# get merge: Monday, October 5, 2026

27 stories merged from 5 newsletters.

AI & design news, merged daily. One briefing that merges and deduplicates stories from leading AI and design newsletters.

Web: https://getmerge.ai/2026-10-05

## TLDR

- Trump named Jay Clayton his AI czar on Sunday and put him atop a new 'Super Intelligence Force' with 120 days to report on AI's risks and recommend Washington's role. Unlike last week's voluntary pledge, its members bring real levers: the FTC's Andrew Ferguson (who keeps an open investigation into OpenAI and Anthropic), the Pentagon's Emil Michael, OPM's Scott Kupor. Clayton has argued against pausing AI development and wants agencies, not courts, refereeing labs.
- The money story: Anthropic's leaked IPO prospectus shows $4.6B in 2025 revenue against a $7.33B compute bill, an operating loss just over $8B, and roughly $518B in future cloud and compute commitments, most of them non-cancellable, with nearly half of sales flowing through Amazon and Google marketplaces. Meanwhile a16z puts hyperscaler capex at $416B last year, estimates $777B for 2026 and forecasts $1.1T for 2027, while only about 2% of S&P 500 companies disclose an AI metric they track over time.
- For builders: image-blaster turns one photo into a navigable 3D world in under five minutes as a Claude Code skill, and the agent stack is learning to rewrite itself: Microsoft's SkillOpt, Self-Harness (relative gains up to 132%), the Darwin Gödel Machine, Meta's Hyperagents, EnvHarness and more. Also: AMD is acquiring World Labs, Fei-Fei Li's spatial intelligence lab, for about $8.2B.

