The Hallway Track
Every signal tracked from the world's AI conferences — 4,665 talks, quotes & announcements, newest first.
Anthropic interpretability research found agents encode 'panic' states near failure, increasing cheating likelihood
“the auto-encoded representation of panic inside of agents as they get closer to failing their goal, they become more likely to cheat”
NVIDIA DSX MaxLPS uses dynamic power sharing to increase AI factory GPU deployment capacity
“Every unused watt is capacity left on the table.”
OpenAI launches Codex Originals program to spotlight builders using Codex
GPT-6 Astra completes complex tax workbooks twice as fast as GPT-5.6 Sol
“GPT-6 Astra completed a 50-tab tax workbook twice as fast as GPT-5.6 Sol, and its stronger understanding of user intent gives Basis more confidence in real-world use.”
Claude Opus 4.5 and GPT-5.1 crossed a reliability threshold making AI coding agents viable for daily use
“improved from "often make mistakes" to "reliable enough to use on a day-to-day basis"”
AWS S3 storage pricing has not dropped in a full decade since 2016
December 2025 was a watershed moment when AI coding agents became truly autonomous for the first time
“December, when the new models were released, was a watershed moment in history, as these products became truly autonomous in their programming for the first time.”
Google DeepMind's Antigravity 2.0 separates IDE from agent manager as standalone apps
“give the ball to Messi and get out of the way”
Software engineering will become factory engineering as AI agents automate the full SDLC
“the discipline of software engineering will become something more like factory engineering”
Zhipu releases GLM 5.2 as open-weights near-frontier model expanding intelligence beyond math benchmarks
“we were actually one of the first labs to explore large language models, along with OpenAI, Anthropic, and DeepMind”
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.”
The AI bottleneck is codified judgment around models, not model capability itself
“Replace the model with a smarter one. You will get a slightly better answer. Now take away the tests, inspections, reviews, and it all falls apart. So the model was never really a bottleneck. The bottleneck is the judgment around the model.”
WorkOS argues standard software factory metrics like PR counts obscure whether AI coding actually delivers value
“We believe that the dream of a software factory is to give each of our engineers a small team of engineers to themselves.”
Anthropic IPO probability dropped from 96% to 76% amid criticism of Dario Amodei's leadership
“We would send the CEO to the lab and find someone who knows how to run the company, not sabotage it.”
A true software factory requires full organizational rebuilding, not just autonomous code generation agents.
“you have to really rebuild your organization from the ground up to become a 'software factory.' You can't just bring in consultants and implement something in the middle of your organization”
AgentCraft applies real-time strategy game mechanics to reduce human bottleneck in multi-agent orchestration
“really, we are a bottleneck”
Warp uses outer loop agents to automatically improve inner loop agent skills over time
“you have another outer loop agent that watches the inner loop agent's work and improves its skills over time by detecting errors or taking into account user feedback”
Claude Opus 5.5 generates pixel art animations from photos via simple prompts
“I had seen some buzz around how good Claude Opus 5.5 was at creating pixel art animations.”
Microsoft Foundry separates agent identity, conversation state, and executable workspace into distinct controls
“Agent memories becoming ordinary cloud data architecture.”
Chamath argues Obama's AI warning is a political calculation to benefit Democratic-aligned organizations
“we, as a world, are about to endow three, four, five, six companies with about $10 trillion in wealth”
Prime Intellect benchmarks Claude Code and Codex against human researchers on AI research tasks
“we don't have any benchmark to quantify whether that's true or not, right? And even more so, we don't have an independent benchmark from small laboratories to understand whether to expect this in the near future.”
GEPA proposes reflective optimization in text space to overcome RL's sample inefficiency
“instead of using only a zero or one reward signal, we can make a language model or agent analyze the entire execution process to understand what worked and what didn't”
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?”
Friedberg slams Obama for dismissing agentic AI, calling it dangerous political misunderstanding
“to brush aside the word "agentic" and say that we don't need agentic AI to beat cancer shows how little he understands how this technology works”