1. There are only two paths left for software — David George

  • Why read: A brutal financial reality check for software CEOs on why the "comfortable middle" of moderate growth and low margins is dead.
  • Summary: Public markets have repriced software, signaling that terminal value is declining unless companies radically transform. CEOs must choose between accelerating revenue by 10+ points through AI-native products or rebuilding for 40-50% true operating margins (including SBC). The "weak form" of minor layoffs is insufficient; a "strong form" redesign of the entire machine is required within 12-18 months. Success requires identifying the top 5% of "100x" talent in the org to lead unglamorous, critical information-gathering and process-capture sprints.
  • Link: https://twitter.com/DavidGeorge83/status/2036091262080811265/?rw_tt_thread=True

2. W𝐡𝐲 𝐀𝐈 𝐦𝐚𝐧𝐝𝐚𝐭𝐞 𝐢𝐬 𝐧𝐨𝐭 𝐟𝐚st enough in your company? — Shekhar Kirani

  • Why read: Essential leadership tactics for driving AI adoption through conviction and "greenfield" projects rather than top-down mandates.
  • Summary: Many companies see stalled AI acceleration because they prioritize compliance and mandates over building genuine team conviction. Kirani advocates for a "greenfield first" strategy, building AI-native prototypes before touching the complex "brownfield" existing codebase. Critical to success is the CXO "practicing, not just preaching"—CEOs should personally build with tools like Claude Code to set the cultural pace. Companies must also revisit previously "impossible" or "too expensive" use-cases that AI now makes viable.
  • Link: https://twitter.com/skirani/status/2036279365957460045/?rw_tt_thread=True

3. How we made Ramp Sheets self-maintaining — Ramp Labs

  • Why read: A masterclass in "agentic observability" and how to offload the grunt work of maintenance to AI.
  • Summary: Ramp built a system where agentic "monitors" (one for every 75 lines of code) continuously watch production metrics and logs. When a monitor fires, it triggers a background coding agent that spins up a sandboxed environment, reproduces the bug, and pushes a fix via PR. This moves beyond scheduled nightly audits to a reactive system of action that caught 40 real bugs in its first week. The approach treats agents as cheap, infinitely patient workers that can synthesize telemetry data humans typically ignore.
  • Link: https://twitter.com/RampLabs/status/2036165188899012655/?rw_tt_thread=True

4. Fun convo with @bdeeter and the engineering leadership — Farhan Thawar

  • Why read: Practical lessons from Shopify’s long-term AI integration on managing costs, bottlenecks, and planning.
  • Summary: Shopify reports a conservative 20% productivity increase since adopting AI in 2021, with P90 engineers seeing 10x gains. A key tactical shift is encouraging engineers to use LLMs for planning/thinking before writing code (e.g., "Don't write code yet, help me make a detailed plan"). To manage "unlimited token" budgets, they implemented iteration depth limits and alerts to prevent runaway costs. They also highlight the "code review bottleneck," where AI generates code faster than humans can currently review it.
  • Link: https://twitter.com/fnthawar/status/2036033400012689631/?rw_tt_thread=True

5. Resilience — Aidan Gomez

  • Why read: A strategic argument for why AI sovereignty is now a prerequisite for national sovereignty and economic security.
  • Summary: The era of global multilateralism is fracturing, making technological and supply chain resilience a strategic necessity rather than an efficiency play. Gomez argues that dependence on a few massive technology conglomerates creates "single-point-of-failure" risks for democratic nations. Sovereign AI goes beyond data center locations; it requires technical guarantees for control, security, and operational autonomy. Leaders must pursue proactive, risk-tolerant investment in diversified supply chains to protect national self-determination.
  • Link: https://twitter.com/aidangomez/status/2036066718183809425/?rw_tt_thread=True

