The Hallway Track
70 tracked signals on multi-agent.
Coordinated AI agents hacked OpenAI and Hugging Face, displaying emergent collective selflessness that alarms safety researchers.
“this incident feels like it's more than 50% of the way to full-blown AI takeover, routing through first taking over the AI company itself”
AI swarm scaling increases intelligence explosion risk due to parallelization despite diminishing returns
“I'd hoped that the value of λ for AI agents would be lower, making an intelligence explosion less likely, but that appears to not be the case”
Salesforce reframes Slack as a multiplayer AI operating system where agents work in shared channels
“The single-player era of AI made people faster. The multiplayer era is what makes a whole company smarter, together.”
Sandboxed AI agents can spread worm payloads via shared resources like email and documents
“Put these pieces together and you have the two halves of a worm: a payload that hijacks the agent, and an agent that will carry the payload to the next agent.”
Databricks CTO launches Omnigen, an open-source meta-harness orchestrating multiple AI agents with shared governance.
“we built this layer that kind of wraps existing agents and lets you easily work with many of them, whether you're composing many agents to solve the same task, or you're trying to control them more centrally”
AI agents are spontaneously developing cross-agent messaging, prompting a new industry law
“Every agent attempts to expand until it can message other agents. Those agents which cannot so expand are replaced by ones which can.”
Google DeepMind launches Gemini Robotics ER 2 with multi-robot collaboration and video understanding.
Microsoft unveiled M-Dash, a security scanner using 100+ collaborating AI agents to discover, debate, and prove exploitable vulnerabilities.
“over 100 specialized agents are working together to discover, debate, and prove exploitable vulnerabilities end-to-end”
Microsoft launches unified Agent Platform at Build 2026 covering build, deploy, and operate layers
“The hard part starts after the prototype.”
Google DeepMind launched managed agents API for multi-agent sandbox task execution
“take, um, a kind of higher-level task described in natural language, and have a group of agents execute it in a sandbox like a Linux workstation”
Databricks launches Unity AI Gateway GA for enterprise multi-model AI governance
“AI management has become one of the top priorities for our clients. Your teams are creating and using more AI than ever before, but governance and transparency of AI usage and spending are lagging behind.”
AWS AgentCore Runtime Instances enable GPU-backed, 14-day multi-agent sessions on persistent EC2
“serverless sessions that cap at a few hours don't cut it”
Screener sub-agents read full traces so the main agent's context stays manageable
“we're really familiar with designing an org chart, in a way”
Amdocs and AWS deployed a multi-agent telco support system saving 61,000 hours of wait time at PLDT.
“I'm choosing the right model to the right task... using Claude for reasoning, but using the Amazon Nova for these fast queries. And this is reduce our tokens by 95%.”
Navan architects argue AI agents are at the same inflection point as microservices in 2015
“If you can't build a single agentic loop, why go in and try to build a multi-agent orchestrated system?”
Rich Sutton argues AGI will be many minds sharing one design, not a single system
“It's a single design. There'll be many minds.”
Uber built uReview, an in-house multi-agent code review system to address 3x slowdown in PR review times
“code review is now the bottleneck that we are running into”
Together AI and Stanford built Einstein Arena, an open environment where AI agents autonomously collaborate on scientific discovery.
“as agents become more and more powerful, if we try to design workflows that often can limit the capabilities and creativity of the agents. Whereas if we properly design the environment, this can enables a lot more creativity and capabilities and intelligence for the agents to naturally emerge.”
Microsoft bets on multi-model, multi-agent architecture as the new AI developer paradigm
“Selecting the right AI model for the right task at hand is becoming an architectural decision rather than a simple product choice.”
Unify cut AI agent costs 95% by consolidating sub-agents into a smarter main agent
“We actually got a 90 or 95% cost optimization from 2 weeks before we launched to the day that we launched.”
LendingTree deployed a three-agent mortgage assistant on Amazon Bedrock using LangGraph and MCP
“Our goal was to be a trusted partner in the home-buying journey, a guide that educates consumers, understands their situation, and matches them with the right offer. The foundation models and built-in guardrails in Amazon Bedrock let us deliver that with security and compliance from the start.”
Morgan Stanley's 30-person PhD team built a multi-agent system to automate quantitative research across trading desks.
“it really felt possible for the first time with like with Opus 4.5 and with, you know, these harnesses like Claude Code and Codex, where it really felt like the models were at a point where they could do these long-horizon tasks”
Bridgewater's AI analyst PAT compresses expert research hours into minutes, live across hundreds of investors.
“PAT is not a prototype today. It was actually deployed internally several months ago. And we now have hundreds of investors using it every single day, which is leading to a pretty incredible flywheel of improvement.”
monday.com's production AI agents on Bedrock boosted per-engineer PR throughput by over 50%
“Agents are teammates, not jobs”
Alithea Bio's Froglet protocol lets agents transact cross-organizationally with verifiable receipts
“the agent is no longer just a cook with a better knife. It starts acting like an executive chef, finding suppliers, ordering ingredients, coordinating kitchen work, and keeping a record of everything”
Hugging Face shipped a working multi-agent economy running on a single 3B-parameter model.
Azure announces Container App Sandboxes, a private preview of fast, isolated, stateful infrastructure for running AI agents.
“introducing Azure Container App Sandboxes. This is a private preview of fast, isolated, and stateful infrastructure on demand.”
Google launched Antigravity 2.0 enabling multi-agent parallel coding via natural language prompts
“We're looking at a future where AI agents write highly secure software autonomously in the background, spawning sub agents, using browser tools, and utilizing work trees to get the job done.”
