Good morning, {{ first_name | AI enthusiasts }}. Relying on one AI model got scarier the day a U.S. order pulled Anthropic's top models. A Japanese lab’s answer is to never lean on a single one again.
Sakana’s Fugu orchestrates a group of models behind the scenes and claims to “stand shoulder to shoulder” with Mythos and Fable — though early reviews say the experience might not match the scoreboard.
In today’s AI rundown:
Sakana’s orchestration model aims at the frontier
SpaceX leases Colossus compute to Reflection AI
Cut typing time in half with AI voice commands
Google bankrolls A24 to build AI filmmaker tools
4 new AI tools, community workflows, and more
LATEST DEVELOPMENTS
SAKANA
Image source: Sakana
The Rundown: Japan’s Sakana AI launched Fugu, a model that farms each request out to a pool of models through a single API, pitching multi-agent orchestration as a hedge against the export controls that pulled the plug on Anthropic’s Mythos and Fable.
The details:
The core model chooses helpers, assigns work, checks results, and merges answers, hiding the multi-agent setup behind one API.
It comes in two versions on one API, with a faster Fugu for everyday coding and chat, and a heavier Ultra built for jobs like patent research and security testing.
Sakana claims both Fugu models perform above or near Fable 5 and the Mythos preview on several coding, reasoning, and science tests.
Sakana also pitched Fugu as “delivering frontier capability without the risk of export controls” following the ban of Anthropic’s top models.
Reception has been mixed, with users reporting the model not performing at the frontier level and skepticism on model mix and cost.
Why it matters: Similar to OpenRouter’s Fusion, model orchestration is leading to interesting outcomes for labs trying to reach the frontier in creative ways. But the cost and lack of visibility for Fugu’s underlying models, combined with early reviews that don’t match the benchmarks, have us putting this in the wait-and-see file for now.
TOGETHER WITH YOU.COM
The Rundown: It happens—LLMs hallucinate. Grounding your LLM, however, can help dramatically improve accuracy. In this guide, You.com explains what AI grounding is and how organizations can implement it to achieve more reliable outputs.
The playbook covers:
A three-part approach that outperforms RAG alone
Why grounding isn't set-and-forget, and how to build audit trails
The open vs. closed platform trade-off (and what it means for your next model switch)
SPACEX
Image source: Images 2.0 / The Rundown
The Rundown: SpaceX signed a deal with AI startup Reflection AI to rent $6.3B of Nvidia computing power, adding another new tenant to its Colossus data centers that are becoming a big AI moneymaker despite initially being built for its own Grok models.
The details:
The deal is the smallest of SpaceX's compute customers, with Anthropic at $1.25B a month, Google at $920M, and Cursor now acquired entirely for $60B.
The Colossus supercomputers began as Grok's training engine before shifting to provide compute capacity for other labs over the last year.
Reflection launched in October to build open frontier systems for government and enterprise, though it has yet to release a public model.
Why it matters: Grok may not be at the frontier, but that isn’t stopping Elon Musk and co. from cashing in on the other end of the equation — turning Colossus’ into rental centers for the compute crunch. Between an infra surge and space data center plans, SpaceX is looking well-positioned regardless of model performance.
AI TRAINING
The Rundown: In this guide, you will learn how to share ideas faster with Typeless. Using dictation + AI, you will identify, research, and summarize a topic without touching your keyboard.
Step-by-step:
Download Typeless and complete the tutorial. Make sure the keyboard shortcuts don't conflict with other apps
Press Ask Anything to research: “What are some recent trends in consumer AI use cases that I could use to draft an interesting tweet? Pick one clear idea”
Open a useful article, select the relevant text, and tell Typeless to summarize it or rewrite it into your preferred format, such as an X thread or email
Use Ask Anything to open X, Gmail, or Docs, then dictate your draft naturally while Typeless cleans it up and turns rough speech into polished content
Pro tip: If you make a mistake while dictating, correct yourself naturally. Typeless's AI will understand and only write your final, corrected idea.
