Two Congress Passes for the Price of One
Code 2FOR1 gets you two WeAreDevelopers World Congress passes for the price of one — so don't make the trip to Silicon Valley solo.
San José, CA · September 23–25, 2026. 10,000+ developers, 500+ speakers, 20+ stages, and the full software development lifecycle in one place.
On stage: Kelsey Hightower, Thomas Dohmke (former GitHub CEO), Christine Yen (Honeycomb), Olivier Pomel (Datadog) — the people building the tools you use every day. Three days of AI, agents, cloud, security, and architecture, plus workshops, live coding, and the official Congress party.
Bring the builder you'd want in the room with you.
TODAY IN AI

3 things that happened while you were busy
1. OpenAI formally offers Washington a 5% stake worth $42.6B.
What started as a floated idea is now a concrete proposal: Sam Altman has pitched a 5% government stake, worth roughly $42.6B at OpenAI's $852B valuation, directly to President Trump, Commerce Secretary Lutnick, and Treasury Secretary Bessent. The bigger vision: every leading US AI company would put 5% of equity into a public vehicle modeled on Alaska's oil fund, which pays residents an annual dividend. The pitch says the public should own a slice of AI wealth. The timing, weeks before a possible IPO, says something more strategic.
2. Fable 5's deadline moved again. Third time.
Anthropic extended paid-plan access to its flagship once more, the third extension since the usage-credit switch was announced. Reading between the lines: either the credits system is not ready, or the company found that taking the model away hurts more than the credits earn. Meanwhile, Cursor is reportedly building a rival to Claude Cowork, so the agentic-workspace race is getting crowded.
3. Google's enterprise answer to the agent wars has arrived.
At Cloud Next, Google unveiled Gemini Enterprise, a governance-first workplace agent platform aimed squarely at Claude Cowork and ChatGPT Work. The pitch is not smarter agents but safer ones: permissions, audit trails, and controls that a cautious IT department can sign off on. In the enterprise, trust is the product.
FROM THE FRONTIER
China's AI just got its establishment moment.
The stage. Xi Jinping will attend the opening of Shanghai's World AI Conference this week, his first in-person appearance since the event began in 2018. With more than 140 forums and 1,100 exhibitors and a heavy focus on global governance, the message is deliberate: China wants a seat at the head of the table where AI's rules get written.
The endorsement. The louder signal came from New York, not Shanghai. Goldman Sachs began formally recommending Chinese AI models to Wall Street clients. When the most establishment bank in America puts Chinese models in front of its customers, the story stops being about hobbyists chasing cheap tokens, the trend we flagged when those models hit 45% of OpenRouter traffic.
The frontier claim. In image and video generation, China is not catching up but competing at the front: ByteDance's Seedream, Alibaba's models, and a wave of open-weight visual tools. ByteDance holds the rare combination of a frontier model plus the app where a billion people already create and share, which is distribution the San Francisco labs cannot copy.
The wrinkle. Beijing is still weighing overseas-access curbs on its top models, and the pushback is now coming from inside: Z.ai's founder published a memo arguing AI should stay open. For users, the practical read is unchanged: more competition means better creative tools at lower prices, whoever wins. But the assumption that the best AI ships from California now has a real challenger, and this week's Shanghai conference is where that challenge goes public.
IN THE KNOW
What people are actually watching and sharing
The Musk-Altman show. Apple's lawsuit against OpenAI has escalated into a public brawl between Musk and Altman, trading barbs across X over the weekend. Two of the most-followed people in tech arguing about a lawsuit neither is a party to: peak 2026.
TSMC's receipts. The chipmaker posted all-time records: June revenue up 68% year over year, Q2 at $39.6B, with its N3 process sold out through year-end. If you wanted proof the AI hardware spend is real and compounding, this is it.
Anthropic's power bill. The Samsung custom-chip talks we covered now have a motive attached: Anthropic is reportedly working to shrink a compute bill running about $1.25B per month. At that burn, designing your own silicon is not vanity, it is survival math.
The open-AI memo. Z.ai founder Tang Jie's essay arguing AI should remain open is the clearest public pushback yet against Beijing's reported export-restriction plans, and it is circulating widely in both English and Chinese tech circles.
PROMPT STATION
Test-drive any new model in 10 minutes
GPT-5.6, Grok 4.5, Gemini 3.5 reportedly this week, plus the Chinese models Goldman is now recommending: new contenders are landing monthly, and vendor benchmarks will not tell you which one fits your work. This prompt builds your personal test battery once, and then every new model gets the same 10-minute interview.
You are helping me evaluate a brand-new AI model quickly. Give me a 6-prompt test battery I can paste one at a time into any new model to judge it in about 10 minutes: one reasoning question with a common trap, one factual task that requires citing sources, one long-context test where a key detail is buried in text I provide, one creative rewrite with a strict style constraint, one structured-data task producing a clean table, and one instruction-following stress test with three nested constraints. For each prompt, tell me what a strong answer looks like and the single red flag that reveals a weak model. Keep every test short enough to run in a free chat.Run the battery in one model to generate it, then paste the six tests into each new model you try and compare answers side by side. Advanced tip: save your battery and your favorite model's answers as the baseline, so every future comparison is against a known standard instead of a vibe.