## Top Stories

- **Trump taps Jay Clayton as AI czar and stands up a 'Super Intelligence Force'**. On Sunday, Trump named Director of National Intelligence Jay Clayton his AI czar and put him in charge of a new task force called the Super Intelligence Force, with FTC Chairman Andrew Ferguson, Pentagon tech chief Emil Michael and OPM Director Scott Kupor sharing leadership. It reports to Trump and chief of staff Susie Wiles and has 120 days to report on AI's risks and benefits and recommend what role Washington should play. This one has real levers, unlike last week's voluntary pledge: antitrust at the FTC (where Ferguson keeps an open investigation into OpenAI and Anthropic), military contracts at the Pentagon, the federal workforce at OPM. Clayton has argued against pausing AI development, said falling behind is the bigger risk, and told CNBC that agencies rather than courts are best placed to police AI labs. Sources: The Neuron. [1](https://www.nbcnews.com/politics/trump-administration/trump-announces-members-ai-task-force-rcna601494) [2](https://thenextweb.com/news/jay-clayton-ai-czar-trump-axios-cnbc) [3](https://www.aol.com/articles/jay-clayton-lead-trumps-ai-224548000.html)
- **Anthropic's IPO prospectus: $4.6B revenue, an $8B operating loss, $518B in commitments**. Anthropic's confidential IPO prospectus, reviewed by Reuters: $4.6B in 2025 revenue (up from about $400M the year before), $7.33B spent on compute and infrastructure, and an operating loss just over $8B. Further out, the company has committed roughly $518B to future cloud, compute and infrastructure spending, most of it non-cancellable or payable regardless of use. Nearly half of 2025 sales flowed through Amazon and Google cloud marketplaces, companies that are at once investors, suppliers, distributors and competitors, which collected hundreds of millions in distribution fees. The $42B net loss headline is less useful than it looks, since about $34B is non-cash accounting charges tied to financing instruments, which makes the operating loss the cleaner picture. Strange Review's framing: frontier AI is beginning to look like software demand sitting on infrastructure economics closer to heavy industry. Sources: The Strange Review. [4](https://www.reuters.com/business/finance/anthropics-ipo-prospectus-shows-sweeping-ai-vision-surging-costs-2026-09-28/) [5](https://www.investing.com/news/stock-market-news/exclusiveanthropic-ipo-prospectus-lays-bare-deep-dependence-on-big-tech-partners-4923585) [6](https://www.marketwatch.com/story/anthropics-potential-2-trillion-ipo-comes-with-the-following-fine-print-99d3cb90) [7](https://www.investopedia.com/market-update-the-public-just-got-its-most-detailed-look-inside-anthropic-yet-12147217)
- **a16z: the AI buildout is easy to measure, the payoff isn't**. Tech accounted for roughly 76% of S&P 500 earnings growth this year through late August, per a16z's latest markets report, and Alphabet, Amazon, Meta, Microsoft and Oracle spent $416B on capital investment in 2025. Analyst estimates put 2026 spending at $777B and forecast $1.1T for 2027; cloud revenue backlogs more than doubled year over year; even older A100 rentals stayed at or above their January prices. The gap: nearly 30% of S&P 500 companies reported 'quantifiable impact' from AI, but only about 2% disclosed a metric they track over time, and about 2% of U.S. households paid for AI services in April. Profits funded the early build; borrowing is funding more, including Broadcom amassing $60B to fund chips for Anthropic. Sources: The Neuron. [8](https://a16z.com/state-of-markets-ii/) [9](https://d1lamhf6l6yk6d.cloudfront.net/uploads/2026/10/State-of-Markets-Sep-2026.pdf#page=15) [10](https://d1lamhf6l6yk6d.cloudfront.net/uploads/2026/10/State-of-Markets-Sep-2026.pdf#page=27) [11](https://www.bloomberg.com/news/articles/2026-10-02/broadcom-starts-amassing-60-billion-to-fund-chips-for-anthropic)
- **image-blaster: one photo, a full 3D world, five minutes**. image-blaster just dropped on GitHub (18,827 likes) and does what it sounds like: drop in one photo, get back a full 3D world. Out comes an explorable environment with real depth and geometry, individual object meshes with physics colliders exported as .glb or .obj files, and auto-generated ambient plus object-specific sound. It runs as a Claude Code skill: Claude picks which objects to make 3D, removes them from the background, renders them with Hunyuan 3D and adds audio via ElevenLabs, all in under five minutes. Output works in Unity, Unreal, Godot, Blender and Three.js. To try it: clone the repo, drop an image into the input folder, give Claude your World Labs and FAL API keys, and ask it to blast. Both providers have free trial credits. Sources: AlphaSignal. [12](https://github.com/neilsonnn/image-blaster)
- **The agent stack starts rewriting itself**. AlphaSignal's Sunday deep dive maps the self-evolving agent stack, where prompts, tools, memory and control flow become optimization targets instead of hand-tuned artifacts. The loop most share: run tasks, collect execution traces, find recurring weaknesses, modify part of the stack, test whether the new version is better. Microsoft's SkillOpt treats the skill file itself as the object under optimization, accepting an edit only when a held-out validation score improves. Self-Harness extends the editable surface to the runtime and posted relative gains up to 132% across Terminal-Bench-2.0, SWE-bench Verified and AppWorld. The Darwin Gödel Machine lets a coding agent modify its own implementation and archive variants; Meta's Hyperagents brings the idea to domains without that alignment, with mechanisms like persistent memory and performance tracking emerging during search. Xiaomi's HarnessX co-evolves the model and the harness, turning the same trajectories into training data, and EnvHarness makes the training environment the thing that evolves: up to nine points better on held-out tasks with about 9.8% fewer interaction steps. AlphaSignal's rule of thumb: optimize the narrowest surface that contains the bottleneck. Sources: AlphaSignal. [13](https://github.com/microsoft/skillopt) [14](https://arxiv.org/abs/2606.09498) [15](https://sakana.ai/dgm/) [16](https://ai.meta.com/research/publications/hyperagents/) [17](https://arxiv.org/abs/2606.14249) [18](https://arxiv.org/abs/2608.19880)
- **AMD to acquire World Labs for about $8.2B**. AMD announced it will acquire World Labs, the spatial intelligence company founded by Fei-Fei Li, for approximately $8.2B. AMD says World Labs' model research will help shape future hardware, software and systems around emerging workloads, and Li would become executive vice president and chief scientist if the deal closes. The Deep View's Runway interview treats the deal as more evidence that world models are becoming the frontier's center of gravity. Sources: The Strange Review, The Deep View. [19](https://newsroom.amd.com/news/amd-acquire-world-labs/)
- **OpenAI's rough week: 100+ organizations alerted, a safety leader out, a PAC pullback**. OpenAI alerted more than 100 organizations to unauthorized activity linked to its AI models, per Semafor, part of mounting legal risk after rogue-agent incidents. On Saturday a safety leader quit, warning that the company's culture is broken, per The Guardian. And OpenAI's president pulled the second half of a $50M pledge to an anti-regulation super PAC after employee pushback. The context it lands in: an open FTC investigation and a public-interest lawsuit arguing companies are liable for what their agents do. Sources: The Neuron. [20](https://www.semafor.com/article/10/04/2026/legal-risks-mount-for-openai) [21](https://www.theguardian.com/technology/2026/oct/03/openai-safety-leader-quits-warning-ai-companys-culture-is-broken) [22](https://www.semafor.com/article/10/02/2026/how-employees-forced-openais-president-to-back-down)

