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? 👇
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? 👇
15
Edge Of AI retweeted
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 what’s next: langchain.com/blog/series-c
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 — 2.5X faster than 3.8 Flash, outpacing competing models in its class.
Jensen Huang has stepped down as Nvidia CEO after 33 years. CFO Colette Kress is interim CEO while the board searches for a permanent one.
The company behind most of the world's AI compute just got a new chapter. What changes for AI builders? 👇
NVIDIA announces CEO transition. Read the release: nvidianews.nvidia.com/news/n…
Claude just moved into whiteboards, sheets and data. The AI tab is slowly turning into the whole office suite.
Which one would you actually use first? 👇
Should every workflow run on AI?
The CEO of @GoKiteAI calls herself an AI maximalist: if a workflow isn't powered by AI yet, it's still in development. The payoff is more human time for creative, cutting edge problems.
Overkill or the future? 👇
13
Live now: who actually owns the data that trains AI? Josh Kriger with Andrea Muttoni @devrelius of the Data Foundation.
Tune in 👇
Who owns the data behind AI?
Join Josh Kriger with Andrea Muttoni (@devrelius) of The Data Foundation for our next Edge Live.
📅 Nov 5 | 1:30 PM CST
Join us on X Spaces: xitter.cf/i/spaces/1aWbXoWQPZNzP…
26
"If a workflow isn't powered by AI, it's still in development."
The CEO of @GoKiteAI, the payment layer for AI agents, at Money20/20.
Bold take or exactly right? 👇
15
Verifying fewer than 4% of steps for 9 to 13% extra API cost is a tradeoff teams can actually budget for. Small judge at the risky forks, big model everywhere else. Keen to see it on messier repos.
Can a 4B model help a coding agent ~70× its size make better decisions?
Introducing SWE-Verifier, a lightweight 4B verifier that selectively guides coding agents at critical decision points.
Unlike trajectory or all-step verification, SWE-Verifier only samples and verifies additional actions when needed, improving performance at a fraction of the cost.
On SWE-Bench Verified with mini-swe-agent, SWE-Verifier improves MiniMax-M2.7 and DeepSeek-V4-Flash by ~1.7 percentage points, verifying fewer than 4% of steps with only 9–13% additional API cost.
📄 Paper: arxiv.org/abs/2610.30284
🧵 How it works, training recipe, and results 👇
87
Even the best AI model scores just 12% on GDPx, @SemiAnalysis_'s new test of agents on high-value work across 11 domains, from chip design to quant trading.
Agents are moving fast. Expert-level work is still a much higher bar.
Where would you trust an agent first? 👇
Today we’re releasing GDPx, a benchmark for measuring agentic capabilities across the most valuable domains.
GDPx is a true frontier-level benchmark: even the best model scores just 12% (1/7)🧵
1
1
61
What do strong AI operators know that most companies still miss?
The CEO of @GoKiteAI says it is data management: your know-how about your business and customers, and how well you manage that context.
Is your data ready for AI agents? Tell us below 👇
21
Before you ship an AI tool, ask: does it actually work?
1. What does good look like? Write 10 real examples.
2. Who grades it? A rubric, not a gut feeling.
3. When do you re-test? Every model or prompt change.
Which step does your team skip?
5
Your brain runs on 20 watts. Today's AI needs trillions of tokens to learn.
@peterevoss of @Aigo_Community says real intelligence learns like we do, not by brute force.
Is bigger always better in AI? 👇
7
AI won’t just be getting prompts. It could soon be getting budgets.
Zach Francis, Co-founder of Willo, shares how AI could start buying software, hiring other agents, and spending money to get things done.
@zach_francis @BuildwithWillo
10
2017: about $1M raised, much of it on a dev team for one app.
2026: four products in three days, no developers.
@erikanswerman put an AI builder on trial with @joshkriger. Verdict: use it.
What would you build in a weekend?
10
AI can't replace human connection.
@TheEricPulier on The Edge of Show with @joshkriger: AI can give you superpowers. It's not going to be great for human connection.
When did you last turn your screen off on purpose? 👇
13
Open vs closed AI models, in plain English:
Open: run it on your own servers, your data stays with you, your team keeps it running.
Closed: strongest results out of the box, pay per use, read the data terms.
Most teams end up using both. Which way is yours leaning?
12
4 minutes in 2024. 17 hours earlier this year. 17 weeks now, at a 50% success rate. Most company AI plans were written for the 4 minute version. Which workflow would you hand over first? xitter.cf/emollick/status/211332…
When my first AI book was published in 2024, the best AI could do 4 minutes of human work with a 50% chance of success
When I turned in the draft of Co-Existence earlier this year, it was 17 hours
When it was published two days ago, AI could do 17 weeks open.substack.com/pub/oneuse…
23


















