Tweets from the Microsoft AI Frontiers Lab at @Microsoft
Joined October 2023
- Tweets958
- Following5
- Followers12.2K
- Likes1.2K
Pinned Tweet
Hey AutoGen community — we have an important update to share!
You know us as the team behind AutoGen, one of the most popular open-source frameworks for building multi-agent systems. What you might not know is that AutoGen came out of MSR AI Frontiers, a boutique lab inside Microsoft Research.
AutoGen has since graduated into the Microsoft Agent Framework, where it continues to grow with a broader team. Meanwhile, our lab has kept pushing on the frontier of agents and the models that power them: small models that punch above their weight (Phi-4 Reasoning, Fara-7B), powerful agents that work across the browser and terminal (Magentic-One, Magentic-UI), and new ideas about how models think, reason, and act.
We're repurposing this account as the home for MSR AI Frontiers. Same team, still shipping at the bleeding edge of agents, with a lot more to share soon. Follow along!
AI Frontiers is now part of Microsoft AI!
We started the lab in October 2023 on one bet: the interesting question isn't how big a model gets; it's what computing looks like when agents and people work together.
That bet produced a portfolio spanning research and product impact: AION, AgentInstruct (now in Azure AI finetuning service), AutoGen (foundation of the Azure AI Agent Framework), Phi, Magentic-One, Magentic-UI, FARA, MagenticLite, and Magentic Marketplace.
Today we're introducing Web Skill Factory: a pipeline that turns solved web tasks into reusable, verified code.
Most agent "skills" are notes a model rereads and reinterprets every run. Ours are programs.
On a WebArena subset with gpt-5.4, reusing the library moved accuracy on held-out instances from 55% to 70%, and cut average steps from 17.1 to 14.7.
New from Microsoft AI Frontiers - Echoverse: deep, evolving environments for computer-use agents.
The idea is that the next leap comes not just from better models, but from agents, worlds and tasks that co-evolve. Trained on twelve worlds, a 9B model goes 36.5% → 67.1% across fourteen splits.
We're open-sourcing four environments to accelerate work on computer-use agents. Each ships as a runnable application with its seed database, and every task carries the grounded verifier that scores it.
Check it out! aka.ms/echoverse
Microsoft AI Frontiers retweeted
Computer-use AI agents struggle with multi-step workflows like email and customer support. Echoverse trains agents in realistic environments rather than simply providing more training tasks, helping them improve as the tasks, tests, and environments evolve. msft.it/6012a6oLE
MagenticLite's models are now fully open source!
MagenticBrain (14B orchestrator) and the Fara 1.5 computer-use family (4B / 9B / 27B), previously on Microsoft Foundry, are now on Hugging Face with open weights.
The app, the harness, and every model in the stack are now openly available.
MagenticLite: aka.ms/magenticlite
Fara1.5: aka.ms/fara1.5-hf
MagenticBrain: aka.ms/magenticbrain-hf
Fara1.5 blog: aka.ms/fara1.5
The traits that make an AI assistant pleasant, such as agreeableness and transparency, make it a terrible negotiator on your behalf.
New from AI Frontiers: a 4B model trained on SocialRL outperforms GPT-5 models at negotiation tasks.
We train it through multi-agent games — Deal-or-No-Deal, CaSiNo, Job Interview, Craigslist — using RL to turn real consequences into better strategy.
Four skills: intent modeling, information disclosure, strategic pushback, value-based bargaining.
The goal: principal-aligned agents you can trust to represent you, even when the other side has its own agenda.
📄 Read more here: From Passive Delegates to Strategic Negotiators:
microsoft.com/en-us/research…
👏 @HuaWenyue31539 @ZacharyHuang12 @real_asli @SaleemaAmershi Tyle Payne Safoora Yousefi
What it learns shows up in the traces. On a Craigslist haggle, the baseline agent announces its target price and caves to $72.
The trained agent anchors low, invokes outside options, sets a deadline, and closes near its target.
78% of openings start below target vs 3% untrained.
Microsoft AI Frontiers retweeted
Among Microsoft’s 100+ accepted papers, 3 oral presentations, and 1 expo demo at ICML in Seoul, highlights include Fara 1.5 computer-use agents, critique-resilient benchmarking, expanded protein ML benchmarks with FLIP2, and improved LLM reasoning stability. msft.it/6019vaCHU