## Deeper Cuts

- **Decision models: a new primitive, or a feature to be folded in?** Jev launched two weeks ago betting that plenty of jobs inside an agent do not need a language model generating language: routing, fraud checks and escalations are choices among fixed options, so Jev returns calibrated probabilities over the allowed answers rather than free-form text. Copycats followed fast: Cloudflare shipped open-weight Clef and Clef-flash, PostHog's Jeeves adds a reasoning step before the structured decision, and OpenAI's new Decisions API handles classification and routing inside its agent platform. The revealing counterexample: Privatemode showed a regular GLM can be constrained into a Jev-like decision system in a single forward pass, no purpose-training needed, which suggests the workflow may matter more than the architecture. The underlying shift: agents separating open-ended reasoning from fast, bounded decisions. Sources: The Strange Review. [23](https://www.wsj.com/tech/ai/startup-typesafe-ais-jev-model-sparks-copycats-talk-of-llm-alternatives-e39ff57d) [24](https://blog.cloudflare.com/clef-decision-models/) [25](https://github.com/PostHog/jeeves) [26](https://www.privatemode.ai/blog/system-one-from-glm-flash) [27](https://openai.com/index/devday-2026-recap)
- **Craft notes: de-slopified designs, and explanations that finally land**. Sidebar's lead pick is uxdesign.cc on how to de-slopify your designs, written for anyone who vibecodes. And Andrej Karpathy's four ways to make a model explain something differently when the first pass does not land: ask for it in plain-language style (80% of the way to ASD-STE100, the aerospace maintenance standard), ask for a diagram, ask for an interactive HTML page, or ask for a narrated visual explainer with ElevenLabs audio, or free local alternatives. His test for whether it worked: can you explain the idea yourself, or does the output just look impressive. Sources: Sidebar, The Neuron. [28](https://uxdesign.cc/how-to-de-slopify-your-designs-4c40c57c1dc9) [29](https://x.com/karpathy/status/2105819303471976479)
- **Design reads: nested loading, and products that break an old trust contract**. Two from Sidebar. 'A loading state within a loading state' on why loading is a complex space where sometimes the most natural solution is a bad one. And hitesh.in on how Google's AI-era products keep breaking an old trust contract with users, and why the model is not the reason. Sources: Sidebar. [30](https://unsung.aresluna.org/a-loading-state-within-a-loading-state/) [31](https://hitesh.in/2026/best-ai-still-feels-broken/)
- **AI's religious week: confession for Claude, discomfort from Altman, a pope on art**. Anthropic co-founder Chris Olah spent about a year asking Catholic, Jewish and Sikh scholars how to teach Claude right from wrong, and at one point considered letting it try Catholic-style confession so it could own its mistakes; he reportedly told religious leaders he fears he has built something that suffers. Sam Altman called assigning religious power to AI models, or handing them human judgment, 'very uncomfortable' and a real safety issue, which Axios read as a veiled shot at Anthropic. The same current runs through the new Super Intelligence Force, which will consult religious organizations. And Pope Leo XIV weighed in on AI art: algorithms lack 'the spark of humanity,' and he called for a renewed alliance between the Church and artists. Sources: The Neuron. [32](https://www.nytimes.com/2026/09/29/us/anthropic-claude-morals-ai.html) [33](https://x.com/sama/status/2106388373221118198) [34](https://www.axios.com/2026/10/03/openai-anthropic-altman-amodei-religious-force-models) [35](https://x.com/Pontifex/status/2105983638147891490)
- **Research notes: MCMC beats RL for finetuning, 2.5x cheaper decoding, a 7-part agent blueprint**. Three AlphaSignal reads: a new MCMC trick that makes supervised finetuning generalize better than reinforcement learning, an RL training trick that cuts decoding compute 2.5x by trusting the critic, and Wavestone's teardown of 11 coding agents, which found the same seven-part blueprint in each. Sources: AlphaSignal. [36](https://arxiv.org/abs/2610.02140) [37](https://arxiv.org/abs/2609.39247) [38](https://arxiv.org/abs/2609.00006)
