1. Salesforce is dying and there’s a way to save it — Animesh Koratana

  • Why read: Understand the "SaaSpocalypse" and why enterprise value is shifting from data storage to automated work.
  • Summary: The "System of Record" moat is collapsing as AI agents begin to handle the work that humans previously used software to perform. Salesforce and other giants have seen massive valuation drops because their role as "filing cabinets" is becoming obsolete in an era where the "System of Action" is the new priority. For operators, this means the focus must shift from data capture to delivering finished outcomes. Companies that fail to integrate agentic work directly into their platforms risk becoming commodity infrastructure. The practical implication is a migration of budgets from the $1 "software" bucket to the $6 "labor" bucket, now captured by AI.
  • Link: https://twitter.com/akoratana/status/2035050678687867079

2. Every Claude Code Hack I Know (March 2026) — Matt Van Horn

  • Why read: Master the specific technical workflows that define high-velocity "compound engineering" in 2026.
  • Summary: The 2026 development workflow flips the traditional ratio to 80% planning and 20% coding using the `/ce:plan` command. By treating every idea, bug, or screenshot as a prompt for a structured `plan.md` file, developers can delegate the mechanical execution to sub-agents. Van Horn advocates for "voice-pilling"—using LLM-aware transcription to dictate complex logic while driving or multitasking. The strategy involves running 4-6 parallel Claude Code sessions to handle research, building, and bug-fixing simultaneously. This approach transforms the developer into a conductor of an automated swarm rather than a manual coder.
  • Link: https://twitter.com/mvanhorn/status/2035857346602340637

3. 6 Levels of Making Claude Code Run Autonomous — Aakash Gupta

  • Why read: Learn how to scale AI agent tasks from 15-minute sessions to 24-hour autonomous cycles.
  • Summary: Most users are stuck at Level 1 (killing permission prompts), but the "pro" tiers involve persistent infrastructure and looping hooks. Level 4 introduces the "Ralph Wiggum loop," a stop-hook that prevents the agent from exiting until a complex task is verified, enabling runs that last for days. Level 6 moves the agent to a VPS via OpenClaw, allowing for 24/7 execution across email, git, and calendars. The core unlock is giving the model a way to verify its own work through structured eval loops. For product teams, this means moving away from "chatting" with AI toward deploying "always-on" digital employees.
  • Link: https://twitter.com/aakashgupta/status/2035805431516246363

4. New model for engineering team structure in 2026 — Dan Shipper

  • Why read: A blueprint for lean, AI-native organizations centered on "Pirates" and "Architects."
  • Summary: Engineering teams are evolving into two-person units: a Pirate who uses "vibe coding" to rapidly discover product-market fit, and an Architect who structures that surface into a reliable machine. The Pirate owns the product end-to-end and moves with extreme speed, while the Architect works across multiple codebases to solve deep technical scaling issues. This model suggests that most startups no longer need full-time traditional engineering teams until well after PMF. It forces a mindset shift where code is treated as disposable exploration until the Architect solidifies it. Practically, this allows teams to stay tiny while maintaining the output of a 50-person org.
  • Link: https://twitter.com/danshipper/status/2035842017553465814

5. Being a human in the singularity — kache

  • Why read: A visceral report on how Codex 5.4 and personal AI puppets are erasing the "cost of doing" for elite individuals.
  • Summary: The author details how a "Codex 5.4 extra high" model autonomously migrated an entire server infrastructure for $0 labor cost by figuring out SSH keys and hitting API endpoints. This illustrates the "singularity" for programmers: the ability to search the space of all possible computer programs instantly. Elite engineers are now "walking the graph" of distributed systems using custom LLM extensions to automate profiling and hotspot discovery. The practical takeaway is that personal information infrastructure is now a reality; individuals can turn any website into a private API using sandboxed browser puppets. We are entering an era where one person's "velocity" is limited only by their ability to direct intelligence.
  • Link: https://twitter.com/yacineMTB/status/2035779012908175740

