1. Agents Will Become Companies — Sreeram Kannan

  • Why read: A strategic thesis on how the intersection of AI (intelligence) and crypto (capital formation) will create a new class of purely software-native, agentic firms.
  • Summary: Kannan argues that while early AI x crypto projects focused on payments or inference markets, they missed the bigger picture: democratizing the creation of software companies. AI provides the intelligence, and crypto provides the digitally native ownership structure, enabling companies with agents coordinated by tokens instead of human founders. These "agentic companies" will have structurally lower operating costs and different capital formation paths than traditional venture-backed startups. The thesis suggests we are entering an era where economic organization itself is fundamentally altered by autonomous, investable software entities.
  • Link: https://twitter.com/sreeramkannan/status/2041228233426018394/?rw_tt_thread=True

2. Memory Bandwidth for Local AI Hardware (2026 Edition) — Ahmad

  • Why read: A practical hardware guide clarifying why memory bandwidth, not capacity or TOPS, is the true bottleneck for running local AI models effectively.
  • Summary: Ahmad explains that while memory capacity dictates whether a model fits, bandwidth determines the generation speed (tokens per second). He breaks down the 2026 hardware landscape into tiers, noting that a 32GB RTX 5090 outpaces larger unified-memory machines for speed, though systems like Mac Studio remain necessary for models that simply won't fit on normal GPUs. The core mental model is: capacity determines fit, bandwidth determines speed, and software dictates utilization. For operators building local AI workflows, understanding this distinction is critical to avoid misallocating budget on high-capacity but slow-bandwidth hardware.
  • Link: https://twitter.com/TheAhmadOsman/status/2041331757329285589/?rw_tt_thread=True

3. some thoughts on anthropic at $30bn run rate — FleetingBits

  • Why read: An analysis of the staggering implications of Anthropic's reported $30B revenue run rate and how AI capabilities are breaking traditional adoption curves.
  • Summary: Anthropic is reportedly growing at an annualized 9,700%, representing the fastest revenue growth at this scale in history. The author notes that AI bypasses traditional adoption hurdles because it provides a general, human-shaped interface with no complex UX to learn. Furthermore, step-changes in capabilities (like Claude Code) create entirely new adoption curves and demand profiles, rather than just cannibalizing existing software spend. This rapid diffusion across private enterprise suggests that as long as Anthropic maintains access to compute, the top-tier AI market will experience unprecedented economic expansion.
  • Link: https://twitter.com/fleetingbits/status/2041289841225474244/?rw_tt_thread=True

4. Why SaaS got priced out — ProducTea with Leah

  • Why read: A sharp critique explaining why classical B2B SaaS valuations have collapsed by 73% despite continued revenue growth.
  • Summary: Leah argues that the market has repriced traditional SaaS (like HubSpot) because the fundamental input model—humans manually entering structured data—is becoming obsolete. The future paradigm involves unstructured input, where AI seamlessly pulls context from voice, calendars, and conversation history without requiring manual field updates. Because legacy SaaS architectures are constrained by their existing customer bases, they face an innovator's dilemma in adopting this AI-first approach. For operators, the takeaway is that building features around manual data entry is a dead end; the focus must shift to high-quality processing of fuzzy, unstructured inputs.
  • Link: mailto:reader-forwarded-email/01d70b59c99e35674340ea41ef98785c

5. My biggest takeaways from @AnthropicAI's Head of Growth Amol Avasare: — Lenny Rachitsky

  • Why read: Insider insights into how the world's fastest-growing AI company manages product, engineering, and its own growth leveraging AI.
  • Summary: Rachitsky highlights that AI leverage is currently disproportionately benefiting engineers, forcing PMs and designers to manage the output of effectively much larger engineering teams. Anthropic automates its own growth through "CASH", an internal system that handles copy and UI tweaks at the level of a junior PM. Avasare argues that AI companies should focus on massive swings rather than micro-optimizations, given the exponential growth of underlying model capabilities. Crucially, the one bottleneck AI hasn't solved is cross-functional human alignment, cementing the enduring need for product management in complex organizations.
  • Link: https://twitter.com/lennysan/status/2041166073794592926/?rw_tt_thread=True

