1. Deep Moats and Platform Shifts in Computing - Part 3 — Pushkar Ranade

  • Why read: NVIDIA has the same ecosystem pull Intel had in the x86 era, but the barriers to entry are higher this time.
  • Summary: AI hardware is following the same path as the 90s CPU wars. Efficiency and clean design usually lose to scale and developer lock-in. Switching costs are high because AI code is harder to port than standard software. NVIDIA stays ahead by moving faster and spending its massive margins on R&D. The "Wintel" equivalent in AI is the NVIDIA and CUDA combination, and it has a tight grip. Hardware fragmentation is unlikely to fix the cost problem soon.
  • Read more

2. The Eternal Sloptember — George Hotz (geohot)

  • Why read: A warning that AI agents might flood professional engineering with "statistical slop" instead of good code.
  • Summary: AI agents in software development might be a mistake because they produce code that looks right but lacks understanding, which hides bugs. They are great for prototypes but fail at finishing the job. Large companies risk drowning in mediocre code as low performers use agents to churn out volume. High performers will still need to check every line. We might be heading toward a period where craftsmanship is buried under low-quality agent output.
  • Read more

3. The Sensemaking Series: How to Make Sense of AI — Cedric Chin

  • Why read: Use a military framework to track tech shifts without getting caught in the hype cycle.
  • Summary: Sensemaking is the effort to understand things during high uncertainty. Experts use mental models to explain data but stay ready to pivot when something doesn't fit. Instead of just updating a single view, try running two parallel models at once. This keeps you grounded during AI shifts. Look for specific case studies from past tech changes to build a better filter for what matters.
  • Read more

4. Infinite Demand for Intelligence — Chris Anderson

  • Why read: Intelligence might be a commodity where demand never stops growing.
  • Summary: Unlike most things we buy, we might never have enough intelligence. Lower prices just create new ways to use it. In traditional economics, markets reach a limit, but intelligence might keep expanding as it opens new markets. There is no "peak AI" because every gain in intelligence creates more problems that need solving. It is the primary asset for everything we build next.
  • Read more

5. Consulting Firms and the AI Bottleneck — Raphaël Dabadie

  • Why read: How AI-native firms are starting to eat the lunch of high-priced consultants at McKinsey and BCG.
  • Summary: Big consulting firms are in trouble because clients don't want to pay human rates for work that AI can do in a fraction of the time. Most firms are just trying to help their staff work faster, but the real shift is redesigning the work for agents. Agents can handle the research and logic while humans act as final reviewers. This moves the business model from selling hours to selling results powered by agent workflows.
  • Read more

6. Agents Want Flexible Schemas: The NoSQL Comeback — Gordon Brander

  • Why read: Why Markdown with Frontmatter is becoming the preferred format for AI-driven data.
  • Summary: Computers need rigid rules, but LLMs work best with free-form text. Markdown with Frontmatter solves this by giving computers the structure they need while letting the AI work in natural language. This lets agents build their own models as they go without breaking the system. For product builders, the future of data looks more like organized notes than strict databases.
  • Read more

7. The Agentic ROI Gap in 2026 — SuspendedCap

  • Why read: A check on whether AI agents are actually making money or just wasting compute.
  • Summary: Five months into the agent era, businesses are spending big on tokens but are still looking for real ROI. Sustainable growth only happens when the math works, like spending $5 to get $20 in value. The real win is when AI cuts operational costs by 30% or fixes insurance pricing. The next year will show which tools are useful and which are just toys.
  • Read more

8. The Outlier Quotient (OQ) for Founder Success — Adam Shuaib

  • Why read: Success is tied to "deviation from the mean" rather than elite credentials.
  • Summary: A study of 15,000 founders shows that success is tied to being an outlier. Look for people with odd hobbies, early wins, or unusual life choices. Standard VC filters often ignore these people because they prioritize conformity. To find the best talent, look for the people the traditional system usually rejects.
  • Read more

9. The Dead Economy Theory — Owen McGrann

  • Why read: The AI industry's massive valuations only make sense if they replace human jobs at scale.
  • Summary: AI labs are valued so high because they are betting on taking over the labor market. High-end roles like law and medicine are the primary targets. If AI doesn't actually remove human costs, these companies are overvalued. The "copilot" label is just a friendly way to sell labor replacement. Be ready for the pressure to move from augmentation to replacement.
  • Read more

10. Dexter: Autonomous Financial Research Agent — virattt

  • Why read: A concrete example of an autonomous agent that handles deep financial research.
  • Summary: Dexter breaks down hard financial questions into research plans and pulls real market data to answer them. It checks its own work and avoids getting stuck in loops. This is a move toward specialized associates that do the heavy lifting of gathering data. It provides a template for how to build agents that use specific tools for specific jobs.
  • Read more

11. Did the Customer Ever Ask for This? — Patrick Berzai

  • Why read: Most AI features are built for investors, not for the people using the products.
  • Summary: Many AI moves are driven by the need to satisfy boards and hit valuation targets. This often leads to a worse experience for customers, like useless chatbots on websites. Real market value comes from what people actually want to use and pay for. Product leaders should check if they are building for their users or for their next funding pitch.
  • Read more

12. Morning Taste, Afternoon Shipping — Carlos E. Perez / Garry Tan

  • Why read: A simple daily schedule for people working with AI.
  • Summary: High performers split their day. Use the morning for strategy, invention, and setting the quality bar. Use the afternoon to manage agents, iterate, and ship. This keeps the human in charge of quality while the AI handles the speed of execution. By separating intent from work, you can stay fast without losing quality.
  • Read more

13. The Scoreboard Doesn't Care — Kahlil Lalji

  • Why read: In a world of risky startups, your reputation is the only thing that lasts.
  • Summary: Startups are judged on results, not stories. While a company might fail, how you treat people during the process stays with you. Investors back founders who are honest and take accountability when things go wrong. Reputation is built by doing the right thing when there is no guarantee of a win.
  • Read more

14. Staying Creative During Travel — David Hoang

  • Why read: How to use travel constraints and boredom to come up with better ideas.
  • Summary: Travel forces you to improvise. Disconnecting on a flight can trigger a state of boredom that leads to new ideas. Use simple tools to capture notes on the go, then use AI to organize them later. This makes travel a source of new ideas rather than a break from work.
  • Read more

15. The AI Data Center Stack — Ali Afridi (SandHill.io)

  • Why read: The latest trends in AI infrastructure and companies that don't sell software.
  • Summary: The best companies might stop selling software and start selling AI-driven results. Infrastructure, power, and data centers are becoming the primary competitive advantages. We are seeing the rise of AI service firms that use agents to deliver work at scale. The next big opportunity is in the physical layer and robotics.
  • Read more

Themes from yesterday

  • ROI Reality Check: Moving from AI hype to demanding clear cost savings and results.
  • AI-Native Structures: Markdown becoming a data standard and new workflows like "Morning Taste/Afternoon Shipping."
  • Human Moats: Sensemaking and reputation are the only things AI can't replace.
  • Hardware Realities: Hardware and infrastructure are the biggest barriers to entry in the market.