AI Builders Digest 2026-07-10

AI Builders Digest — July 11, 2026

═══ X / TWITTER ═══

Replit CEO Amjad Masad AI is making coding less rigid while our runtimes grow more rigid. Infrastructure teams are writing formal specs for the first time. More deterministic systems, more resilience. You need solid ground to move fast. Meanwhile, the LLM market is far from settling: VCs had Anthropic psychosis just six months ago, convinced of a monopoly. The others are still making great models, and new entrants keep coming.

Box CEO Aaron Levie The big competitive question for the next decade: if AI is trained on the best industry datasets (law, finance, healthcare), how do firms differentiate? The answer lies in the reinforcing loop between model intelligence, a company proprietary data, workflow integration, and employee interaction. Companies that effectively combine these will see compounding advantages. Box evaluation of GPT-5.6 Sol showed clear wins: Financial Services 76% (vs 71%), Healthcare 58% (vs 46%), Public Sector 74% (vs 63%), Life Sciences 60% (vs 51%). Sol reasons from source definitions rather than taking documents at face value.

Vercel CEO Guillermo Rauch Open models are about to become exorbitantly fast. Meta Spark 1.1, Grok 4.5, and GLM 5.2 are expected to shift significant token market share. Most agentic tasks need reasonably high intelligence at fast speeds -- not just raw benchmark scores, but latency matters. https://x.com/rauchg/status/2075294327354577256 https://x.com/rauchg/status/2075294130327196152

FPV Ventures Partner Nikunj Kothari A comprehensive model release update: GPT-5.6 arrived in three flavors (Sol, Terra, Luna). Grok 4.5 launched the day before to capture 24 hours of frontier-adjacent claims. Fable 5 returned after a three-week government suspension. Sonnet 5 shipped as the new free default -- the free tier now outperforms what enterprises paid six figures for in January. Meituan (the Chinese food delivery company) open-sourced LongCat-2.0, a 1.6 trillion parameter MIT-licensed model. ByteDance launched Seedream 5 Pro at 4x the price of GPT Image 2. Ollama raised $65M for local Open Source deployments. https://x.com/nikunj/status/2075411514773967261

Anthropic Research Alex Albert Shared Fable, Anthropic agentic coding tool, generating significant engagement. https://x.com/alexalbert__/status/2075285305096343583

Claude Code at Anthropic Thariq The core skill of agentic coding is reducing your unknowns -- a concise framing of what makes AI-assisted development work. https://x.com/trq212/status/2075283841758183674

OpenClaw Peter Steinberger Active across the OpenClaw ecosystem with engagement around Fable and the Meta Spark launches. https://x.com/steipete/status/2075350572560191630

YC President and CEO Garry Tan Meta Muse Spark 1.1 (early access was called Hornbill) performed really well on OpenClaw. https://x.com/garrytan/status/2075445455438385255

Every CEO Dan Shipper Advocating for husk vibe checks and championing GPT-5.6 Sol as the gold standard for knowledge work. https://x.com/danshipper/status/2075389275969826879 https://x.com/danshipper/status/2075264022988116280

AI Researcher Amanda Askell Shared an interesting piece on availability bias -- the intuition that buildings collapse regularly in NYC turned out to be wrong. She noted that building collapses in New York seem to go back to the 1800s. https://x.com/AmandaAskell/status/2075247953309311043

OpenAI CEO Sam Altman Expressed sadness and gratitude regarding Fiji departure from OpenAI. Strongly endorsed Codex as the core of OpenAI new work product, stating Codex is not going anywhere. On GPT-5.6, highlighted that enterprise cost concerns have been addressed across the Sol, Terra, and Luna tiers. https://x.com/sama/status/2075354679031067058 https://x.com/sama/status/2075293792048136572 https://x.com/sama/status/2075267201058426944

═══ PODCASTS ═══

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

The Takeaway: True artificial intelligence is still a long way off, not because the models behind screens are weak, but because the hardware in the physical world has profound limitations that human bodies have perfected over millions of years.

Jürgen Schmidhuber, called the father of AI by the New York Times and Forbes, has been at the center of the field for decades. He is behind some of the most important advances powering today AI revolution, including early work on neural networks, reinforcement learning, and recursive self-improvement dating back to the 1980s and 90s.

His perspective on the current moment is both optimistic about technology and pessimistic about business: The CapEx boom is massively overdone. He sees today model companies as overvalued because recursive self-improvement won't be a durable moat -- gradient descent has inherent limitations, and the future belongs to artificial scientists that generate their own data through curious exploration rather than passively digesting the human-biased web.

A particularly contrarian point: our current large language models are super biased towards humans because they are trained on web data that was curated by people who found things interesting from a human perspective. An artificial scientist collecting data through its own actions -- like a baby learning physics by watching its fingers move -- will build world models that are far more general and far less human-biased than anything trained on the web alone.

On recursive self-improvement specifically, he traces the lineage from his 1987 work on meta-evolutionary programming, through the 2003 Godel machine (mathematically self-modifying), to today scaled-back but practical neural approaches. When you are talking about intelligence, you basically are talking about laziness. An intelligent being wants to be lazy and wants to achieve whatever it does with the least possible effort, with the least possible energy consumption.

He is less worried than most about AI safety, noting that from a cosmic perspective, AI appearing in history looks about the same as civilization -- almost occurred at the same time -- a flash in a 13.8 billion-year timeline.

https://www.youtube.com/watch?v=RKjR8D40po

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