#1 [AINews] Anthropic @ $30B ARR, Project GlassWing and Claude Mythos Preview — AINews

  • Why read: Unpacks the massive revenue jump and leaked details of Anthropic's most powerful (and restricted) model yet.
  • Summary: Anthropic has reportedly jumped to $30B ARR in April, strategically positioning itself against OpenAI's stalled growth. They have internally confirmed "Claude Mythos", a model trained on their largest run ever, which demonstrated dangerous capabilities like zero-day vulnerability discovery across major OSs. It's currently restricted to 40 partners under "Project Glasswing" instead of general release. This level of growth and capability suggests a rapid shift in the frontier AI balance of power.
  • Link: mailto:reader-forwarded-email/eea4eebca6ba26551cf4b1d46133c711

#2 Systems Engineering: Building Agentic Software That Works — Ashpreet Bedi

  • Why read: A crucial reminder that building agentic applications requires rigorous traditional systems engineering, not just prompt hacking.
  • Summary: The agentic software ecosystem is repeating past mistakes by ignoring systems engineering principles and optimizing components in isolation. Building reliable agentic software requires addressing five layers: Agent Engineering, Data Engineering, Security Engineering, Interface Engineering, and Infrastructure Engineering. Developers shouldn't rely on file-system-backed memory in shared sandboxes or bypass proper RBAC just because they are using LLMs. Treating software engineering as systems engineering ensures that your agents run deterministically, securely, and scalably.
  • Link: https://twitter.com/ashpreetbedi/status/2041568919085854847/?rw_tt_thread=True

#3 Can new companies be built as MCPs? — nihar bobba

  • Why read: Introduces a novel framework for building businesses entirely at the Model Context Protocol (MCP) layer.
  • Summary: As workflows centralize into AI harnesses like Claude Cowork, users increasingly depend on vendors for their ability to package enriched, workflow-ready context rather than their UI. This shift opens the door for purely "MCP-layer" companies that forgo traditional web dashboards. "Read-MCP businesses" can build moats through proprietary data acquisition and enrichment, feeding context to agents. "Write-MCP businesses" act as systems of record or executors of action, competing on trust, compliance, and authorization. It challenges founders to consider if their application UI is actually necessary.
  • Link: https://twitter.com/nbobba/status/2041557222698713402/?rw_tt_thread=True

#4 The Building Block Economy — Mitchell Hashimoto

  • Why read: Explains why creating high-quality, composable software primitives is the fastest path to massive adoption in the AI era.
  • Summary: The dominant way to build software is shifting from high-quality mainline apps to gluing together proven "building blocks." AI excels at assembling well-documented components, drastically lowering the barrier to entry and flooding the market with customized, niche software. By building primitives instead of monolithic apps, creators can outsource R&D to the community and capture a much wider array of use cases. This implies that modern product strategy should prioritize developer and AI ergonomics over broad, end-user feature parity.
  • Link: https://twitter.com/mitchellh/status/2041566958681014418/?rw_tt_thread=True

#5 Extreme Harness Engineering: 1M LOC, 1B toks/day, 0% human code, 0% human review — Latent.Space

  • Why read: A glimpse into the absolute cutting edge of AI-driven software development at OpenAI.
  • Summary: Ryan Lopopolo from OpenAI's Frontier team details how they operate a 1 million lines-of-code codebase with zero human-written code and zero human review before merging. They rely heavily on "Harness engineering," spending up to 1 billion tokens a day per developer to automate code generation and validation. Operating at this extreme scale pushes the boundaries of what is possible with models like Codex. It suggests that any team not aggressively maximizing their token spend on agentic workflows is falling behind.
  • Link: mailto:reader-forwarded-email/abda9f131be16bd35d23c441f37cf202

#6 How to Replace Your Entire GTM Engineering Layer With Claude Code — Michel Lieben

