AI This Week: Agents That Remember, AI in the Lab, Votes on Data Centers

A newer Claude model wrote nine memory files and earned four Pokémon badges, where an older one wrote 413 files and earned one. Uber says AI agents booked 4 rides among about 320 million trips in one week, and voters chose data center limits in 44 of 45 counted local measures.
artificial-intelligence
entrepreneurship
Author

Kabui, Charles

Published

2026-11-09

Keywords

ai-weekly, agent-memory, wet-lab-benchmark, agent-orders, data-center-votes

Download as PDF

Anthropic published two studies of AI agents, one on memory and one on money. In Pokémon Red, about 10,000 steps in, Claude Sonnet 3.7 had written 413 notes to itself and earned one badge. A newer model, Fable 5, wrote nine and earned four (one run each). An agent that rereads its past work and rewrites its notes, which Anthropic calls dreaming, scored 61% on company files with search plus a wiki it built, against 53% with search alone. In a replay of real EVE Online market offers, not a live market, three Claude models profited in all nine two-week windows, growing their starting capital, worth only about $16 to $31, by 16%, 22% and 43% on average.

OpenAI’s BioBasicsBench gives AI agents 100 routine wet-lab tasks that use raw data such as gel images. At its highest reasoning setting, GPT-6 Astra earned 66.7% of the available credit, against 43.8% for GPT-5.6 Sol, and all seven models tested were OpenAI’s own. In a preprint not yet peer reviewed, Anthropic’s biology team says Claude helped it purify a vault, a hollow protein barrel found in cells, in about 90 minutes instead of three days, and reach a first 3D model 34 hours after image processing began.

Agent shopping is still small. Uber’s finance chief said AI agents placed about 500 Uber Eats orders and 4 rides in one week, when Uber ran around 320 million trips. Redbud’s unofficial tracker shows voters chose limits on data centers in 44 of 45 local measures with complete counts. Democrats won governor races in Texas, Georgia, Ohio and Arizona, and Ars Technica cites Bloomberg’s estimate that 13 gigawatts of Texas data center capacity is now exposed to delays.

Read More: OpenAI Dreaming V3: ChatGPT Now Builds and Updates Its Own Memory of You and DoorDash Counts Several Hundred AI Agent Orders as Websites Limit Access

Sources:

Disclaimer: For information only. Accuracy or completeness not guaranteed. Illegal use prohibited. Not professional advice or solicitation. Read more: /terms-of-service

Reuse

GNU GENERAL PUBLIC LICENSE v3.0(View License)

Citation

BibTeX citation:
@misc{kabui2026,
  author = {{Kabui, Charles}},
  title = {AI {This} {Week:} {Agents} {That} {Remember,} {AI} in the
    {Lab,} {Votes} on {Data} {Centers}},
  date = {2026-11-09},
  url = {https://toknow.ai/posts/ai-weekly-agent-memory-eve-trading-biobasicsbench-uber-data-center-votes/},
  langid = {en-GB}
}
For attribution, please cite this work as:
Kabui, Charles. 2026. “AI This Week: Agents That Remember, AI in the Lab, Votes on Data Centers.” https://toknow.ai/posts/ai-weekly-agent-memory-eve-trading-biobasicsbench-uber-data-center-votes/.