The Hallway Track
Every signal tracked from the world's AI conferences — 5,359 talks, quotes & announcements, newest first.
Microsoft Discovery brings agentic, VS Code-based AI to scientific research, running specialized agents that hypothesize, simulate, and create new agents on the fly.
“If Microsoft Discovery does not find an agent for a task, it will create one on the fly”
Microsoft Foundry adds centralized tools, governance, evals, and an agent optimizer to make LangGraph agents enterprise-ready.
“Everyone is building agents, but that's not the hard part anymore. The hard part is integrating them into your business and governing them at scale.”
Microsoft launches Microsoft IQ to unify enterprise intelligence (Web IQ, Foundry IQ, Fabric IQ) as grounded context for agents.
“Agents are only as good as the context we give them.”
Axiom argues formal verification, not just coding ability, is the key bottleneck on the path to AGI.
“Verification to me is about scaling brilliance, compounding brilliance,”
Microsoft Build Day 2 hosted a live vibe-coding competition to build a cloud-based collaborative markdown editor with AI.
“Is there something like Google Docs, but for markdown? I need a cloud-based collaborative markdown editor, please.”
Fundamental's NEXUS, a deterministic foundation model for tabular data, is now on Amazon SageMaker JumpStart.
“It's pre-trained on billions of real-world prediction tasks across structured datasets, so it arrives already knowing how to find signal in your data.”
Listen Labs replaced manual trace review with LangSmith Engine, an agent that auto-surfaces bad agent behaviors.
“One big aha moment was really just my change in workflow and how much it sped up time for me.”
Microsoft is positioning itself as an ecosystem 'Frontier Intelligence Platform' rather than a single AI model or platform.
“a platform is defined by fundamentally its ability to create more value about the platform versus what's captured in the platform”
Google launched Managed Agents in the Gemini API: one API call spins up an autonomous agent in a sandboxed Linux environment.
“you make a single API call and you get an autonomous agent that can work on behalf of you and solve like problems creatively.”
Microsoft's Project Lobster is a cloud-based, always-on AI assistant with agency and memory that acts without user babysitting.
“I don't know that people fully grasp what it means to have a personal assistant that is has agency and judgment.”
GitHub Copilot's new desktop app is an agent-native experience for managing the full coding workflow, not just an editor or CLI.
“It's an agent native desktop experience for building things.”
Google's co-scientist, a fine-tuned Gemini with extra tools, helps generate hypotheses and analyze scientific data.
“Now you have a system called co-scientist that can invent new things.”
Frontier models now generate high-fidelity UI, enabling a shift from static to generative UI for MCP apps.
“And now it can write better front-end code than me.”
Azure Translator's Document Translation adds GA synchronous and batch image translation plus LLM-powered translation, announced at Microsoft Build.
VS Code team to share its internal AI adoption story in a Build day 2 breakout session.
“It's like literally like our team uses AI now. Okay, now what?”
AWS Deep Learning AMIs and Containers now support SOCI snapshotter for lazy-loading container images to cut cold start times.
“Standard Docker image pulls of 15–20 GB can take 4–6 minutes per instance, delaying training jobs and inference endpoints during scaling events.”
Google I/O attendees debate whether spending hours to automate a 2-minute task is a red or green flag.
“Green flag, honestly. Might You only have to do it once, but someone else might have to do it a bunch of times.”
Microsoft launches Foundry Managed Compute to run and customize open-source AI models on managed elastic GPU capacity.
“Managed Compute is built to lift that work off your team so the model, not the infrastructure, is what you spend your time on.”
Microsoft launches Agent Optimizer in Foundry Agent Service to automate optimizing hosted agents.
“the hard part isn't building the agent, it's getting the agent to behave correctly across all the scenarios it needs to handle.”
Microsoft Foundry Agent Service adds procedural memory so agents reliably reuse proven execution patterns in production.
“agents often know the right facts and still fail the task because they do not execute the right procedure”
Microsoft Foundry adds Toolboxes capabilities—Skills, Work IQ, Fabric IQ, Browser Automation, Tool Search, and Routines—to scale production agents.
“Tooling doesn’t break at a small scale—it breaks when teams move to production.”
Microsoft Foundry adds enterprise agent distribution: publish to M365 Copilot/Teams, autopilot agents, and Agent-to-Agent.
“The past year was about building agents. The next year is about putting them to work.”
Combining SFT and DPO on SageMaker AI improves a small model's tool-calling accuracy for agents.
“When an agent picks the wrong tool, formats parameters incorrectly, or breaks a workflow chain, task completion times grow, error rates rise, support costs increase, and user experiences degrade.”
NVIDIA announces Nemotron open model family for agent orchestration on Azure AI Foundry with Microsoft.
“DeepSeek published open models that were able to reason. And in terms of open models, this was a breakthrough.”