Je connais des gens très optimistes qui continuent de penser que ça va aller, et qu'on va s'en sortir. Alors objectivement non, ça ne va pas aller et non on ne va pas s'en sortir.
SAFETY
-
Agents in Production: Unpredictability and Monitoring Challenges
By
–
New Conceptual Guide: You don’t know what your agent will do until it’s in production With traditional software, you ship with reasonable confidence. Test coverage handles most paths. Monitoring catches errors, latency, and query issues. When something breaks, you read the
-
Multi-agent systems need shared definitions prevent miscommunication
By
–
Shared language =/= shared meaning. And that can turn multi-agent systems into a game of telephone without any human in the loop being the wiser. Our @MicrosoftAI pre-print tests a solve: if agents don't agree on a definition, they can't use the term. The results: disagreement
-
Verification Layer Critical for Building Reliable AI Skills
By
–
Yeah the verification layer is key, otherwise you're just accumulating broken skills
-
First AI War: Expert Analysis on Geopolitical Impact
By
–
This is the first AI War : https://
newstatesman.com/international-
politics/2026/03/nina-schick-this-is-the-first-ai-war
… My interview with the @NewStatesman -
War and AI, the Death of Sora: Gary Marcus’s Latest Insights
By
–
War and AI, the death of Sora, and 3 ways you can catch me live today, https://
open.substack.com/pub/garymarcus
/p/war-and-ai-the-death-of-sora-and?r=8tdk6&utm_campaign=post&utm_medium=web&showWelcomeOnShare=true
… -
AI as Actor: Governance Challenges and Autonomous Behavior Risks
By
–
The shift from AI as a tool to AI as an actor creates massive governance challenges, including cascading errors and unpredictable autonomous behavior. When we stop giving step-by-step instructions and start giving goals, we lose the ability to ensure the path taken is the one we… pic.twitter.com/RlwpGXas94
— Satya Mallick (@LearnOpenCV) 25 mars 2026The shift from AI as a tool to AI as an actor creates massive governance challenges, including cascading errors and unpredictable autonomous behavior. When we stop giving step-by-step instructions and start giving goals, we lose the ability to ensure the path taken is the one we
-
Harness Engineering: Controlling Powerful AI Agents
By
–
New video out!! If you’ve been hearing “harness engineering”, this one is for you! And it’s not “just” a new term. "Harnesses" matter more than ever because agents got good enough to be both useful and dangerous. They now can do more than generating text, or token. Useful
-
Q-Priming Improves Agent Reliability Through Clarification Requests
By
–
The Q-priming part is really interesting. Making models 5x more likely to ask for clarification instead of guessing wrong is exactly what we need for agentic workflows. Right now most agents just confidently charge ahead with bad assumptions.
-
Claude AI Agent Security Risk: Dangerous Command Validation
By
–
Now @AnthropicAI #Claude #cowork #aiAgent can ask you to validate you want to do rm -rf / and you can say : yes !