1. [AINews] OpenAI launches GPT 5.6 Sol/Terra/Luna, Codex becomes ChatGPT superapp — Substack

  • Why read: Catch up on OpenAI's new model lineup and its push into the enterprise workspace.
  • Summary: OpenAI launched the GPT-5.6 family (Sol, Terra, and Luna) to compete with Claude. The flagship Sol model scores high on coding and agent benchmarks, driven by an "ultra" setting that manages parallel subagents. OpenAI also merged ChatGPT Work and Codex into a single desktop app. The release lowers developer costs while giving users better tools to orchestrate tasks.
  • Read more

2. [AINews] not much happened today — Substack

  • Why read: Look at the early user friction and UX mistakes from the GPT-5.6 rollout.
  • Summary: Early GPT-5.6 users struggled with complex settings and confusing defaults. After complaints about unexpected limits, OpenAI quickly reset usage caps and promised UI rollbacks. Users like the model's orchestration and computer use, but hidden costs from background subagents are causing billing headaches. The backlash shows how hard it is to design clear interfaces for complex agent systems.
  • Read more

3. Clouded Judgement 7.10.26 - Own Your Weights — Substack

  • Why read: Why companies should own their AI training pipelines instead of static model weights.
  • Summary: Static model weights lose value quickly as newer models arrive. Real cost control comes from owning the entire process, including data, reinforcement learning, and evaluations. Fine-tuning open models on specific tasks often beats generalist models on performance and inference cost. But managing this infrastructure is hard, opening the door for startups focused on model governance and version control.
  • Read more

4. We Forgot to Talk About Governance. Our Bad. — Substack

  • Why read: A look at the financial and operational risks of running AI agents without strict rules.
  • Summary: Companies are deploying autonomous agents fast and ignoring governance. Unsupervised agents can delete data or rack up huge token bills from minor errors. To fix this, every agent needs a "constitution" that sets hard limits on actions like spending money or deleting files. Operators need to plan for failure to keep agents from becoming expensive mistakes.
  • Read more

5. Congratulations! You are now the Head of AI — Substack

  • Why read: A guide for managers trying to push AI adoption in their departments.
  • Summary: If you manage a team today, you are effectively in charge of AI integration. Start by building your own agents to learn the tech, then write a team manifesto to set expectations. Spend the first three months automating repetitive work to free up time. To make it stick, give your team protected time to experiment and tie AI use to performance metrics.
  • Read more

6. Inside Docusign's Enterprise Credit Model — beehiiv.com

  • Why read: How and why Docusign moved from per-user SaaS pricing to a consumption-based credit system.
  • Summary: Docusign switched its enterprise agreement product to a pure credit-based model. Scrapping user limits removes the friction of seat management and ties pricing directly to usage. This encourages wider adoption across a company and makes it easier to charge for AI features as they are used. It offers a useful template for SaaS companies trying to fit AI into old pricing structures.
  • Read more

7. Three Years In — tomtunguz.com

  • Why read: How fast AI growth and specialized infrastructure are changing venture capital.
  • Summary: AI startups are hitting revenue milestones faster than previous generations. This speed is blurring VC funding stages and driving massive early rounds. Meanwhile, focus is shifting from foundation models to specialized inference infrastructure for batch, real-time, and local processing. High inference costs are also pushing companies toward AI-native advertising to subsidize user growth.
  • Read more

8. RAM Fever — Substack

  • Why read: Why rising memory chip prices will squeeze margins across the AI hardware industry.
  • Summary: AI data center demand has nearly tripled DRAM costs in six months. Samsung and SK Hynix are using their pricing power to force double-digit hikes. These costs act as a tax on anyone scaling compute without hyperscaler volume discounts and will soon hit consumer electronics. Hardware and compute-heavy companies need to budget for these component spikes now.
  • Read more

9. Weekly Dose of Optimism #201 — Substack

  • Why read: The latest on advanced nuclear reactors being built to power AI data centers.
  • Summary: Tech hyperscalers are funding a boom in advanced nuclear energy to power their data centers. Four companies hit criticality ahead of 2026 targets, turning a science project into a manufacturing race. Data centers are the perfect early customers, offering enough capital to bypass traditional utilities and build new domestic supply chains.
  • Read more

10. California Will Tax SaaS?! | A Massive Change for Tech Companies — Substack

  • Why read: An early warning about new software taxes hitting California tech companies in 2027.
  • Summary: Starting January 2027, California will add a 7.25% or higher sales tax to prewritten software and AI services. This hits both internal tech budgets and customer invoices. Finance teams need to plan for the cost increase and track tax nexus carefully, as higher prices could drive customer churn.
  • Read more

11. Hello, may I scan your chats real quick? — Substack

  • Why read: An update on the EU's push to scan private messages under Chat Control legislation.
  • Summary: The EU narrowly extended Chat Control 1.0, letting tech platforms voluntarily scan private messages for illegal material until 2028. The move bypasses recent rejections and sets up a fight over mandatory scanning in Chat Control 2.0 this fall. While the current law protects end-to-end encryption for now, privacy advocates warn it normalizes warrantless surveillance of private chats.
  • Read more

12. The Second-Time Founder List VCs Can't Buy — So I Built It — Substack

  • Why read: How to use data scraping to find successful founders before they launch their next startup.
  • Summary: VCs want second-time founders who just finished their acquisition earnouts because they tend to outperform. Standard databases like LinkedIn and Crunchbase lag in surfacing them. By using AI to scrape acquisition histories, track vesting timelines, and check social signals, investors can find these founders exactly when they are ready to build again.
  • Read more

13. Why We Still Pay Human Designers in 2026 — Substack

  • Why read: An argument for hiring human designers despite cheap AI generation.
  • Summary: AI tools are great at churning out disposable assets for ad testing, but they struggle to create distinct brand identities. Real branding acts as a competitive moat. It takes personality and detail that AI still gets wrong. If the goal is long-term trust rather than a quick click, human designers are still required. Use AI for assets meant to be tested and broken; hire humans for assets meant to last.
  • Read more

14. How to Cash Out Your Early Backers Without Selling the Whole Company — Substack

  • Why read: A financial maneuver to buy out early startup investors without selling the company.
  • Summary: Software founders at $5M to $20M ARR often get pressured by early angels wanting an exit. Instead of selling the whole business, founders can do a majority recapitalization with private equity. This buys out early backers and adds new growth capital, while letting the founder keep operating control and a large equity stake. It cleans up the cap table and sets the company up for a bigger exit later.
  • Read more

15. NIO Just Gave Its Founder the Elon Musk Deal — Substack

  • Why read: A look at NIO's all-or-nothing CEO pay package designed to force massive growth.
  • Summary: Chinese EV maker NIO gave its founder a $1.2 billion share package modeled on Elon Musk's 2018 Tesla deal. The shares only vest if NIO hits a $120 billion market cap and $6 billion in net profit. By linking the payout entirely to extreme growth, the board aligns executive pay with shareholder returns. It highlights the pressure to scale in the global EV market.
  • Read more

Themes from yesterday

  • The Agentic Pivot: OpenAI's GPT-5.6 release shows the industry moving from chat interfaces to multi-agent workspaces. This transition is immediately causing UX friction and billing headaches.
  • Hardware & Energy Realities: Triple-digit jumps in DRAM costs and a rush to build advanced nuclear reactors show the physical limits and economic bottlenecks of scaling AI.
  • Maturing AI Operations: Companies are shifting from prototyping to strict operations. They are writing firm rules for autonomous agents, rethinking SaaS pricing around credits, and taking ownership of their training pipelines.