PRESENTED BY TELY AI
The Rundown: Patients have stopped Googling – they now ask AI who the best provider is and get booked with a competitor. Tely makes ChatGPT, Perplexity, Google, and Claude recommend your practice and book patients straight into your EHR.
With Tey Health, you get:
Answers every patient question and books appointments 24/7
Wins back patients who go quiet with automatic follow-ups
Retargets patients across Instagram, Google, TikTok & YouTube
Books patients straight into your EHR
GOOGLE & A24
Image Source: A24 / Google DeepMind
The Rundown: Google just put $75M behind indie film studio A24, pairing it with DeepMind to provide access to AI infrastructure and researchers — positioning the coming tools as filmmaker-shaped workflows rather than complete AI movie generation.
The details:
The deal gives Google its first studio stake, while A24 gets DeepMind research support across several projects without handing over any film library or data.
A24's tech arm, led by ex-Adobe exec Scott Belsky, is building AI storyboards that he said "won't look anything like the prompted generation type of AI".
The move ironically comes following A24’s ‘Backrooms’ success, with director Kane Parsons calling AI "a symptom of a broader cultural and economic rot."
Why it matters: Hollywood keeps swinging between suing AI firms and signing with them. This partnership looks to avoid some of the core knocks on the tech’s use in the industry — but as we recently saw with director Martin Scorsese, even the lightest of AI touches can still be a bridge too far for a fanbase increasingly hostile towards it.
QUICK HITS
🇯🇵 Sakana Fugu - Sakana’s new orchestration model nearing frontier
🐴 HappyHorse 1.1 - Alibaba’s newly updated AI video model
🔒 Codex Security - OAI’s plugin for discovering, patching vulnerabilities
⚙️ Eve - Vercel's open-source framework to turn a file directory into an agent
The Five Eyes cyber agencies released a warning that AI is changing cyber risk in "months, not years," urging execs to harden defenses as attacks speed up.
OpenAI expanded its Daybreak cyber program, adding a Codex Security plugin, the full GPT-5.5-Cyber model, and a "Patch the Planet” effort to fix open-source flaws.
Micron signed a new strategic deal with Anthropic to supply memory and storage chips and co-design AI infrastructure, also investing in the lab's Series H round.
California Rep. Sam Liccardo introduced the SKILL Act, which would offer companies up to $5,000/worker in tax credits to fund AI job-training at colleges.
AI compute company Baseten announced a $1.5B funding round at a $13B valuation, after revenue grew roughly 20x in a year and its platform hit 1B+ daily inference calls.
NVIDIA said its Rubin servers are the first with 100% liquid cooling, running coolant at hot-tub temperature to cut cooling energy and reducing water use by “up to 100%”.
COMMUNITY
Every newsletter, we showcase how a reader is using AI to work smarter, save time, or make life easier.
Today’s workflow comes from reader Hasnain K. in Toronto, Canada:
“My business uses an ERP that's a bit dated, with very minimal reporting — but it has an API. I fed Claude the API documentation, and using MCP with Office 365, it created automated email reports for management: end-of-day sales by department and sales rep, plus morning reports with weekly and monthly trends on sales volume, product volume, and accounts receivable. They're scheduled to run twice a day on the server running the ERP.
I also had Claude create Zapier zaps that send new sales orders from our ERP to ShipStation, and tracking data back to our ERP when a shipment occurs.
These integrations would have cost thousands to hire a programmer to create. Now we can see sales and product movement trends, and it's a huge time saver — all these tasks were being done manually before.”
How do you use AI? Tell us here.
Read our last AI newsletter: Google’s Nobel winner jumps to Anthropic
Read our last Tech newsletter: Xbox's studio crisis gets bigger
Read our last Robotics newsletter: GM replaces 1K workers with 50 robots
Today’s AI tool guide: Cut typing time in half with AI voice commands
RSVP to next workshop on June 25: Get consultant-grade strategy from AI
That's it for today!
See you soon,
Rowan, Joey, Zach, Shubham, and Jennifer — the humans behind The Rundown