6. Clawcard: Infrastructure for Agents — Christian Bryant

  • Why read: Critical infrastructure for closing the loop on agentic internet operations (payments/ID).
  • Summary: Clawcard provides agents with the missing primitives of the internet: email, phone numbers for SMS verification, and virtual credit cards. This allows agents to sign up for services, pay for APIs, and handle checkouts without human intervention. It integrates stablecoin wallets (USDC on Base) and secure credential storage into a single CLI. By removing the "human in the loop" for administrative hurdles, agents can move from "planning" to "fully operational." For builders, this is the foundational layer that enables truly autonomous e-commerce and SaaS-handling agents.
  • Link: https://twitter.com/cblovescode/status/2035804007151878383

7. 9 Lessons from GTM Agents at Scale — Brendan J Short

  • Why read: Evidence-based insights from companies like Vercel and Ramp on deploying AI in sales.
  • Summary: Growth-stage companies are replacing complex human GTM (Go-To-Market) chains with "system-thinking" agents. Vercel notably reduced its inbound SDR count from 10 to 1, while LangChain saw a 250% increase in lead quality. These agents aren't "exotic AI infrastructure" but are assembled using existing building blocks connected via MCP (Model Context Protocol). The key lesson is that agents thrive when they have access to clean data systems and specific "intent" triggers. For operators, the message is clear: stop hiring for high-volume, repetitive sales tasks and start building the agentic workflow.
  • Link: https://www.thesignal.club/p/9-lessons-from-11-growth-stage-companies

8. World Models vs. LLMs: Saining Xie Interview — Kyle Chan

  • Why read: A contrarian view on the path to AGI from a co-founder of AMI Labs (with Yann LeCun).
  • Summary: Saining Xie argues that Silicon Valley is currently "LLM-pilled" and that text-based models are insufficient for true AGI. He believes the true frontier is "World Models"—predictive brains that understand physical, high-dimensional, continuous signals rather than just digital text. Xie famously turned down Ilya Sutskever twice because of a disagreement over whether vision is a "solved" problem. AMI Labs is betting on JEPA (Joint Embedding Predictive Architecture) to create models that can reason about the physical consequences of actions. For strategists, this highlights a looming shift from generative AI to predictive physical intelligence.
  • Link: https://twitter.com/kyleichan/status/2035493986530718078

9. Short the tool, Long the work — Animesh Koratana

  • Why read: A framework for investing and building in an economy where software costs are approaching zero.
  • Summary: Applying the "Progression of Economic Value" framework, Koratana argues that AI is compressing commodities, goods, and services into a single infrastructure layer. Value is migrating to "Outcome Companies" that use tech to deliver finished work and "Experience Companies" that provide immersive human moments. Software as a standalone product is losing its premium because it is no longer scarce. The winning strategy is to move up the stack: don't sell a CRM (a tool), sell the "closed sale" (the outcome). Practical builders should focus on capturing the end-to-end labor value rather than selling another seat-based SaaS.
  • Link: https://twitter.com/akoratana/status/2035783146223096223

10. Revenue Durability is Failing in the Age of AI — OnlyCFO's Newsletter

  • Why read: Essential financial metrics for evaluating the survival of software companies in 2026.
  • Summary: The "SaaS glory days" are over, and the new critical metric is "Revenue Growth Endurance"—the ratio of this year's growth to last year's. AI is creating a durability crisis where high-growth companies are seeing massive deceleration because their moats are being bypassed by agents. Investors are now heavily discounting companies where 90% of the valuation relies on cash flows 5+ years out, as long-term predictability has vanished. For CFOs, the priority is automating high-volume, repetitive finance tasks (like AR collections) using agents to protect margins. If your revenue isn't tied to an "AI winner" narrative, your valuation is likely at risk of a major correction.
  • Link: mailto:reader-forwarded-email/a71b9b4634d7a6350ce7d13561006df4

Themes from yesterday

  • The SaaSpocalypse & Systems of Action: A major consensus that "Systems of Record" (Salesforce, etc.) are being replaced by agentic "Systems of Work" that prioritize outcomes over data storage.
  • Vibe Coding Org Design: The emergence of "Pirate/Architect" team structures, where AI allows tiny teams to manage massive product surfaces through high-velocity iteration.
  • Autonomous Agent Infrastructure: The shift from "copilots" to "autonomous agents" enabled by persistent VPS loops, OpenClaw, and financial primitives like Clawcard.
  • Physical World Models: A growing technical debate (led by Saining Xie and LeCun) that LLMs are a dead end for AGI, and the next leap requires modeling the physical, high-dimensional world.