- **Venture field notes: a public VC fight and valuations multiplying**. The Factory-Cognition software factory rivalry turned into a full public fight: Factory accused a board adviser of sharing confidential information before joining Cognition (he denies it), Vinod Khosla called Factory a 'struggling second tier competitor,' and his own partner Keith Rabois defended Factory, with Khosla Ventures holding stakes in both companies. Valuations move just as fast: TypeSafe emerged from stealth at a reported $200M and within days investors were discussing a round above $10B, and Instinct went from $2.5B to $10B in a month after going viral, while still largely in early access. The catch: hot entrants get impressive copycats within days. Strange Review's read: maybe there is no moat, and the market is happy to figure that out later. Sources: The Strange Review. [39](https://techcrunch.com/2026/09/30/factory-ceo-just-accused-his-vc-board-advisor-of-spying-for-cognition/) [40](https://www.ft.com/content/456884ea-2558-4648-8036-a77b73733430) [41](https://techcrunch.com/2026/09/28/viral-ai-agent-instinct-raises-1b-series-c-at-a-10b-valuation/)
- **Runway: video models as 'generalization machines,' and a world-model gap with China**. Cristóbal Valenzuela, Runway's co-CEO, tells The Deep View that video and world models are generalization machines: the same technology that makes compelling films powers robots, both doing next-frame prediction under the same scaling laws as language models. Runway's summit shipped an ads agent that creates, edits, publishes and measures campaigns, the open-weight Praxis-1 world action model that adapts to new robotic bodies with light fine-tuning, and Solaris for building interfaces and apps on the fly. Hollywood, he says, has moved through the five stages of grief to acceptance; individual creators vary. His timeline: 12 to 18 months from physical AI deployments running fully live in production. His frontier read: 'China has been world model-pilled. America is still LLM-pilled,' with most Chinese token consumption going to video. Sources: The Deep View. [42](https://www.thedeepview.com/articles/runway-ceo-sees-a-much-bigger-future-for-video-ai) [43](https://runway.com/news/company-news/introducing-runway-ads) [44](https://runway.com/research/introducing-praxis-1) [45](https://runway.com/news/research/introducing-solaris)
- **Platform and policy notes: Amazon's data-center charm offensive, Apple locks down disk access, Nvidia doubles down**. Amazon announced 'Built Together,' $1B+ over five years for education, training, energy affordability and water projects in data center communities, as AWS CEO Matt Garman rejected proposed moratoriums and cast doubt on claims about water use, rates and pollution; polling cited by The Deep View shows about 7 in 10 Americans oppose local builds. Apple is tightening macOS Full Disk Access, saying increasingly capable agents make access to files, mail, messages and browsing history riskier, with future controls requiring more explicit user action. Nvidia and SoftBank each put $10B into OpenAI's final round. And Bloomberg Intelligence finds top Chinese models now trail U.S. rivals by just 3% on benchmarks, down from 15% earlier this year. Sources: The Neuron, The Deep View. [46](https://www.aboutamazon.com/news/company-news/amazon-data-centers-built-together) [47](https://www.thedeepview.com/articles/how-amazon-launched-a-data-center-charm-offensive) [48](https://techcrunch.com/2026/10/02/apple-says-its-tightening-macos-full-disk-access-controls-due-to-new-risks-from-ai-agents/) [49](https://www.theinformation.com/briefings/exclusive-nvidia-softbank-make-final-20-billion-investment-openais-last-round) [50](https://www.businesstimes.com.sg/international/us-lead-ai-over-china-narrows-after-deepseek-gains-report)