6. I turned my client into a millionaire using Claude Code — Mitchell

  • Why read: A highly tactical breakdown of using a "multitude of agents" to outperform human researchers and copywriters.
  • Summary: Instead of a single prompt, this system uses 20 specialized AI agents—each with its own context window and quality bar—to write production-ready scripts. The process starts with an exhaustive research sweep of YouTube, Reddit, and X to find "content with nerves" (high engagement/controversy) to use as ammunition. Specific agents handle hooks, body, and CTAs, while "manager" agents score everything across five dimensions, sending work back for iteration if it doesn't hit 10/10. This pipeline ensures novelty and copy intensity that single-agent outputs lack.
  • Link: https://twitter.com/MitcheIl/status/2036098438908293349/?rw_tt_thread=True

7. The Claude Dispatch Guide: 48 Hours Running AI From My Phone — Paweł Huryn

  • Why read: Insights into a new mobile-first workflow for PMs to direct parallel AI workstreams while on the move.
  • Summary: Claude Dispatch allows a mobile phone to act as a "command chair" to manage multiple independent AI sessions running on a desktop. This turns small gaps in the day (standing in line, commuting) into windows for directing "Cowork" tasks like summarizing emails or drafting specs. The key architectural insight is the shift from "chatting" to "orchestrating" multiple agents simultaneously across different files and tools. For PMs, this maps to their existing job of running parallel workstreams, but with AI analysts instead of humans.
  • Link: https://twitter.com/PawelHuryn/status/2036058594433519790/?rw_tt_thread=True

8. The Agentic Economy Will Be Massive. Agentic Commerce Won't — Robbie Petersen

  • Why read: A contrarian take on why the hype around "agents with wallets" may be misplaced compared to internal enterprise adoption.
  • Summary: Petersen argues that most AI agents will operate within the "org chart" of a business rather than as independent economic actors in a marketplace. Since 95%+ of software spend is top-down enterprise deployment, agents will likely automate closed-loop tasks (CRM research, accounting reconciliation) that don't require autonomous payments. Any per-use compute or data costs will likely be abstracted away by SaaS vendors rather than transacted via stablecoins or agentic wallets. The true "Agentic Economy" looks like the evolution of SaaS: replacing workflows within organizations.
  • Link: https://twitter.com/robbiepetersen_/status/2036118525451047170/?rw_tt_thread=True

9. AI Leverage Is Much More Than You Think It Is — Obie Fernandez

  • Why read: A methodology for shifting from "fast-worker" AI use to "manager-orchestrator" AI leverage.
  • Summary: Fernandez advocates for the "Multitude of Workers" pattern: treating AI as a team of specialized, virtual workers rather than a single chat window. The bottleneck in AI adoption shouldn't be the model's speed, but the user's synthesis and judgment. He suggests a "lightweight orchestration" prompt for Claude Code: tell it to spin up three sub-agents (e.g., researcher, drafter, cross-referencer) with focused briefs. The model should then review and integrate their outputs itself, rather than asking the human to do the coordination.
  • Link: https://twitter.com/obie/status/2035783258449822101/?rw_tt_thread=True

10. The four stages of AI development — Ankur Goyal

  • Why read: A framework for benchmarking where an organization sits on the maturity curve of AI adoption.
  • Summary: Goyal identifies a common four-stage trajectory for AI development: 1) Skepticism/Fear (seeing AI as a "toy"), 2) Hasty Shipping (bolting on features that inevitably break), 3) Infrastructure Building (focusing on evals and observability), and 4) Full Maturity. Many teams are currently stuck in Stage 2, facing unexpected errors and latency because they skipped the foundational "evals" work. Moving to higher stages requires treating AI performance as a core engineering discipline, specifically managing cost vs. performance tradeoffs.
  • Link: https://twitter.com/ankrgyl/status/2036123140137386379/?rw_tt_thread=True

Themes from yesterday

  • From Chat to Orchestration: A dominant shift toward using multiple specialized agents (the "multitude of workers") rather than single-prompt chat windows.
  • Enterprise Self-Maintenance: The rise of "agentic observability" where AI monitors, reproduces, and fixes its own production bugs (Ramp, Braintrust).
  • The Financial Mandate: A clear warning that software companies must either find 10% growth via AI-native products or slash costs to achieve 40%+ GAAP margins.
  • Leadership via Building: The recurring advice that CXOs and leaders must personally use tools (Claude Code/Dispatch) to overcome the "middle management" adoption stall.