You can't prohibit agile software development, but you kanban it.

The Senior Engineer Crisis Isn't Coming. It's Already Here.

• 📉 Shrinking pipeline: The industry is facing a structural talent shortage caused by a 60% drop in entry-level hiring since 2022 and the loss of "natural" on-the-job training.
• 🤖 AI-driven atrophy: By removing the "friction" of manual coding and debugging, AI tools allow junior engineers to ship output without building the deep mental models and system intuition necessary for long-term seniority.
• ⚖️ Emerging bifurcation: The labor market is splitting into two tiers: a small, highly paid group of deep systems thinkers and a larger group of AI-assisted developers who struggle with complex, non-routine problems.
• ⏳ Inevitable talent war: A massive, competition-heavy talent war for senior engineers is likely between 2029 and 2031, as companies realize too late that they stopped developing their own internal pipelines.
• 🏗️ Intentional development: To survive, leaders must move beyond measuring sprint velocity and instead "reconstruct friction" by assigning juniors to high-ambiguity problems that require deep reasoning rather than just AI-assisted coding.
• 🧠 Knowledge preservation: Organizations must actively capture and distribute institutional knowledge—which is currently trapped in the heads of senior staff—before it leaves the company or degrades through attrition.

Tech CEOs are apparently suffering from AI psychosis

• 🤖 Tech executives are suffering from "AI psychosis" — they see AI's potential in controlled demos but don't understand the messy reality of implementation that happens "at the last mile" of actual work.
• 👨‍💼 CEOs like Aaron Levie argue that top leaders are "sufficiently distant" from day-to-day operations to know what can actually be automated — they don't review buggy code, catch hallucinated library calls, or spend days combing through contracts.
• 📉 The tech industry is experiencing record layoffs in 2026 (115,430 people in just five months) with companies crediting AI productivity gains, though research suggests these claims are overstated or unproven.
• 📊 Research contradicts CEO assumptions: A UC Berkeley meta-analysis found "no robust relationship between AI adoption and aggregate productivity gain," and MIT researchers predict AI agents won't achieve human-quality work until around 2029.
• ⚠️ The real risk isn't that AI can't work — it's organizational chaos. When everyone uses AI to produce more, the bottleneck shifts to executives who must approve everything, potentially creating an unmanageable workload.
• 💡 Levie's advice: CEOs should experiment with AI extensively to understand "both the upside and the real work" before making major decisions — but most seem to be skipping this step.
• 🚨 Some CEOs like ClickUp's Zeb Evans are already betting big, laying off 22% of staff after deploying 3,000 internal AI agents, believing this creates a "100x org" — a claim unsupported by current data.

How to share your point of view (even if you’re afraid of being wrong)

• 🤐 Silence costs you — Not sharing your perspective robs your organization of valuable insights and prevents you from being noticed as a high performer.
• 🧠 Unlearn the "stay safe" mentality — Speaking up is a learnable skill that can be built with practice through intentional principles and frameworks.
• 📊 You have unique proximity to problems — Junior employees often have the only close-up view of certain issues and context that others lack, making their insights invaluable.
• 🎯 Frame your contribution beyond just facts — Don't just document what happened or summarize data — share what it means for the business and why the problem matters.
• ⚖️ Match your language to certainty — Use accurate language reflecting your confidence level, and remember that controversial ideas require higher burden of proof.
• 💡 Ground hunches in evidence — Always articulate where your instinct is coming from (data points, patterns, experiences) to give your perspective credibility.
• 🏃 Your perspective IS part of your role — In the relay race of work, sharing your "so what" and interpretation isn't passing the baton — it's completing your leg of the race.

Team cohesion: what anthropology teaches us about holding organizations together

• 🔗 Team cohesion is built through daily "conformations" — small concrete acts like votes, shared meals, pricing agreements, and turn-taking that make relationships visible and real, not through abstract trust or communication workshops.
• 🏢 Two fundamental organizational challenges require different solutions — division of labor (solved through org structure) and integration of effort (solved through relational design), yet most organizations focus heavily on the first and neglect the second.
• 🤝 Three distinct types of cohesion work through different conformations: equality-driven organizations use matching acts like rotating votes and shared profits; market-driven organizations use pricing acts and internal currencies; community-driven organizations use sharing acts and collective responsibility.
• 🌍 Anthropology, not management theory, best explains how decentralized organizations actually hold together — Alan P. Fiske's concept of "conformations" reveals the invisible mechanism that synchronizes mental models across autonomous teams without middle managers.
• 📊 Real-world proof exists at scale — organizations like Buurtzorg (10,000+ nurses), NER Group companies (Basque Country), and Disco Corp (Japan) demonstrate that radical decentralization works when the right conformations are consistently practiced.
• ⚠️ Forcing one cohesion type onto an organization designed for another creates "corporate theater" — the key is identifying which relational logic fits your organization and designing conformations accordingly.
• 💡 Remove middle managers and conformations become the organization itself — these daily acts aren't nice-to-haves; they're the actual glue holding decentralized systems together.

AI demands more engineering discipline. Not less

• 🤖 AI code generation reached viability in late 2025 — it now produces code "approximately as good as that of the median software engineer" for common patterns, much faster and cheaper than before
• 📚 The shift mirrors previous infrastructure changes — just as immutable infrastructure replaced handcrafted servers, AI-generated code should be treated as "disposable and regenerable" rather than precious artifacts to be preserved
• 🧠 Software's true product is "shared understanding" — but code has been a poor container for this knowledge; we need better evaluation methods and observability instead of relying on human code review as the primary quality gate
• 🔄 Mutation accumulates entropy — when code regeneration is cheap, "editing in place becomes risky"; code should be treated as "a materialized view of understanding that is useful while current, disposable when stale"
• ⚙️ AI demands MORE engineering discipline, not less — we need stronger specs, better architecture artifacts, behavioral testing, characterization tests, and production observability to validate nondeterministic systems
• 🚀 The real bottleneck shifted — production (not code production) is now the constraint; engineers should focus on encoding knowledge into systems, specifications, and tests rather than manually reviewing generated code
• 🎯 Validation is not where humans should focus — "human brains are not good at validation"; invest human creativity and reasoning into architecture, specifications, and understanding systems instead

That’s all for this week’s edition

I hope you liked it, and you’ve learned something — if you did, don’t forget to give a thumbs-up, add your thoughts as comments, and share this issue with your friends and network.

See you all next week 👋

Oh, and if someone forwarded this email to you, sign up if you found it useful 👇