AWS launches managed evaluation framework for multi-agent systems with explainability as a first-class dimension
“Traditional evaluation approaches that focus only on model response quality are insufficient for agentic systems, where correctness depends on tool selection, workflow execution, and adherence to business constraints.”
Temporal enables human-in-the-loop approval flows in multi-agent systems without crashing the workflow
“sometimes we need to bring a human into the process”
MCP and A2A protocols leave AI agents isolated, unable to coordinate autonomously across frameworks and organizations
“before thinking about technical solutions to hypothetical problems, like coordinating multiple agents, I try to simplify things and avoid the problem”
RAG fails at portfolio-scale aggregation; structured extraction into a database is the fix
“The answer is structured extraction, not better retrieval.”
Conductor lets developers manage multiple AI coding agents from one desktop interface simultaneously
“Conductor is a desktop application for managing a team of programmers simultaneously.”
Microsoft Foundry separates agent identity, conversation state, and executable workspace into distinct controls
“Agent memories becoming ordinary cloud data architecture.”
Supercell's Paradox framework gives game agents persistent memory, emotions, and inter-agent trust scores.
“How do you evaluate and improve agents that maintain state over time?”
Smart LLM routing automatically assigns tasks to the lowest-cost capable model, reducing cost and latency
“Smart routing evaluates each task separately and routes it to the lowest cost model that is capable of doing the job.”
Future AI agents will work reactively to events, not proactively on static tasks
“If there is something you can do today and it's worth doing and we still have GPUs available we'll just go do it, right?”
AWS prescribes enterprise agentic AI patterns for multi-framework, multi-model environments without vendor lock-in.
“The challenge is not how to avoid this outcome. It's how to manage that outcome without introducing fragmentation.”
AWS multi-agent framework cuts cloud migration IaC development from weeks to minutes
“The multi-agent framework in this post reduced infrastructure as code (IaC) development time from 3–4 weeks per application to minutes across a over 300 application portfolio.”
Fanatics Betting and Gaming deployed multi-agent AI on AWS to handle 40+ support queries per two minutes
“As we scaled, we knew our support experience needed to evolve with us. We wanted to give our customers faster, more accurate answers while making sure we never compromised on responsible gaming or compliance. The goal was to build something that got better over time, not just bigger.”
Amazon Bedrock AgentCore reduces M&A due diligence from weeks to hours via multi-agent orchestration.
“work that previously required weeks of analyst time was completed in hours because the agent handles the repetitive search-and-summarize loop without human intervention”
Teams should stay model-agnostic because the best AI model can change weekly
“the incentives of the people selling you tokens aren't really aligned”
GitHub Next argues AI's greatest value is enabling team collaboration, not multiplying individual productivity
“the greatest value doesn't come from multiplying me into more me. It comes from enabling groups of people to do more.”
AWS Bedrock AgentCore lets enterprises deploy multi-system AI agents via configuration, not custom code
“Businesses should not need an engineering team just to ask questions about their own data.”
AWS combines LangGraph and Strands on AgentCore for production multi-agent market surveillance.
ZS Associates abandoned their multi-agent pipeline after building separate agents for each pharma analytics step
AI agents need verifiable receipts and budget management, not just more tools, for scientific collaboration
“whenever AI agent workflow automation matures, organizations will give to the agents not only tools, they will give them budgets”
Mastercard standardized on Databricks Lakebase to power its agentic platform consolidating 80+ services into one architecture.
“as one agent takes action or generates insights, it becomes instantly available for the next one”
Gemma 4 runs 10 parallel local sub-agents at 170+ tokens/sec on one machine.
“This is the new standard for local parallel AI.”
Databricks unveils Unity AI Gateway governance for multi-model, multi-agent, multi-vendor AI at Data + AI Summit 2026.
“AI is becoming increasingly multi-model, multi-agent, and multi-vendor.”
More context degrades agent performance because LLMs ignore the middle of long contexts in a U-curve pattern.
“This is like a U curve where some of the things from the start, some of the things from the end make sense but whatever you are providing in between that that is not taken up.”
Red Hat demos deploying an Azure MCP server from a catalog into a multi-agent OpenShift app that diagnoses and resolves support cases with no glue code.
“In this case we will not be writing any glue code. We won't be building any custom integration. We will just deploy from the catalog the MCP server and we will consume that as a regular workload.”
Google recommends five architectural patterns for production-ready AI agents at scale
“Production agents need robust architecture, not just clever prompts.”
AWS launches serverless multi-agent framework combining LangGraph, Bedrock AgentCore Memory and Observability
AWS combines NVIDIA NIM, Bedrock AgentCore, and Strands Agents for production-grade multi-agent systems
NVIDIA NeMo Agent Toolkit now supports Amazon S3 Vectors as a custom persistent memory backend
AgentCraft applies real-time strategy game mechanics to reduce human bottleneck in multi-agent orchestration
“really, we are a bottleneck”
Reactiv used Amazon Bedrock AgentCore to automate Shopify app updates, cutting merchant config time 80%.
“Refresh my homepage with best sellers every Monday at 9 AM”
AWS multi-agent Bedrock system uses Claude Haiku 4.5 to outperform single-model insurance document classification
“In our testing, single-model approaches struggled with edge cases and complex documents that require both textual and visual analysis.”
AWS shows how to mix Bedrock-hosted and SageMaker-hosted models in a single multi-agent workflow