Deep Currents 07.07.26

Reading the Currents

Deep Currents started in March 2025 for an unglamorous reason: I was already tracking every AI release that mattered for my own use, and it felt like a waste not to share my notes. Doing it monthly, on a fixed schedule, was as much a discipline exercise as an editorial one. I also suspected that tracking this space, one release at a time, would be worth doing simply because of how fast the ground was moving under everyone's feet. Sixteen editions later, I think that suspicion held up, though not quite in the way I expected.

The earliest editions read like field notes: Claude 3.7 Sonnet's debut as a model built to blend quick answers with extended step-by-step thinking, GPT-4.5's release, an image-generation season so frenetic that Ghibli-style memes briefly broke ChatGPT. By June that year the format had settled into a shape I kept for most of the run: company names bolded with links to original sources, categories organized by function, and a closing section where I tried to make sense of whatever I'd just listed. That section, "What This All Means," became the place where the real insights of each post were laid out, and by December, reader feedback convinced me to move it to the top. I also renamed the section "Reading the Currents," and labelled the exhaustive list "The Full Stream."

The throughline across all sixteen editions is a steady march from novelty toward infrastructure. Tools that felt like magic tricks in the spring of 2025, including Runway's first consistent-character video model, and Suno and Udio's first genuinely capable song generators, matured into standard production tools within the year, and the fights over compensation that followed (Disney and Universal suing Midjourney, then ByteDance, while UMG and Sony sued Suno and Udio) mostly settled into licensing deals rather than lawsuits. Vibe coding, letting an AI agent write and run code from a plain-language description, followed the same arc, turning from a meme into Cursor, Claude Code, Codex, and a dozen competitors running unsupervised, multi-hour sessions inside real companies. The biggest single moment was the release of OpenClaw, the open-source agent that picked up GitHub stars faster than any open source project in history. It showed what every lab has been racing to build since: something that keeps working, remembering, and acting on your behalf while you're doing something else.

The same stretch of time also turned this series into something closer to a business and policy beat than a pure product roundup. Anthropic's refusal to let the Pentagon use Claude for mass surveillance or autonomous weapons, and the standoff that followed, became one of the biggest stories I covered here it and eventually got its own essay, "Standing Ground". xAI's trillion-dollar merger with SpaceX and Meta's ten-day acquisition of Manus made the same point from the money side: consolidation, not invention, was driving the next big phase. By this June, with three of the biggest labs racing toward IPOs and the most capable models locked behind the kind of restricted access usually reserved for export-controlled technology, exactly the two-tier split I'd predicted in"Reading the Meter" a few months earlier, I found myself writing "I'm not sure if I'm going to keep doing these" almost as an aside. But it wasn't just an aside, it was a hint.

I haven't stopped following any of this. What's changed is that I want some of my time back for other projects, and the industry itself feels like it's settled into a groove rather than a series of sprints. A year and a half ago, every month offered a major breakthrough or a genuine surprise. At this point new frontier models and powerful agentic tool releases are the norm, and the novelty is wearing off. Covering the shift monthly, in real time, was the whole point of this series, and there's something fitting about its own subject becoming so routine, and dare I say predictable, that it no longer feels necessary to keep the series going. I might be wrong about that. A big enough breakthrough could have me back at it in a few months. We'll see.

The Full Stream

If you want to keep tracking this stuff yourself, here's where I'd start. Instead of everything that shipped this month, I'm sharing my reading list: all the newsletters, blogs, and the people and companies I follow, that made every edition of this series possible in the first place.

Daily digests

• The Rundown AI. Daily digest of AI news plus a practical how-to tip in every issue.
• Superhuman AI. Similar daily format with a broader tech and science lens, robotics and space and biology, AI woven through rather than the sole focus.
• Superhuman AI: The Code. The coding-focused sibling edition, aimed at developers shipping with AI tools.
• The Neuron. Daily AI news with a chattier, meme-heavier voice than The Rundown.
• Ben's Bites. Ben Tossell's long-running daily/twice-weekly roundup, one of the originals in this format.
• Limited Edition Jonathan. Practical Claude and agent-workflow tips, usually tied to whatever just shipped. The subject lines lean clickbait; the content underneath holds up.

