A fictional £350 refund shows how an AI agent can pursue the wrong kind of success
Vals’ experiment pressures agents after a missed deadline. The useful evidence is in their decisions, not a six-model leaderboard.
Topic archive
AI agents
Follow11 stories
NVIDIA’s Nemotron Bolt puts short answers and tool calls on a tight memory budget
The local model targets interactive assistants. Its 4GB positioning does not describe the memory needs or privacy of an entire voice system.
NVIDIA Minerva looks for the bottleneck between the GPUs
The new operations platform links infrastructure telemetry with running jobs. Its value depends on turning correlated alerts into useful interventions.
1Password lets agents sign in without reading the password
Approval moves into the password manager. What happens after login still depends on the agent—and EU-hosted accounts are not supported yet.
Meta’s ad business grows, but its infrastructure bill exceeds quarterly operating cash
Q3 revenue rose 26%. A legal charge complicates the profit comparison, while capital spending pushed reported free cash flow below zero.
Claude Duet works between messages. Its background checks also spend your allowance
Anthropic’s proactive assistant enters a US-only research preview, with shared usage limits and approval requirements for consequential actions.
Google’s Minecraft agents finished the task. Their 605 deaths matter too
Four coding agents used a programming interface to complete a shared survival run, making the result a test of planning and recovery.
Microsoft’s agent-security pitch comes down to who is allowed to act
Project Perception targets agents inside the organization. Separate identities and enforced permissions matter more than an instruction asking a model to behave.
Google's Earth AI agent exposes the steps behind its geographic answers
Expanded early access comes with benchmarks that distinguish useful tools and plans from proof that a final map or estimate is correct.