1. Microsoft climbs hills, Anthropic files first — Matt Slotnick

  • Why read: See how Microsoft is moving away from OpenAI dependency and how Anthropic's revenue is hitting record speed.
  • Summary: Microsoft is shifting its weight toward in-house models. They just launched seven "MAI" models, including one that goes head-to-head with Claude Opus. They are also using "Frontier Tuning" to help teams pick the right model for the job based on cost and speed. Anthropic, meanwhile, has filed for an IPO after reaching a $65B valuation. Their revenue run rate hit $47B in two quarters. The industry is moving from research experiments to production at scale.
  • Read more

2. How we automated technical implementation - Antimetal — Sai Naidu

  • Why read: A look at how to use software to handle customer onboarding instead of hiring more engineers.
  • Summary: Antimetal built a tool called "Anvil" that handles the technical setup for new customers. Instead of a basic tracker, it actually connects to production systems. It uses MCPs to run health checks and check customer logs automatically. This turns onboarding from a hiring problem into a software problem. For product teams, it shows how agents that "know" your infrastructure can get customers live much faster.
  • Read more

3. An executive IC is either the best idea I've had... — Jean-Michel Lemieux

  • Why read: A new way for senior leaders to drive results by getting back into the daily work.
  • Summary: Former Shopify CTO Jean-Michel Lemieux is testing a role he calls the "Executive IC." He joins sales calls and product reviews to give immediate feedback, skipping the usual management layers. Because of his experience, he can make quick calls on quality or code that others might avoid. It’s a way for veteran leaders to have an impact by coaching and shipping rather than just managing people.
  • Read more

4. Your AI strategy has a trust problem, not a tooling problem — Elena Verna via Lenny's Newsletter

  • Why read: Why your company's culture is likely slowing down your AI adoption more than the technology is.
  • Summary: AI only works if people have the power to use it without asking for permission every time. Many companies treat employees as risks to be managed, which creates bottlenecks. The goal should be "Employee Agency," where people own their workflows and make decisions locally. To win, leadership has to trust their best people to move fast and own their work end-to-end.
  • Read more

5. Your AI needs more context — Greg & Taylor

  • Why read: Prompting isn't the problem. You need to give your AI access to your actual work data.
  • Summary: AI gets more useful when it has access to Slack, email, and meeting notes. Engineering AI moved fast because code is easy to read, but business context is usually trapped in conversations. Connecting tools like Granola lets the AI see why decisions were made. This turns a basic chatbot into a partner that understands your daily reality and can handle tasks for you.
  • Read more

6. Why your open rates dropped — Jamal Reimer

  • Why read: Buyers are now using AI to block sales emails that sound like they were written by an AI.
  • Summary: Sales emails are getting buried by AI filters that spot generic language. To get through, you have to talk about the account's actual financial goals, not just insert a name or company. Generic subject lines don't work anymore. Sellers have to write high-signal messages that sound like the buyer's internal team and focus on the bottom line.
  • Read more

7. How to Stop Shipping Low-Quality RL Environments — Latent.Space

  • Why read: Why bad testing environments are the real reason many reinforcement learning models fail.
  • Summary: Models often learn the wrong things because the environments they train in have bugs or race conditions. A broken training harness ruins data and wastes money. The fix is to treat these environments like production software. Better data from better environments is the fastest way to improve model performance.
  • Read more

8. The Minimill of AI — Tomasz Tunguz

  • Why read: How to save money by using small, local models for easy tasks and saving the cloud for the hard stuff.
  • Summary: Tomasz Tunguz suggests a "Minimill" approach where a local router decides if a task is easy or hard. Local models handle about 80% of the work, which cut his queue times by 94%. This setup is faster and cheaper because you only pay cloud rates when you absolutely have to. Expect to see more of these local routers used to manage the costs of running agents at scale.
  • Read more

9. The Cannonball GTM Agent Landscape for Growth Leaders — Cannonball GTM

  • Why read: A simple guide for choosing which GTM agents to use and which to ignore.
  • Summary: Most sales stacks are messy, so growth leaders should focus on agents that handle specific tasks or interpret data. AI SDRs often fail if the target list is bad. Instead of a full system swap, start with small wins. The future is an orchestration layer that uses the CRM as an input rather than the only source of truth.
  • Read more

10. Tokens Over Humans: The AI Budget Fight Is Here — SaaStr

  • Why read: Engineering budgets are shifting from hiring new people to paying for AI tokens.
  • Summary: A new fight is happening over whether to spend money on headcount or tokens. Uber hit its yearly token limit in four months, leading to caps for developers. The best move is to use expensive models for research and switch to efficient, tuned models for production. Companies are now deciding if they trust agents to handle work that humans used to do.
  • Read more

11. Power Shifts (This Week in Stratechery) — Ben Thompson

  • Why read: Why Google is gaining on Microsoft and how YouTubers are beating Hollywood.
  • Summary: Google is catching up to Microsoft in market value, even with Microsoft’s head start in AI. At the same time, Gen Z creators on YouTube are making movies that perform better than old franchises. This shows that having direct distribution is more important than being a traditional gatekeeper. In AI, platform power now depends on how well you can reach users directly.
  • Read more

12. Deep Dive: How Close Are We to Living Longer? — Chamath Palihapitiya

  • Why read: Why we should treat aging as the root cause of most diseases.
  • Summary: Social Capital argues that aging is the single driver behind cancer and heart disease. Slowing aging by one year would add $400B to the economy, but research in this area gets almost no funding. The biggest hurdle is that regulators don't see aging as a disease yet. It’s a high-risk, high-reward area that could change the economy if the science catches up.
  • Read more

13. Fast Mode: Anthropic’s New Pricing Lever — Rob Litterst

  • Why read: How AI companies are starting to charge more for speed.
  • Summary: Anthropic’s "Fast Mode" shows that latency is the new way to make money. Since many models are starting to perform similarly, companies are charging a premium for faster answers. This moves pricing away from just counting tokens and toward the value of an operator's time. Users are proving they will pay for tools that help them work faster.
  • Read more

14. Where Are the American Open Source Models? — Jamin Ball

  • Why read: Why "source available" licenses are replacing traditional open source in AI.
  • Summary: New AI licenses are more restrictive to stop big tech companies from taking all the value for free. In the past, being the "default" was the goal, but now companies need to protect their commercial upside. The lack of a major American open-source champion is leaving a gap for international companies to fill.
  • Read more

15. Recursion Starts at Home (Liberty's Highlights) — Liberty’s Highlights

  • Why read: How Anthropic uses Claude to build the tools that make Claude better.
  • Summary: Anthropic follows a philosophy where they use their own models to automate their engineering. The labs that move the fastest are the ones that have automated their own internal work. For everyone else, the lesson is to use AI to build the specific tools your team needs, creating a loop of better productivity.
  • Read more

Themes from yesterday

  • Context over prompts: AI utility comes from connecting it to real-time work data like Slack and transcripts, not just better prompt engineering.
  • The token vs. human budget fight: Engineering leads are now choosing between hiring more people or increasing their spend on high-end AI models.
  • Local AI and the Minimill: Teams are moving toward local routers to handle simple tasks, saving the cloud for the most complex work.
  • Agency over Agents: Technology isn't the bottleneck for AI; it's a lack of organizational trust and decentralized decision-making.
  • The end of generic sales: AI filters are killing automated "personalization." To get through, messages must be high-signal and tied to the buyer's finances.