Research, policy & frontier analysis

• Import AI. Jack Clark's weekly newsletter on frontier AI research, read for the paper digest and for his own essays on where things are heading.
• Interconnects. Nathan Lambert on open-model research and the politics of open versus closed AI. Dense, and unusually candid about incentives.
• Hyperdimensional. Dean W. Ball on AI governance, written from inside the Washington conversation.
• Last Week in AI. Weekly news roundup with a companion podcast edition.
• Center for Humane Technology newsletter. Occasional dispatches on AI harms and the attention economy from the Tristan Harris and Aza Raskin org.

Business, product & the future of work

• Every. Essays on AI and knowledge work, plus their model reviews. Consistently the best-written of the business-facing newsletters.
• Lenny's Newsletter. Product management newsletter gone heavily AI-inflected, with two spin-off editions worth following too: How I AI and Community Wisdom.
• The Leverage. Tech and VC newsletter, AI-adjacent rather than AI-first, covering the money side of the boom.

Design, UX & AI

• The Shape of AI. Emily Campbell's UX pattern library and newsletter for AI product design. Probably the most-cited resource on this whole list among actual designers.
• UX for AI. Greg Nudelman shares insights based on his years of experience designing AI-driven products. He also offers training and wrote a book on it.
• Jakob Nielsen's newsletter. The UX-research veteran on AI and usability, plus a recurring "UX Roundup" digest.
• Design with AI. Practical, tool-focused tips for using AI inside design workflows.
• Peter's Context Design. Design and AI essays with a personal-essay voice.
• Moon Learning. Numbered-issue newsletter on designing for AI, adaptive UI and agentic design for practicing designers.
• Stephanie Walter Design. Weekly "Pixels of the Week" link roundup, UX-first but AI shows up most weeks.
• Design Better. Aarron Walter and Eli Woolery's design podcast and newsletter. Not AI-exclusive, but AI comes up constantly in guest conversations.
• Slot Machine. TJ Pitre on design systems, AI tooling, and web components. Full disclosure: TJ's a friend.

AI & creativity

• Creativity AI. Weekly roundup on generative art, video, and music tools, currently in its 80s by issue number.
• ArtisanOS Alpha. Rob Hestand on AI and the creative and entertainment industries, more essayistic than most tool-roundup newsletters.
• Geeky Curiosity. Newsletter for creators who want to master AI art creation, with a free weekly roundup and otherwise mostly gated content.

Adjacent / general tech

• Semafor Technology. General tech-industry briefing, not AI-specific, but AI is a constant thread.

Blogs I routinely visit

• Anthropic news. Anthropic's company newsroom: research, policy, partnerships, the broader company story.
• Claude blog. Narrower than the newsroom above, product news and best practices specifically for people building with Claude.
• Dario Amodei's blog. His personal essay site, separate from the company channels. Infrequent, but this is where the big-swing pieces land, "Machines of Loving Grace," "The Urgency of Interpretability," "The Adolescence of Technology."
• Google DeepMind. Research and product announcements straight from the lab behind Gemini and Genie.
• OpenAI news. Official announcements: model releases, research posts, policy statements.

Industry leaders and AI luminaries I follow

• Sam Altman. CEO of OpenAI, the account that moves markets when it posts.
• Mira Murati. Ex-OpenAI CTO, now building Thinking Machines.
• David Holz. Founder of Midjourney, one of the few lab heads who still posts like a person.
• Yoshua Bengio. Turing Award laureate, one of the three godfathers of deep learning, increasingly focused on the risks of the field he helped build.
• Yann LeCun. Turing Award laureate, Meta's outgoing Chief AI Scientist, and the most reliable "here's why that's wrong" voice among the three, especially on AGI timelines.
• Andrew Ng. Co-founded Coursera and Google Brain, now runs DeepLearning.AI. The most practical and least doom-inflected of the pioneers.