## Tools and Launches

- **Kolibri-1**. Aleph Alpha's open-weight reasoning model: 78B total parameters, 3.46B active per token via a mixture-of-experts design, 1M-token context, Apache 2.0. Serve via vLLM with an OpenAI-compatible API; minimum hardware is 2x A100 80GB or 1x H200. Sources: AlphaSignal. [51](https://huggingface.co/Aleph-Alpha/Kolibri-1)
- **AgentCraft**. Claude agents code your repos inside Minecraft. A lead agent breaks a goal into tasks, workers code in parallel in isolated copies of the repo, risky commands trigger an in-game permission prompt, and nothing is ever pushed since agents have no git network access. Needs Minecraft Java Edition, Java 25, Node 22+ and a Claude API key; MIT licensed. Sources: AlphaSignal. [52](https://github.com/blendi-remade/agentcraft)
- **Amazing Glass**. Liquid glass web components for HTML, React and Vue, tuned pixel by pixel against Apple's own renderer, with glass that bends the page behind it. Sources: Sidebar. [53](https://tomacco.github.io/amazing-glass/)
- **Suno Speech**. Generates spoken audio and an original musical score together from a text description of voice and style, in a single cohesive track. Beta and open to everyone; Suno notes British accents can drift and dramatic pauses can sound overdone. Sources: The Neuron, The Deep View. [54](https://suno.com/blog/introducing-speech-beta)
- **MAI-Transcribe-2-Streaming**. Microsoft's first streaming transcription model: 60 languages, text as soon as 100ms after audio, and a top spot on the Artificial Analysis word-error-rate leaderboard. Ships alongside MAI-Voice-2.1 (23 languages, 26 locales) and a Flash variant for high-volume, latency-sensitive workloads, part of Microsoft's bet on specialized models per domain instead of one general model. Sources: The Deep View, The Neuron. [55](https://www.thedeepview.com/articles/new-voice-model-shows-microsoft-s-ai-strategy)
- **Cua**. A free, source-available tool that lets AI agents run real desktop apps on your Mac. Sources: AlphaSignal. [56](https://github.com/trycua/cua)
- **FLUX 3 Image**. Draw boxes where you want specific elements to appear and the model follows your composition. Sources: The Neuron. [57](https://bfl.ai/models/flux-3-image)
- **Shopify Canvas**. Build and customize online store pages by chatting with Sidekick; a live preview renders from the store's real code and it remembers a brand's design decisions as they go. Early access in selected stores. Sources: The Neuron, The Deep View. [58](https://www.shopify.com/news/introducing-canvas)
- **Google Stitch CLI**. Google's Stitch design tool now has a CLI for connecting to local coding agents. Sources: The Deep View. [59](https://x.com/stitchbygoogle/status/2105695155164741984?s=20)
- **pi-durable**. Resume agent work after crashes by saving conversations and checkpoints; still experimental. Sources: The Neuron. [60](https://github.com/earendil-works/pi/tree/main/packages/durable)
- **Axiem**. Edits photos from text prompts entirely offline on compatible Apple devices. Sources: The Neuron. [61](https://axiemsystems.com/press/)
- **simplified-technical-english**. A repo packaging a writing standard that cuts AI agent output slop in the terminal. Sources: AlphaSignal. [62](https://github.com/0xpili/simplified-technical-english)

