The transition from speculative AI to hard-boiled implementation is bringing the “degree of difficulty” into sharp focus this week. Whether it’s Starbucks pulling back on computer vision in retail or enterprise CEOs admitting that AI is changing the moral fiction of management, we are moving past the “vibe-coding” era into one defined by structural moats, regulatory clarity, and capital efficiency. The signal this week: Execution is the only moat that lasts.
Signals from the Stream#
Starbucks kills AI system#
🦔This one happened yesterday but is still worth flagging. Starbucks killed its AI-powered inventory counting tool after nine months in North American stores. The system used LiDAR sensors and cameras to count syrups and milks, routinely confused similar products, and missed… pic.twitter.com/beOSIQArgt
— Hedgie (@HedgieMarkets) May 23, 2026
Summary: Starbucks has reportedly discontinued its AI-powered inventory counting tool after a nine-month pilot in North America, citing inaccuracies with LiDAR and camera systems.
Why it Matters: Real-world computer vision in “dirty” environments (retail, warehouses) remains significantly harder than digital-native AI tasks. Edge-case density is the silent killer of retail automation.
My Take: Hardware is the bottleneck for the next phase of AI. Until sensors can distinguish between different syrup flavors in low light as reliably as a tired barista, humans remain the most efficient “computer” for the last mile of physical operations.
Enterprise software business model upended#
love that AI is revealing managerial effectiveness to be a moral fiction https://t.co/qovn8VBmYe
— Danielle Morrill (@DanielleMorrill) May 23, 2026
Summary: A discussion on how AI is disrupting the traditional “Roadmap Defense” in enterprise software, forcing a shift from long-term promises to immediate, generative value.
Why it Matters: For decades, SaaS vendors used roadmaps to freeze customer budgets. AI has collapsed the time-to-value loop, making “we’ll build that next year” an unacceptable answer.
My Take: The Roadmap is a moral fiction. In a world where agents can build features in weeks, the three-year vision is a liability. Speed of shipping is now a defensive requirement, not a competitive advantage.
Headcount cuts while business is booming#
Today we reduced headcount by 22%. The business is the strongest it's ever been. So I think it's important to be direct about what I'm seeing and why.
— Zeb Evans (@DJ_CURFEW) May 21, 2026
First, I made this decision and I own it. I did it because the way to operate at the highest level of productivity is changing,…
Summary: ClickUp CEO Zeb Evans announced a 22% headcount reduction despite the business being at its “strongest ever,” citing a fundamental change in how to operate at peak productivity.
Why it Matters: We are seeing the first wave of “AI-native rightsizing.” High-growth companies are realizing that the headcount-to-revenue ratio of the ZIRP era is obsolete.
My Take: Growth without efficiency is technical debt. Expect more “healthy” companies to cut deep as they re-architect their operations around agentic workflows rather than middle-management coordination.
Headcount cuts (again!) while business is booming#
Cloudflare CEO Prince on how AI changes who gets laid off first:
— Wall St Engine (@wallstengine) May 21, 2026
Two weeks ago I laid off more than 20% of my workforce. I didn’t do it because Cloudflare is struggling. We posted record revenue growth, have strong free cash flow and are adding an unprecedented number of… pic.twitter.com/PH7Qa0m7Cu
Summary: Cloudflare CEO Matthew Prince discusses how AI is changing layoff criteria, shifting the focus toward roles that are easily replaced by automated efficiency tools.
Why it Matters: Layoffs are no longer just about cost-cutting; they are about re-skilling the organization for an AI-first architecture.
My Take: Infrastructure is Strategy. When your infrastructure handles the complexity, your “talent” needs to shift from specialists to system orchestrators.
The fallacy of product-market fit?#
Why founder conviction matters more than ever
Summary: An argument for why founder conviction matters more than market validation in the current landscape, where data is often a lagging indicator.
Why it Matters: In rapidly shifting markets (like AI or Climate), “traction” is often a mirage. Deep conviction allows a founder to stay the course through the “trough of disillusionment.”
My Take: Conviction is the ultimate hedge. Market data tells you where the ball was; conviction tells you where the field is being rebuilt.
Let’s build a compiler!#
in 1988 a physicist named Jack Crenshaw got tired of compiler textbooks being unreadable
— trish (@TrisH0x2A) May 19, 2026
so he wrote his own tutorial on a BBS called "Let's Build a Compiler"
he starts from a parser that handles exactly one digit and adds one feature per installment until you have a real… pic.twitter.com/uIL6fNiH9s
Summary: A reminder of Jack Crenshaw’s 1988 tutorial “Let’s Build a Compiler,” which taught complex systems through radical simplification.
Why it Matters: As we build increasingly abstract AI systems, the foundational understanding of how code is parsed and executed is becoming a lost art.
My Take: Complexity is a tax on understanding. Returning to first principles—like building a compiler from scratch—is the only way to ensure we aren’t just building “black boxes on top of black boxes.”
Deeper Dives from the Library#
Tagging my blog posts with BERTopic and LLMs by Vicki Boykis#
We kind of forget how hard something as seemingly pedestrian as “tagging” can be!
Summary: An exploration of using topic modeling and LLMs to automate content taxonomy, arguing that manual tagging is becoming less relevant for readers but more vital for internal knowledge management.
Why it Matters: As search moves from keyword matching to semantic retrieval, the structure of our data matters more than the specific labels.
My Take: Taxonomy is the new search. If you don’t structure your internal knowledge now, no agent will be able to find it later. Semantic density is the goal.
Filipinos LOCKED OUT of US Markets: Why the SEC Banned IBKR (& What’s Next) by John Dang#
I did an article on this, but really worth highlighting again!
Summary: An analysis of the regulatory hurdles preventing emerging market investors from accessing global capital markets directly.
Why it Matters: Financial inclusion is often throttled by national regulators trying to prevent capital flight, which perversely traps local capital in low-growth environments.
My Take: Capital wants to flow to the highest utility. Regulatory moats only work in the short term; eventually, DeFi or alternative rails will bridge the gap.
The Apollo Guidance Computer#
Summary: A technical review of the hardware and software constraints of the AGC, focusing on real-time task management and error recovery.
Why it Matters: We often forget that we landed on the moon with a fraction of the compute power in a modern toaster. It was a triumph of software optimization over hardware abundance.
My Take: Constraint breeds excellence. Our current “compute abundance” has made us lazy. We need more “Apollo-era” thinking in our software architectures.
PS: you can review the original Apollo 11 guidance computer course code!
Deep Moats and Platform Shifts in Computing - Part 2 by Pushkar Ranade#
Summary: A breakdown of NVIDIA’s CUDA moat and why vertical integration in the AI stack is so difficult to disrupt.
Why it Matters: Hardware is only half the battle; the software ecosystem (CUDA) is the true barrier to entry for any NVIDIA challenger.
My Take: Software is the gravity of hardware. You can build a faster chip, but you can’t easily build a better developer ecosystem. Ecosystem is the ultimate moat.
What builders need to know about the CLARITY Act by a16z crypto#
Summary: A guide to the proposed CLARITY Act and its potential impact on the U.S. crypto regulatory environment.
Why it Matters: Regulatory uncertainty has been the primary brake on institutional blockchain adoption. Clarity (even if strict) allows for long-term capital planning.
My Take: Rules are better than silence. Builders can navigate a maze; they can’t navigate a void. Governance is a feature.