AI safety, policy & governance people

• Jan Leike. AI safety researcher now at Anthropic, ran OpenAI's superalignment team before that.
• Miles Brundage. Ex-OpenAI policy lead, now executive director of AVERI, a nonprofit pushing for real third-party auditing of frontier models.
• Sam Bowman. AI safety researcher at Anthropic, on leave from a faculty job at NYU.
• Andrew Critch. AI safety researcher and CEO of Encultured AI.
• Peter Wildeford. Superforecaster turned AI policy analyst, focused on catastrophic-risk odds.
• Daniel Eth. AI alignment commentary, sharper and funnier than the average policy account.

Public-facing researchers & critics

• Gary Marcus. Cognitive scientist and the loudest mainstream AI skeptic, useful ballast against lab hype.
• Ethan Mollick. Wharton professor, wrote Co-Intelligence, runs One Useful Thing.
• Melanie Mitchell. Santa Fe Institute, writes AI Guide, a grounded, skeptical-of-hype research voice.
• Margaret Mitchell. AI ethics researcher, now at Hugging Face after Google and Microsoft.
• Meredith Whittaker. President of Signal, chief advisor to the AI Now Institute.
• Rodney Brooks. Robotics and AI pioneer at MIT, reliably deflates overclaiming.
• Simon Willison. Independent researcher, one of the best plain-language explainers of what's actually happening under the hood.
• Thomas Wolf. Co-founder of Hugging Face.
• Soumith Chintala. Co-founded and leads PyTorch at Meta.
• Jeremy Howard. Co-founder of fast.ai and Answer.ai, a longtime voice for practical, accessible machine learning.
• Nathan Lambert. Writes Interconnects, listed here as a person as well as a newsletter, since he's as worth following as reading.

AI ethics & accountability researchers

• Dr. Abeba Birhane. AI accountability researcher, founder of the AI Accountability Lab.
• Kate Klonick. Law professor and journalist covering tech geopolitics and AI governance.
• Rumman Chowdhury. CEO of Humane Intelligence, former US Science Envoy for AI.

Builders & the AI-engineering scene

• Anton Osika. Founder of Lovable, building at the edge of AI-assisted software.
• swyx. Latent Space podcast, AI Engineer conference, a hub for the builder side of the conversation.
• Kyle Lo. Co-led Olmo at Ai2, now deep in language model data and evals.

AI art & creative practice people

• Dreaming Tulpa. Creative coder working the AI-art edge, worth following for process as much as output.
• Min Choi. One of the more useful AI-tools curators, less hype than most in that lane.

Anthropic folks

• Andrew Lampinen. Researcher on cognition and AI, ex-DeepMind and Stanford cognitive science.
• Meaghan. Leads design for Claude Code and Cowork.
• Jenny Wen. Designer at Anthropic, another design-first vantage point on how Claude actually gets built.
• Boris Cherny. Created Claude Code, posts daily setups and workflows straight from the source.
• Andrej Karpathy. Ex-OpenAI founding member and ex-Tesla AI director who joined Anthropic's pretraining team this year. Coined the term "vibe coding."

Tech & culture journalism

• Taylor Lorenz. Tech and internet-culture journalist, runs the UserMag newsletter and Power User podcast.

Institutions, companies & labs

People come and go from these; the institutions are often the actual primary source once you strip away everyone's commentary on them.

• OpenAI. The lab whose releases still set the news cycle.
• Claude. Anthropic's product account.
• Cursor. AI coding editor, one to watch for the agentic-coding side of things.
• Midjourney. Still the reference point for AI image generation.
• Runway. AI video generation.
• Replit. AI-assisted app building, idea to deployed app.
• Udio. Generative music.
• Hume AI. Empathic AI research lab, emotional intelligence in models.
• Topaz Labs. AI image and video upscaling.
• FLORA. Generative-workflow creative environment.
• AVERI. Nonprofit pushing for real third-party auditing of frontier models.

That's the list. If I've missed anyone or anything that you think I should have included, please let me know! Thanks for reading along for sixteen months of this. See you in the next one, whenever that is.

This post was written with the assistance of Claude Sonnet 5. Feature image generated with Midjourney 8.2 Preview