  • Why read: A highly practical breakdown of orchestrating Go-To-Market workflows via Claude Code and MCPs.
  • Summary: Managing a complex tech stack for lead generation can be vastly simplified by using Claude Code as a central hub. By feeding platform API docs to Claude, you can orchestrate workflows across tools like Clay, Wiza, and Google Analytics without deep technical knowledge. Providing Claude with specific "skills" gives it the context needed to build entire outbound campaigns and analyze ICPs directly from the terminal. This approach has drastically increased GTM engineer capacity, allowing one person to manage significantly more clients.
  • Link: https://twitter.com/MichLieben/status/2041172240956600636/?rw_tt_thread=True

#7 Finding the Right Altitude — Alfred Lin

  • Why read: A powerful mental model for solving complex strategic problems by first identifying the correct "toy problem."
  • Summary: Most strategic errors are failures of altitude—looking at a problem from the wrong distance. Iconic successes often began by focusing on a "toy problem" that contained the essential structure of the ultimate ambition, like Amazon starting with books or Apple with the iPod. Solving the toy problem builds the intuition, infrastructure, and muscle memory needed to tackle full complexity later. Before scaling a solution, founders should ask if they have found the right wedge that makes the broader architecture possible.
  • Link: https://twitter.com/Alfred_Lin/status/2041513637685407829/?rw_tt_thread=True

#8 When Will Anthropic Surpass NVIDIA? — Tomasz Tunguz

  • Why read: Provides perspective on the unprecedented speed at which foundation model companies are scaling revenue.
  • Summary: Anthropic reportedly added $10B in revenue in a single month, crossing a threshold in four years that took traditional SaaS giants nearly two decades. If growth continues at this breakneck pace, Anthropic could theoretically surpass NVIDIA's market cap in three to seven years, depending on deceleration rates. While customer concentration remains a risk, the sheer velocity of this scale-up is historic. It highlights the unique economic dynamics and massive addressable markets captured by leading AI labs.
  • Link: mailto:reader-forwarded-email/d4db3269d65224424e7b8303716865e0

#9 Cycles of disruption in the tech industry — The Pragmatic Engineer

  • Why read: Contextualizes the AI revolution by comparing it to historical tech shifts, featuring insights from Kent Beck and Martin Fowler.
  • Summary: The current AI shift mirrors past disruptions like the advent of OOP and Agile, but at a much faster pace. Just as with Agile, the AI ecosystem is seeing misaligned incentives, "snake oil" vendors, and resistance from mid-pack developers whose careers may suffer. Companies are struggling to adapt AI to complex legacy codebases and are wrongly resorting to measuring input metrics like PR frequency instead of outcomes. To avoid burnout, engineers must set boundaries and recognize when they are producing "negative value."
  • Link: mailto:reader-forwarded-email/2b5113d773233b7e718050fc982f88f6

#10 A few not so obvious points from implementing AI on enterprise — giyu_codes

  • Why read: Grounded, hard-learned lessons on the realities of deploying AI within large corporations.
  • Summary: Rushing into "low-hanging fruit" AI use cases often backfires as teams get bogged down in corporate systems and compliance. Proper IT infrastructure, accountability, and governance frameworks are vastly more important than basic context management. Instead of trying to automate and replace workers, enterprise AI initiatives should focus on empowering them to avoid massive compliance failures like expensive TCPA violations. Hiring long-term local talent pays off more than relying on agencies promising quick automation.
  • Link: https://twitter.com/giyu_codes/status/2041368678307070378/?rw_tt_thread=True

#11 wisdom is the new intelligence — Ole Lehmann

  • Why read: A profound take on how human value shifts when AI commoditizes raw intelligence and knowledge work.
  • Summary: Throughout history, the most valuable human edge has shifted from physical strength, to learned skills, and then to raw intelligence. As AI outsources cognitive tasks like coding, writing, and analyzing, the new human edge is wisdom. Wisdom in this context means having the taste and emotional intelligence to make high-quality decisions, see hidden patterns, and direct AI effectively. We are transitioning from knowledge workers to "wisdom workers" whose primary value is judgement rather than execution.
  • Link: https://twitter.com/itsolelehmann/status/2041180695268323331/?rw_tt_thread=True