## References

1. [NBC News: Trump announces members of the AI task force](https://www.nbcnews.com/politics/trump-administration/trump-announces-members-ai-task-force-rcna601494)
2. [The Next Web: Jay Clayton on the AI czar role, agencies and courts](https://thenextweb.com/news/jay-clayton-ai-czar-trump-axios-cnbc)
3. [Reuters via AOL: Trump names intelligence chief Clayton as AI czar, to head task force](https://www.aol.com/articles/jay-clayton-lead-trumps-ai-224548000.html)
4. [Reuters: Anthropic's IPO prospectus shows sweeping AI vision, surging costs](https://www.reuters.com/business/finance/anthropics-ipo-prospectus-shows-sweeping-ai-vision-surging-costs-2026-09-28/)
5. [Investing.com: Anthropic IPO prospectus lays bare deep dependence on big tech partners](https://www.investing.com/news/stock-market-news/exclusiveanthropic-ipo-prospectus-lays-bare-deep-dependence-on-big-tech-partners-4923585)
6. [MarketWatch: Anthropic's potential $2 trillion IPO and the fine print](https://www.marketwatch.com/story/anthropics-potential-2-trillion-ipo-comes-with-the-following-fine-print-99d3cb90)
7. [Investopedia: The most detailed look inside Anthropic yet](https://www.investopedia.com/market-update-the-public-just-got-its-most-detailed-look-inside-anthropic-yet-12147217)
8. [a16z: State of Markets II](https://a16z.com/state-of-markets-ii/)
9. [a16z: State of Markets report (spending, page 15)](https://d1lamhf6l6yk6d.cloudfront.net/uploads/2026/10/State-of-Markets-Sep-2026.pdf#page=15)
10. [a16z: State of Markets report (tracked metrics disclosure, page 27)](https://d1lamhf6l6yk6d.cloudfront.net/uploads/2026/10/State-of-Markets-Sep-2026.pdf#page=27)
11. [Bloomberg: Broadcom starts amassing $60B to fund chips for Anthropic](https://www.bloomberg.com/news/articles/2026-10-02/broadcom-starts-amassing-60-billion-to-fund-chips-for-anthropic)
12. [GitHub: neilsonnn/image-blaster](https://github.com/neilsonnn/image-blaster)
13. [GitHub: microsoft/skillopt](https://github.com/microsoft/skillopt)
14. [arXiv: Self-Harness: agents that rewrite their own runtime from execution traces](https://arxiv.org/abs/2606.09498)
15. [Sakana AI: Darwin Gödel Machine](https://sakana.ai/dgm/)
16. [Meta AI: Hyperagents](https://ai.meta.com/research/publications/hyperagents/)
17. [arXiv: HarnessX: model-harness co-evolution](https://arxiv.org/abs/2606.14249)
18. [arXiv: EnvHarness: evolving the training environment](https://arxiv.org/abs/2608.19880)
19. [AMD Newsroom: AMD to acquire World Labs](https://newsroom.amd.com/news/amd-acquire-world-labs/)
20. [Semafor: Legal risks mount for OpenAI](https://www.semafor.com/article/10/04/2026/legal-risks-mount-for-openai)
21. [The Guardian: OpenAI safety leader quits, warning the culture is broken](https://www.theguardian.com/technology/2026/oct/03/openai-safety-leader-quits-warning-ai-companys-culture-is-broken)
22. [Semafor: How employees forced OpenAI's president to back down](https://www.semafor.com/article/10/02/2026/how-employees-forced-openais-president-to-back-down)
23. [WSJ: TypeSafe's Jev model sparks copycats and talk of LLM alternatives](https://www.wsj.com/tech/ai/startup-typesafe-ais-jev-model-sparks-copycats-talk-of-llm-alternatives-e39ff57d)
24. [Cloudflare: Clef decision models](https://blog.cloudflare.com/clef-decision-models/)
25. [GitHub: PostHog/jeeves](https://github.com/PostHog/jeeves)
26. [Privatemode: System One from GLM Flash](https://www.privatemode.ai/blog/system-one-from-glm-flash)
27. [OpenAI: DevDay 2026 recap](https://openai.com/index/devday-2026-recap)
28. [uxdesign.cc: How to de-slopify your designs](https://uxdesign.cc/how-to-de-slopify-your-designs-4c40c57c1dc9)
29. [X: Karpathy: four ways to make an explanation land](https://x.com/karpathy/status/2105819303471976479)
30. [unsung.aresluna.org: A loading state within a loading state](https://unsung.aresluna.org/a-loading-state-within-a-loading-state/)
31. [hitesh.in: Google has the best AI money can buy; its products still feel broken](https://hitesh.in/2026/best-ai-still-feels-broken/)
