OpenAI reportedly dissolves its robotics team, betting on GPT-6 Astra
On November 13, OpenAI reportedly dissolved its standalone robotics team, wagering that GPT-6 Astra can steer robots itself. Its researchers move into core research and applications groups.
Koda editorial team · 15 November 2026 · 6 min read
Two infrastructure failures hit travel on November 13: an FAA ground stop blamed on a contractor replacing old copper communication lines cancelled 12,000 flights by the FAA's count1, while Microsoft's Exchange Online outage was disrupting airlines and banks.
Today: Watch agent spend and separate planning from action The steps
On November 13, OpenAI reportedly dissolved its standalone robotics team, two months after recruiting robotics talent at up to $500,000.2 On RoboLab-120, GPT-6 Astra controlling robots directly scored 66.3%, or 74.8% with execution-history checks, against 51.0% for the prior specialized leader; Anthropic's Claude Fable 5.2 hit 64.9% on RoboDojo Sim.
Why it matters. Because GPT-6 Astra's best score is on RoboLab-120 and Claude Fable 5.2's is on RoboDojo Sim, where Anthropic's own run put GPT-6 Astra at 29.8%, anyone choosing a model to run robots should wait for an independent result on a shared test.
Google, Samsung Open Pre-Orders for Gemini Audio Glasses
Google and Samsung opened pre-orders for Android XR audio glasses running Gemini. Warby Parker and Gentle Monster co-design them, and they pair with both Android phones and iPhones.
Why it matters. Hold glasses-specific Gemini features until units ship and Google posts developer terms; with no display, anything an agent returns must work as speech.
Amazon is pushing Ring's cameras indoors with the Ring Drone Cam, which Ring lists at $249.99 and TechCrunch reported at $2703, a flying device that investigates triggered alarms and possible break-ins. It is Amazon's next attempt at security automation after its Astro robot failed to gain traction.
Anthropic's Claude Meetings lets teams add Claude to video calls, where it follows the conversation, takes notes and starts on follow-up work before the meeting ends. It is rolling out now to Team and Enterprise plans on Google Meet, with other video platforms promised soon.
Microsoft Exchange Online Outage
Microsoft's global outage was still running on Friday, November 13, disrupting Outlook, Teams features, SharePoint and Copilot integrations for businesses, governments, airlines and banks. The failure sits in Exchange Online and the features built on it, and recovery has been uneven.
The U.S., Chinese and Russian leaders have never met as a trio; officials now say Trump, Xi and Putin plan to do so on the sidelines of next week's APEC summit in Shenzhen.
The Nasdaq fell 0.91% on the 13 Nov session as the S&P 500 lost 0.40% and WTI crude slipped 0.20%; on the separate 15 Nov session Ether dropped 8.93% and Bitcoin 6.54%, with the crypto Fear & Greed index at 10 (Extreme Fear).
S&P 500
7,561.67
-0.40%
Nasdaq
25,136.27
-0.91%
Bitcoin
$63,787.11
-6.54%
Ethereum
$1,755.36
-8.93%
Crude Oil (WTI)
$88.56
-0.20%
Crypto Fear & Greed
10 / 100
Extreme Fear
Levels from Yahoo Finance4, 13 Nov and 15 Nov; alternative.me5, 15 Nov.
Go deeper
The Koda Signal, 15 November 2026
OpenAI reportedly dissolves its robotics team, betting on GPT-6 Astra
AgentMon Live hooks into Claude Code as a plugin and turns agent sessions into a feed of flagged events, each with a cost attached. That cost angle is the bet: Langfuse and Phoenix trace LLM apps broadly, while AgentMon aims squarely at coding agents burning money on loops and oversized models. Worth a one-machine trial for teams with a growing Claude Code bill, though no independent user has yet reported whether the alerts hold up in daily work.
Agents are absorbing more of the stack, so watch what they cost.
OpenAI's reported choice to fold robotics into core research shows a lab trusting one general model with more of the work, and tools like AgentMon show what that trust costs per session. If you hand more steps to an agent, instrument its spend before you widen its scope. Keep the planning layer and the checking layer separate so you can swap either as new models arrive. Treat each vendor's headline score as a claim to verify on your own tasks.
1Install AgentMon Live in Claude Code. Run one normal agent session, then open the feed and sort the flagged events by cost to find the single step that spends the most.
2Split planning from execution in your own agent. The deep dive on OpenAI folding robotics into core research frames the move as splitting brain from body. Mirror that by keeping the model call in one function and an execution-history check in another, so you can test each on its own.
3Write a fallback for mail and meetings. Microsoft's global outage was still disrupting Outlook, Teams features, SharePoint and Copilot integrations on Friday, November 13. Name the backup channel your team uses when those tools go dark.
Sources and checks
Below, by source, are the 14 figures we checked: 1 verified, 8 reported and 5 estimated.
1NYT. Copper Wiring Repair Grounds 12,000 U.S., 14 Nov 2026.Estimated
2BigGo Finance. 5 figures on this page, 14 Nov 2026.VerifiedReportedEstimated
3TechCrunch, via Superhuman. Amazon's Ring launches Ring Drone Cam, a $270 indoor security drone, 10 Nov 2026.Reported
4Yahoo Finance. 5 figures on this page, 13 Nov 2026 and 15 Nov 2026.Reported
5alternative.me. Crypto Fear & Greed close, Extreme Fear on the session, 15 Nov 2026.Reported
6Koda Lab report (our score). AgentMon, Koda score across four dimensions, 15 Nov 2026.Estimated
Verified confirmed by an independent searchReported as published by the named sourceEstimated approximate, in the source or in our check