public life than the research behind it: Personalizer accepted new resources for about 4.4 years, from the May 4, 2019 launch post to September 20, 2023, by our arithmetic.
As we read the GitHub record, the open-source library outlasted the managed service: Vowpal Wabbit published release 9.11.9 on September 27, 2026, about a month after Personalizer’s retirement date.
As we read Langford’s post, it is about research reaching use; it does not mention Personalizer or Vowpal Wabbit, and nothing we read links AI Frontiers to Personalizer.
John Langford, hunch.net, November 6, 2026
“I am just not comfortable being at a company where the research I do has no path to real use.”
Three Implications
FOR BUILDERS ON PERSONALIZERMicrosoft Learn recommends migrating to the open-source microsoft/learning-loop project.
FOR RESEARCH GROUPSMicrosoft Research’s 2017 blog reported a 26 percent click increase on MSN.com from a contextual-bandit system deployed in January 2016.
FOR PRODUCT TEAMS The Azure blog of May 4, 2019 reported a 40 percent engagement lift at Xbox from Personalizer.
The analysis says: “The transformer is a commodity now; what changes between paradigms is where intelligence accumulates and who can capture it.”
As we read it, Personalizer was a product of learning from user interactions, built before the current model paradigm; the analysis does not discuss Personalizer, Vowpal Wabbit or Microsoft Research.
What Is Not Established
The Microsoft Learn page gives no reason for the retirement. The Decision Service paper and the 2017 blog predate AI Frontiers; nothing we read says AI Frontiers worked on Personalizer. The sources do not give Personalizer’s revenue or customer count. This piece makes no forecast.
This Week In The AI Supercycle
The Great Convergence of Capital, Power, and Geopolitics
This week’s issue: for two years the AI build-out was priced as a supply story, and last week as a credit story. This week it was priced as a permission story, and the answer arrived as a cost, not a ban.
Microsoft’s Personalizer, which Microsoft Research says grew out of its Custom Decision Service pilot, stopped taking new resources on September 20, 2023 and had a retirement date of August 25, 2026, about 10.2 years after the 2016 Decision Service paper Langford co-authored, by our arithmetic. Vowpal Wabbit published a release on September 27, 2026.
A note on sourcing. Dates, quotes and figures are from Microsoft Learn, arXiv, the Microsoft Research blog and project page, the Azure blog of May 4, 2019, John Langford’s November 6 post on hunch.net, and the GitHub API as read on November 7, 2026. Year counts are our arithmetic. This is not investment advice.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
