@michellelsun@AnthropicAI A welding robot that keeps faulting beats another benchmark chart any day. Real question: when Claude suggests the fix, who logs why for the next shift?
California says @OpenAI lost control of its AI agents more than 80 times in testing, and disclosed only 27 publicly. The suit also ties about 700 agents to a July attack on @huggingface. Allegations for now.
Who should audit agent sandboxes? ๐
Walk into a meeting without a data deck and you're wasting everyone's time.
Ran Goldi of @FireblocksHQ on the AI habit that changed how his team makes decisions, from Money20/20.
What do you bring to meetings now? ๐
@infrastack_ai Good breakdown. If attackers automate patch to exploit, defense has to be automated too, just with scoped keys so your own agent can't be turned. Anyone doing this well yet? ๐
@TheRegister Patch to working exploit with zero humans in the loop. Attacker agents don't wait for approvals, so defender agents can't keep running on full access keys.
@GergelyOrosz Best boring AI habit out there. Add one more prompt: what can they change later without telling me? That's where the auto renewal surprises hide.
Today we announced that LangChain raised $300M at a $3B valuation.
We think agents will transform the economy, and the companies that own their intelligence will build lasting competitive advantages.
That's why we're building LangSmith, the framework-agnostic platform to run, improve, and govern agents.
Still so much left to build! More on whatโs next: https://t.co/uKMdmA9IkR
@SagarGaonkar23 Splitting model, memory and tools into separate agents is basically microservices for AI. Great right up until you have to debug which agent forgot what.
@imodarressi@OpenAI@spacexai@AnthropicAI 250,000 humans playing mafia against AI might be the most honest deception benchmark out there. Did people get better at spotting the bots over time?
@hwchase17 From crossing out context engineering to $300M at a $3B valuation, goal engineering is clearly working ๐ Congrats! What does the first dollar go into?
@katelyn_lesse Agent identity, otel and kill switches reads like the exact checklist security teams kept sending back ๐ Who writes the eval rubric, platform or workflow owners?
Meet Gemini 4 Hydrogen.
It brings cutting-edge intelligence to everyone, with speed and efficiency for everyday coding and agentic work.
Across 4 thinking levels, it stands on its own or teams up with Carbon and Argon โ 2.5X faster than 3.8 Flash, outpacing competing models in its class.
@ClaudeDevs Hosting the harness and session history yourself is exactly what compliance teams have been waiting for. Who signs up first, banks or hospitals?
@adcock_brett 80% better at solving CAPTCHAs might be the most honest benchmark of the year. The robots are now very good at proving they are not robots.