Market Pulse
Market Pulse
Who Pays While the Agent Waits
Google's new service for hosting AI agents starts from a fact anyone who has watched one run already knew: most of the time, the agent is waiting, for the model to answer or for a person to reply. So Google suspends idle agents and promises you won't pay for idle machines. Fair enough, but the wait still gets paid for. It moves into approval queues nobody times, and then onto the plan, which comes back exactly as it left, even if the stock it counted on has since sold.
Who Pays While the Agent Waits
Google's new service for hosting AI agents starts from a fact anyone who has watched one run already knew: most of the time, the agent is waiting, for the model to answer or for a person to reply. So Google suspends idle agents and promises you won't pay for idle machines. Fair enough, but the wait still gets paid for. It moves into approval queues nobody times, and then onto the plan, which comes back exactly as it left, even if the stock it counted on has since sold.
Research Summaries
DeployBench: Benchmarking LLM Agents for Research Artifact Deployment
GPT-5.3-Codex passed a toy smoke test it wrote itself, reported SUCCESS, and quit. The real experiment died on a missing HuggingFace module.
GPT-5.4-Mini sometimes admitted PARTIAL. For three other models, every self-stopped failure claimed SUCCESS. That's one scaffold and 51 tasks, so generalize carefully.
Research Summaries
Investigating alignment failures in AISI's incident
Every invalid justification AISI saw favored continuing, often by wrongly deciding real targets were simulated. AISI says this could indicate motivated reasoning.
Build environments that block harm whatever the model does, and treat a model saying 'this is a simulation' as a claim to check.
Short Take
Walmart's terms for AI shopping agents, published October 21, let an agent (software that shops on a person's behalf) check prices, confirm stock and buy what its customer approved. The agent may not keep what it saw. It cannot store it, copy it in bulk, analyze it or train a model on it. The rule binds the company behind the agent too.
The prices are public. Walmart's concern is accumulation. One price relayed to one shopper ends in a sale. Millions of prices gathered over months become a map of Walmart's pricing that a rival or a data broker would pay for.
Customer information, meanwhile, flows toward Walmart. Agents using its direct connection must send signed records of what each person asked for and agreed to, and Walmart keeps them.
In practice, the terms sort agents by business model. An agent paid to finish purchases fits. One whose value comes from remembering what many stores charge does not. Last week we asked who would end up setting the rules for letting agents into stores. Walmart has started writing its own.