#12 Patrick Collison on what he wishes he did differently when scaling Stripe — Startup Archive

  • Why read: A contrarian view on growth models that emphasizes go-to-market structure over passive organic curves.
  • Summary: Stripe's founder warns against believing your company is on a natural, passive "growth curve." Instead, growth is a function of the go-to-market apparatus you build to capture specific segments. Collison wishes Stripe had explicitly mapped out concentric circles of their total addressable market early on, working backward to build the organizational structure needed for each segment. This mental model puts growth firmly within a company's control rather than leaving it to market momentum.
  • Link: https://twitter.com/StartupArchive_/status/2035022174030073963/?rw_tt_thread=True

#13 Harvey: ~$1B raised across 4 rounds in 14 months — Art Levy

  • Why read: Explains the "Capital Wars playbook" being run by enterprise AI startups to lock out competition.
  • Summary: AI legal companies like Harvey and Legora have raised nearly $1.7B combined in a short window, utilizing capital as a deep defensive moat. This aggressive fundraising mirrors Uber's early days, subsidizing growth to capture market share, lock up enterprise customers, and acquire top-tier talent. Over time, these highly capitalized businesses intend to transition into massive free cash flow generators once distribution is secured. The playbook indicates that in the AI app-layer, financial firepower is a core strategic advantage.
  • Link: https://twitter.com/artlevy/status/2041584508714262852/?rw_tt_thread=True

#14 The Founders Who Got Storytelling Right — Hiten Shah

  • Why read: Demonstrates how a simple, consistent narrative acts as an operational alignment tool, not just marketing copy.
  • Summary: A founder’s ability to tell a clear, repetitive story directly impacts how fast a company aligns, hires, and scales. Great storytellers, like Marc Benioff with Salesforce's "No Software," ensure every department works off the same mental framework. Without a core narrative, product teams build disconnected features, and sales teams struggle to pitch effectively. Founders must treat storytelling as a strategic priority, answering exactly what job they do, why they win, and where they are going.
  • Link: https://twitter.com/hnshah/status/2041586743326208293/?rw_tt_thread=True

#15 AI is NOT killing the software developer — James Bessen

  • Why read: Provides a data-driven argument for why AI-driven productivity gains will increase, not decrease, software jobs.
  • Summary: Despite fears of a "job-pocalypse," historical data and current trends indicate AI is actually boosting developer demand. As AI takes over mundane tasks, developer productivity increases, which lowers the cost of software and spurs the creation of new and better products. Similar to past tech leaps like cloud computing or open-source, this dynamic creates a Jevons paradox where efficiency drives up overall consumption. Unless AI can instantly perform every single task a developer does, the industry will continue to add jobs.
  • Link: https://twitter.com/JamesBessen/status/2041534963057655917/?rw_tt_thread=True

Themes from yesterday

  • The Shift from Coding to Orchestrating: As seen in "Extreme Harness Engineering" and "Replacing GTM with Claude Code," the frontier of software is rapidly shifting toward building massive systems via AI agents without human-written code.
  • Re-evaluating the Application Layer: Businesses are rethinking traditional SaaS models, either utilizing "Capital Wars" to build insurmountable enterprise moats or creating invisible "MCP-layer" companies that serve enriched context directly into AI harnesses.
  • The Premium on Judgment: With raw intelligence and execution becoming commoditized, strategic clarity—such as finding the right "toy problem," embedding wisdom, and cohesive storytelling—is becoming the ultimate human edge.
  • Systems Thinking for AI: Building with AI requires returning to fundamental engineering principles—like robust IT governance, proper data layers, and security engineering—rather than relying on fragile prompt hacks.