The Hallway Track
As of October 07, 2026
A living roundup of what's being said across the world's AI conferences — the latest signals, the most-active events, trending topics and notable quotes, rebuilt every week from the archive.
Google DeepMind releases new audio models including TTS, live translation, and Gemini Live upgrades
“Our new text-to-speech models are capable of doing really compelling voice personalization, which means that you can create a voice and guide it to have specific emotions, specific resonance, and even to incorporate things like pauses.”
Enterprises must redesign operating models around agents, not just add agents to existing workflows
“Where do I best insert agents working side by side with humans to ultimately optimize humans?”
Speculative decoding cuts LLM inference cost but only when GPU memory has spare capacity
TypeSafe's Jev model uses RLCD to deliver reliable machine-native intelligence for software automation
llm-mistral 0.16 adds reasoning model support including Mistral Large 4
Adecco used Salesforce AgentForce to save recruiters 35-40% time and place 20,000 more people
“AI had to happen with people and not to people.”
Google releases EmbeddingGemma 2 under Apache 2.0, enabling portable open-weight embeddings
“I don't think it makes sense to use a closed, proprietary, hosted-only model.”
Mistral releases 1 trillion parameter Large 4 model, reclaiming competitive position
“it's great to see Mistral put out a model that's back to being maybe about 6 months behind the frontier”
AI agent demand is causing 10-100x user growth that database infrastructure cannot keep pace with
“the number of daily active users has doubled, tripled, 10 or 100 times. This is largely driven by the demand for AI agents. The infrastructure simply cannot keep up with this development.”
Google DeepMind releases EmbeddingGemma 2, an open lightweight multimodal embedding model
AWS AgentCore and OpenClaw enable persistent-memory AI assistants via single CloudFormation deployment
“AgentCore memory, a capability of Amazon Bedrock AgentCore, turns disposable chats into durable knowledge.”
Simon Willison integrates Parseable observability platform with Datasette OpenTelemetry traces using Codex.
NVIDIA DOCA GPUNetIO enables GPUs to control networking directly, eliminating CPU bottlenecks
“GPU applications increasingly need networking and data movement to behave like first-class GPU-controlled operations rather than host-driven services.”
GitHub Copilot agent converts meeting transcripts into project backlogs and risk assessments automatically
“it gives you a competitive edge”
Microsoft IQ gives enterprise agents grounded context via four intelligence sources for meaningful outcomes
“Models give agents intelligence but contexts give that intelligence meaning”
“Our new text-to-speech models are capable of doing really compelling voice personalization, which means that you can create a voice and guide it to have specific emotions, specific resonance, and even to incorporate things like pauses.”— Google Developers (Google I/O)
“Where do I best insert agents working side by side with humans to ultimately optimize humans?”— Salesforce
“AI had to happen with people and not to people.”— Salesforce
“I don't think it makes sense to use a closed, proprietary, hosted-only model.”— Simon Willison
“it's great to see Mistral put out a model that's back to being maybe about 6 months behind the frontier”— Simon Willison
“the number of daily active users has doubled, tripled, 10 or 100 times. This is largely driven by the demand for AI agents. The infrastructure simply cannot keep up with this development.”— AI Engineer