The Hallway Track
Every signal tracked from the world's AI conferences — 4,665 talks, quotes & announcements, newest first.
Stanford HAI proposes first-of-its-kind governance agenda for AI world models beyond language.
“we gradually move from AI that operates primarily through language or a combination of images, audio, and video, to a new class of AI that can maintain a holistic view of the physical environment over time and predict how it might change in response to actions”
Trustpilot deployed Salesforce Agentforce in 9 weeks, cutting case handling time 45%
“If someone came to me and asked me should they move towards being an agentic enterprise and add digital labor to their workforce, my answer to them would be, why haven't you started yet?”
LangChain advocates fine-tuned evaluation judges over expensive frontier LLM-as-judges for production agent monitoring at scale
“Using frontier judges, LLM-as-judges, online evaluators, obviously is very effective, but also very expensive at scale.”
Airbyte redesigned its CLI and MCP as agent-first interfaces built on OpenAPI specs
SMB tiling company achieved 25x admin efficiency gain deploying five Salesforce AI agents in 15 months
“It's been 15 months of working pretty hard to get there. We got the five agents live. It was a global first how we put it all together.”
Gemini enabled real-time remote robot arm teleoperation across 1700 miles via browser
“I worked with Gemini to create a 3D visualizer of the robot.”
HelloFresh uses Databricks Genie agents to optimize 100% of Meta ad spend
“What used to take 40–50 hours a week, our colleagues now do in one or two hours.”
AWS Bedrock now supports OpenAI models in managed agents while keeping enterprise data within AWS.
“AI agents are already moving into workflows that require them to reason over enterprise knowledge, operate across long-horizon tasks, and act in systems where security, accuracy, and governance matter.”
Postman runs AI Agent Mode for 40 million developers on Amazon Bedrock at production scale.
“Building an AI agent for a demo and operating one for 40 million developers are different engineering problems.”
OpenAI's Codex application server is open-source and powers all Codex interfaces via a shared protocol.
“It is important to remember that both the Codex shell and the application server itself are open source.”
No content was provided to extract a signal from.
AI may break public-key encryption faster than humans can replace cryptographic standards
“The problem here is that the speed of AI producing surprises, and the speed of human beings replacing standards (even with the very best AI assistance) are just orders of magnitude different.”
LangChain's Managed Deep Agents bundles harness and infrastructure to simplify production agent deployment
“a task that used to take them weeks, to spin up a new agent for a researcher, now just takes a few hours by putting this all into the Deep Agents harness and deploying it onto LangSmith”
FIDE partnered with Salesforce at Dreamforce to create AI-powered immersive chess broadcasts.
“AI and Salesforce, you're going to create immersive ways to broadcast chess.”
Microsoft launches open-source QDK for Chemistry bridging quantum algorithms and classical HPC workflows
“quantum computers act as a critical accelerator, which treats a specific part of the problem to which it is best suited, rather than as a general purpose replacement for compute in complex scientific workflows”
Replace repeated agent prompting with codified rules via tests, typing, and linters
“you should stop prompting, you should start codifying the actual rules that the agent should follow”
Databricks proposes governed data sharing to solve defense supply chain visibility gaps.
Simon Willison built a working Django feature entirely by voice using GPT-6 Astra High
“I had a pretty good idea of what I wanted to build, and it's a simple enough Django feature that I was certain the model (in this case GPT-6 Astra High) would be able to do it.”
Sophos uses OpenAI Daybreak to cut cyber-threat investigation time by 96%
Asana achieved 76x cost reduction and 5x speed gain using GPT-6 Astra in Codex for browser agents
Databricks Excel add-in supports cascaded filters, saved queries, and custom SQL functions
AI-generated PRs are overwhelming code review pipelines, creating a quality validation crisis for engineering teams.
“Thanks to AI, more PRs are being generated than before, and they are getting larger, each of which may contain defects.”
ttok 1.0 ships with GPT-5/6 tokenizer default after community experiment confirms GPT-6 tokenizer unchanged
“All seven GPT models (5.5, 5.6 Sol/Terra/Luna, 6 Astra/Sol/Luna) report 44,794 tokens and match each other on every one of the 31 fixtures. GPT-6 introduces no input-count change on this corpus.”
Static AI benchmarks create a false reliability signal that collapses under real production usage
“I like to call this the benchmark illusion.”