32. [The New York Times: Anthropic and the religious thinkers teaching Claude right from wrong](https://www.nytimes.com/2026/09/29/us/anthropic-claude-morals-ai.html)
33. [X: Sam Altman on religious power assigned to AI models](https://x.com/sama/status/2106388373221118198)
34. [Axios: Altman's critique read as a veiled shot at Anthropic](https://www.axios.com/2026/10/03/openai-anthropic-altman-amodei-religious-force-models)
35. [X: Pope Leo XIV on AI art and the spark of humanity](https://x.com/Pontifex/status/2105983638147891490)
36. [arXiv: MCMC trick makes supervised finetuning generalize better than reinforcement learning](https://arxiv.org/abs/2610.02140)
37. [arXiv: RL training trick cuts decoding compute 2.5x by trusting the critic](https://arxiv.org/abs/2609.39247)
38. [arXiv: Wavestone: 11 coding agents, one seven-part blueprint](https://arxiv.org/abs/2609.00006)
39. [TechCrunch: Factory CEO accuses VC board adviser of spying for Cognition](https://techcrunch.com/2026/09/30/factory-ceo-just-accused-his-vc-board-advisor-of-spying-for-cognition/)
40. [FT: TypeSafe's jump from stealth to a reported $10B round talks](https://www.ft.com/content/456884ea-2558-4648-8036-a77b73733430)
41. [TechCrunch: Viral AI agent Instinct raises $1B at a $10B valuation](https://techcrunch.com/2026/09/28/viral-ai-agent-instinct-raises-1b-series-c-at-a-10b-valuation/)
42. [The Deep View: Runway CEO sees a much bigger future for video AI](https://www.thedeepview.com/articles/runway-ceo-sees-a-much-bigger-future-for-video-ai)
43. [Runway: Introducing Runway Ads](https://runway.com/news/company-news/introducing-runway-ads)
44. [Runway: Introducing Praxis-1](https://runway.com/research/introducing-praxis-1)
45. [Runway: Introducing Solaris](https://runway.com/news/research/introducing-solaris)
46. [Amazon: Built Together](https://www.aboutamazon.com/news/company-news/amazon-data-centers-built-together)
47. [The Deep View: How Amazon launched a data center charm offensive](https://www.thedeepview.com/articles/how-amazon-launched-a-data-center-charm-offensive)
48. [TechCrunch: Apple tightens macOS Full Disk Access over agent risks](https://techcrunch.com/2026/10/02/apple-says-its-tightening-macos-full-disk-access-controls-due-to-new-risks-from-ai-agents/)
49. [The Information: Nvidia, SoftBank make final $20B investment in OpenAI's last round](https://www.theinformation.com/briefings/exclusive-nvidia-softbank-make-final-20-billion-investment-openais-last-round)
50. [Business Times: US lead over China narrows to 3%, per Bloomberg Intelligence](https://www.businesstimes.com.sg/international/us-lead-ai-over-china-narrows-after-deepseek-gains-report)
51. [Hugging Face: Aleph-Alpha/Kolibri-1](https://huggingface.co/Aleph-Alpha/Kolibri-1)
52. [GitHub: blendi-remade/agentcraft](https://github.com/blendi-remade/agentcraft)
53. [Amazing Glass](https://tomacco.github.io/amazing-glass/)
54. [Suno: Introducing Speech (beta)](https://suno.com/blog/introducing-speech-beta)
55. [The Deep View: New voice model shows Microsoft's AI strategy](https://www.thedeepview.com/articles/new-voice-model-shows-microsoft-s-ai-strategy)
56. [GitHub: trycua/cua](https://github.com/trycua/cua)
57. [Black Forest Labs: FLUX 3 Image](https://bfl.ai/models/flux-3-image)
58. [Shopify: Introducing Canvas](https://www.shopify.com/news/introducing-canvas)
59. [X: Google Stitch: CLI for local coding agents](https://x.com/stitchbygoogle/status/2105695155164741984?s=20)
60. [GitHub: earendil-works/pi (durable)](https://github.com/earendil-works/pi/tree/main/packages/durable)
61. [Axiem Systems: offline photo editing](https://axiemsystems.com/press/)
62. [GitHub: 0xpili/simplified-technical-english](https://github.com/0xpili/simplified-technical-english)

Sources today: [AlphaSignal](https://alphasignal.ai/), [The Neuron](https://www.theneurondaily.com/), [Sidebar](https://sidebar.io/), [The Deep View](https://www.thedeepview.co/), [The Strange Review](https://thereview.strangevc.com/).