6. The founder of Postman says you have to kill your... — Ivan Burazin

  • Why read: A blueprint for the modern, AI-era organizational chart that eliminates middle management layers and requires everyone to build.
  • Summary: According to Postman's founder, companies must abandon pre-AI hierarchies in favor of wide spans of control where leaders work directly with individual contributors. In this new model, PMs prototype directly in Claude rather than writing PRDs, designers ship PRs via Cursor, and staff engineers lead projects with high agency across the full stack. The mandate is binary: you are either building or you are selling. For management, the role shifts away from task coordination toward developing and exercising better strategic judgment.
  • Link: https://twitter.com/ivanburazin/status/2041199368296931595/?rw_tt_thread=True

7. some more ramblings from working at @AnthropicAI — austin

  • Why read: A practical framework for how growth marketers and operators should actually be using AI, moving beyond basic automation.
  • Summary: Austin breaks AI usage into four dimensions, arguing that simply automating manual tasks is the least interesting application. Instead, operators should use AI for breadth (e.g., generating 20 keyword angles to overcome individual bias) and to execute strategies that were previously below the ROI threshold due to manual costs. Most importantly, he advocates for building hyper-personalized tools and agents tailored to your exact stack and edge cases, rather than relying on generic plugins. The true leverage of AI lies in expanding the frontier of what is economically feasible to attempt for a solo operator.
  • Link: https://twitter.com/helloitsaustin/status/2041218663827812815/?rw_tt_thread=True

8. Pocket Power : From State of the Art to Your Phone in 23 Months — Tomasz Tunguz

  • Why read: A compelling visualization of how rapidly AI models are compressing, bringing frontier intelligence directly to consumer devices.
  • Summary: Tunguz highlights that Google’s new Gemma 4 E4B model matches GPT-4o performance but runs entirely locally on a phone, representing a 450x compression in just 23 months. This acceleration is driven by better distillation algorithms, extreme talent density, and massive capital investments in training. The implication is that any current state-of-the-art capability will be available on a smartphone within two years. For builders, this timeline guarantees that highly capable, zero-latency, offline AI applications are on the immediate horizon.
  • Link: mailto:reader-forwarded-email/46cca43e732d1c3f08eddd9217de82e5

9. No "New Deal" for OpenAI — Will Manidis

  • Why read: A stark warning about the rising local and political opposition to the massive infrastructure requirements of the AI industry.
  • Summary: Manidis contrasts OpenAI's polished policy proposals in Washington with the ground reality of communities actively blocking AI data centers due to power consumption concerns. With $162 billion in AI infrastructure projects delayed or blocked over the past year, public backlash is becoming a tangible bottleneck. The author highlights the disconnect between AI executives dismissing energy concerns and the growing grassroots consensus against resource extraction by tech giants. This suggests that the ultimate constraint on AGI may not be algorithmic, but physical and political resistance to infrastructure expansion.
  • Link: https://twitter.com/WillManidis/status/2041262402889457721/?rw_tt_thread=True

10. Confessions of a Millennial in Tech — Elena Verna

  • Why read: A relatable, introspective look at the psychological toll and shifting leverage of knowledge workers in the AI transition.
  • Summary: Verna articulates the disorientation experienced by seasoned tech workers as AI flattens the value of hard-earned, craft-based intuition in fields like growth, marketing, and product management. As junior employees using AI produce surprisingly solid work in minutes, the traditional signals of seniority and linear career ladders are rapidly weakening. Furthermore, the economic gains of AI efficiency are being absorbed by the system as demands for greater output, rather than resulting in more leisure time. This forces operators to confront how their professional identities and compensation models must evolve when execution becomes abundant and cheap.
  • Link: https://www.elenaverna.com/p/confessions-of-a-millennial-in-tech

Themes from yesterday

  • The Collapse of the Traditional Org Chart: AI is flattening hierarchies, forcing a shift where PMs code, designers ship PRs, and middle management is replaced by high-agency ICs.
  • Physical & Political Bottlenecks: While AI software scales exponentially, physical infrastructure is hitting real-world constraints as communities block data centers and energy demands face intense scrutiny.
  • Execution as a Commodity: The premium on manual execution and structured data entry is evaporating, forcing legacy SaaS to reprice and knowledge workers to tie their value to strategic judgment rather than output speed.
  • Pocket-Sized Frontier Intelligence: The rapid compression of models (450x in 23 months) is shifting the center of gravity toward powerful, zero-latency local models running on consumer edge devices.
  • Agentic Capital Formation: The intersection of AI and crypto is moving beyond payments to create entirely new corporate structures—software-native firms managed by tokens rather than human founders.