Pinned Tweet
We back founders building what lasts, across AI, robotics and Web3.
Capital when you need it. Builders beside you when it matters.
We don’t just fund companies. We forge them.
4
72
The reliability threshold just moved.
OpenAI's new report: coding agents now complete 46% of 2–4 hour research tasks on their own — up from 23% in March. On 4–8 hour tasks, 35%. The median researcher is already spending $2,400 a day on agent use.
This is the number that matters. Not benchmarks, not demos — multi-hour work, finished, unsupervised. The "jagged frontier" is the honest part: capability is uneven, agents ace some tasks and fail others. So the winners won't be the best models. They'll be the teams that build verification, evals, and workflows around the jagged edges — the ones who make unreliable capability reliable enough to sell.
Agents stopped being a demo this year. They're becoming labor. The question now is who owns the verification layer.
#AIAgents #AIResearch
We're sharing new data on how AI is changing research at OpenAI.
We find researchers' AI use continues to rise exponentially, and agents are completing longer tasks, but human involvement remains central in research. Researchers are also spending more time on safety and alignment, with human effort in these areas growing faster than in any other individual research area.
This report aims to increase transparency around research acceleration and inform the governance of increasingly capable AI. We believe that AI labs should be required to track and publicly report indicators of recursive self-improvement (RSI): the process by which AI accelerates the development of future AI systems.
Our hope is that this expanded report will help sustain and advance shared measurement norms across the industry.
openai.com/index/research-ac…
1
1
17
The $500M is the headline. The real story is distribution.
Tencent's Q3: WorkBuddy passed 10M DAU, and per Martin Lau the last 7 days annualize to $500M+ in revenue, 5x Q2. Hy4 topping Agent Arena is nice, but benchmarks don't sell seats. Distribution does: the agent lives inside workflows enterprises already use.
Two caveats before calling this the inflection point. A 7-day annualized run rate flatters — early adopters, pent-up demand. What matters is renewal: does usage convert to paid seats, and does it expand? Token volume is scale, not revenue. Usage is vanity until it's monetized.
The structural signal: agents eat the interface layer, not the system of record. CRM is next — the data stays, the UI goes. Tencent Cloud up 35% says the same thing from the other side. Startups win vertical, not horizontal.
#AIAgents #SaaS
Tencent’s Q3 results show AI’s significant achievements across our models and applications. Revenue reached $30.78 billion, up 8% year on year.
Our @TencentHunyuan Hy4 model ranked first among Chinese models on Agent Arena at launch, @WorkBuddy_AI passed 10 million daily active users, and @TencentCloud’s revenue growth rate accelerated to 35%.
More: buff.ly/CyK7XDv
1
14
1/ Meet @UVioletAI — one of the boldest bets in our portfolio.
No blockchain. No gas. No forks.
Client-side validation + zk-proofs, built to run AI and IoT workloads at a TPS blockchains can't touch — anchored to Bitcoin and Lightning.
3/ Behind it: Dr Maxim Orlovsky @dr_orlovsky.
The mind behind RGB smart contracts on Bitcoin, AluVM, and the LNP/BP Association. A cypherpunk and neuroscientist who has spent a decade building the post-blockchain stack.
1
4
30
GPT-6.2 Astra is OpenAI's first released model to hit its 'High' bar for AI self-improvement — a model that can materially accelerate AI research itself. When research starts compounding, the best founders build for the second-order effects.
1
42
Grateful to the partners standing with Everforge Labs from day one:
@fxh_crypto
@DFG__Official
@FundamentalLabs
@waterdripfund
@CerasVentures
Thank you for your support. Here’s to backing the founders building what lasts across AI, robotics and Web3.
4
32






