AI Builders Digest 2026-07-20

AI Builders Digest — July 20, 2026

=== X / TWITTER ===

Swyx (swyx on X) — AI Engineer & Latent Space Podcast host Europe has top-tier AI engineers that the global arena overlooks. We're running the most competitive global arena for AI talent and it will be interesting to see how this plays out as WF enters the eval window. https://x.com/swyx/status/2078628617987518855

Thibault Sottiaux, Codex & ChatGPT @OpenAI Thibault is using ChatGPT Work as his primary interface — dictating into it constantly and delegating tasks. He processed thousands of DMs mentioning ChatGPT Work by simply dictating the instructions. ChatGPT Work now handles sites, emails, documents, spreadsheets, and slides, included in Plus/Pro/Business/Enterprise plans. His approach of delegating to ChatGPT Work is becoming his full-time job. https://x.com/thsottiaux/status/2078702412085498087 https://x.com/thsottiaux/status/2078697741455356367 https://x.com/thsottiaux/status/2078697631019303273

Guillermo Rauch, @vercel CEO Guillermo shared internal evals showing Kimi K3 is top-tier at cybersecurity and Sol is a leap ahead in cyber capability (at higher cost but remarkable). Fable remains restrictive, refusing to complete runs. He also wrote a long-form post arguing that "AGI" has aged poorly — AI is far better than human intelligence for most economically-relevant tasks, but AI can't replace the "proverbial you." Quality and humanity will prevail in the face of AI-generated content. https://x.com/rauchg/status/2078647648307880209 https://x.com/rauchg/status/2078548458714406959

Aaron Levie, CEO @Box Aaron outlined the future of the AI ecosystem beyond the frontier labs. He sees five key layers diffusing AI: companies tuning models for specific use cases, applied AI delivery tools, new vertical labs (life sciences, finance, healthcare), new infrastructure for running models and agents, and new services firms driving enterprise change management. China has crossed the rubicon to compete at near-frontier levels, and the solution is to keep a high rate of progress and drive diffusion. https://x.com/levie/status/2078567715544121815 https://x.com/levie/status/2078481578779685245

Zara Zhang (zarazhangrui on X) — Builder Zara makes a practical point: everyone should develop their own "personal eval set" for AI models — a few tasks relevant to your day-to-day work/life. Industry benchmarks help but don't reflect what makes a model actually useful to you personally.

Zara also identified the biggest barrier for enterprise AI adoption: people who understand AI don't understand the business, and vice versa. https://x.com/zarazhangrui/status/2078666187026911488 https://x.com/zarazhangrui/status/2078492577788268549

Matt Turck, VC @FirstMarkCap In three consecutive years the prediction has been "The model layer is commodotizing" (2024, 2025, 2026) and Matt observes it is still not commodotized. https://x.com/mattturck/status/2078520552680046920

@trq212 (Thariq, Claude Code @anthropicai) Thariq praised the Anthropic team's heroic effort behind the Fable launch, noting it was "not at all clear that we'd be able to do this in time." The post received over 6,700 likes with 146 retweets. https://x.com/trq212/status/207851418006864

Peter Yang (petergyang on X) — Practical AI tutorials Peter and his 8-year-old built a ChatGPTapp site to help her learn multiplication tables, using ChatGPT Images for UI/characters, adding music and a timed boss level. https://x.com/petergyang/status/2078638568784994686

Swyx (swyx on X) Swyx noted that AEO (AI Engagement Officer) will be fully responsible for M in his revenue next year. https://x.com/swyx/status/2078581967768166591

=== PODCASTS ===

Unsupervised Learning — Ep 90: AI Pioneer Jürgen Schmidhuber on the State of AI Today

The Takeaway: The future of AI lies in artificial scientists that collect their own data through experiments, rather than being trained on the human-biased World Wide Web.

Juergen Schmidhuber, set as the father of AI by the New York Times, Forbes, and more, shares perspectives on what is missing in today's models, why today's CapEx boom is overdone, and why he is optimistic on AI technology but pessimistic on the model companies.

Schmidhuber makes the case that current LLMs are "super biased towards human" because they are trained on web data that was created by people who found it interesting from a human perspective. An artificial scientist living in an unknown environment and building world models through its own actions will create data that is "much less human biased." He calls this "artificial curiosity" — a concept he proposed in 1990.

On recursive self-improvement, he argues that the most popular current approach (gradient descent-based weight modification) is limited but works well in practice. His 2003 Godel machine offered a mathematically optimal approach, but it was less practical.

On the AGI timeline question, he gives a cosmic perspective: from a 13.8 billion year world history, civilization's 13,000 years of innovation was just a flash, and so too will AI appear as a flash in hindsight. "A couple of hundreds of years ago, the first calculators... and now you can calculate much more than a hundred years ago for the same price. And then suddenly, it was there."

On robotics: "Robot hardware is really inferior compared to human bodies, and there's no human made technology that compares to this hand. It seems a lot more than a hardware problem. But you can't have AGI without hardware like that."

https://www.youtube.com/watch?v=RKjR8DQ40po

Generated through the Follow Builders skill: https://github.com/zarazhangrui/follow-builders