Gianfranco's Best of March 2025 Reading List

Welcome to the March 2025 edition of my monthly reading list.

This month, I've curated my favorite essays across categories like demand discovery, industrial automation, and the mental models that help us better navigate complex systems.

What if complexity isn’t the problem—it’s the point?

3 Essays from my March reading list exploring why real insights come from embracing friction, misunderstanding customers, and paying attention to inconvenient details.

• Speedrunning the Skill of Demand - Cedric Chin
  • Chin argues against talking directly to customers—observe their behavior instead. His “Sales Safari” method focuses on ethnographic detective work, uncovering hidden truths. A visualization software firm doubled revenue when they realized their product was used more for board presentations than operational reports. Knowing demand isn’t about features; it’s about divorcing your assumptions from reality.
  • Read More

• America Is Missing The New Labor Economy – Robotics Part 1 ($) - Dylan Patel
  • Robotics isn’t about tech—it’s manufacturing capacity. China’s control of 90% of global permanent magnets means robot arms cost half as much there. The detail that hits hardest: DJI crushed GoPro’s drone ambitions because Chinese engineers iterated in hours, while Americans shipped prototypes across oceans. Robotics won’t simply replace workers—it’ll build systems that make even better systems.
  • Read More

• Reality has a Surprising Amount of Detail - John Salvatier
  • Salvatier dismantles our illusion of simplicity: building stairs reveals endless tiny complications. Wood warps unexpectedly. Screws rarely go straight. Water doesn’t just boil—it shifts subtly through multiple stages. His sharpest insight? Important details are invisible until noticed; afterward, they’re impossible to ignore. Mastery isn’t about big ideas, but noticing the small things everyone overlooks.
  • Read More

Artificial Intelligence

• Anthropic Economic Index: AI’s Impact on Labor - Anthropic
  • Anthropic analyzes anonymized Claude.ai usage across millions of conversations to reveal which occupational tasks are most often supported by AI (particularly coding and technical writing), how AI is more about augmentation (57%) than total automation, and why mid-range salary roles are among the heaviest adopters – all to inform ongoing policy and research on AI’s labor transformations.
  • Read More

• Apple AI’s Platform Pivot Potential - Ben Thompson
  • Ben Thompson contrasts Apple’s delayed “Apple Intelligence” with the impressive M3 Ultra chip, arguing that instead of pushing a Siri-centric aggregator model, Apple should embrace a true platform play – letting developers build powerful local AI experiences on Apple Silicon, which uniquely blends high-performance hardware and integrated system design.
  • Read More

• Gemini 2.5 Pro and Google’s Second Chance with AI - Nathan Lambert
  • Nathan Lambert explains how Google’s Gemini 2.5 Pro delivers a substantial performance leap, recapturing its standing among “frontier” AI labs while reminding leaders that a single breakthrough must be matched by equally bold product and platform execution. He notes that Google’s real challenge lies in unifying its vast ecosystem—balancing deep expertise, swift updates, and the internal resolve to exploit AI’s full potential. This moment underscores how leadership demands not merely technological prowess but also the willingness to pivot business models and organizational priorities in pursuit of transformative innovation.
  • Read More

• Inference Scaling and the Log-x Chart - Toby Ord
  • Toby Ord highlights that recent AI "inference scaling" charts often use a log scale on the x-axis, making progress look linear (impressive) when it's actually exponential in cost. He cautions that better benchmarks could clarify how new models (like o1/o3) differ from brute force, but without overlapping data or direct comparisons, such charts risk overstating AI gains from simple test-time compute.
  • Read More

• Elicitation, the Simplest Way to Understand Post-Training - Nathan Lambert
  • Nathan Lambert proposes that large language models already contain vast capabilities from pretraining, and that post-training (via instruction tuning or reinforcement learning) mainly draws out these latent skills rather than creating them anew. He counters the idea of “superficial alignment,” showing how even smaller datasets or RL can unlock major performance gains when carefully applied. This approach reframes post-training as harnessing existing potential rather than teaching completely new behaviors.
  • Read More

Startups

• Speedrunning the Skill of Demand - Cedric Chin
  • Cedric Chin shares how a focused “Sales Safari” method—observing real-world customer pains and worldviews—can rapidly sharpen a team’s skill at uncovering and meeting market needs. He demonstrates that true mastery requires detachment from one’s product and a relentless hunt for genuine customer motivations, ultimately tying effective sales, marketing, and product design to deeper empathy for the buyer’s self-improvement goals. The series foreshadows a deeper dive into the Jobs to Be Done framework and how it can further refine this approach.
  • Read More

• Lopsided AI Revenues - Tomasz Tunguz
  • Tomasz Tunguz compares Q4 revenue from AI hardware (notably NVIDIA’s $31B data center business) and software/consulting (Microsoft, IBM, OpenAI), finding hardware income at least four times larger, signaling GPU sales far outstrip AI software services – though he expects years of strong software growth ahead.
  • Read More

• Klarna Chameleon - Net Interest
  • Born from an invoice-factoring insight in Sweden, Klarna evolved from funding merchants with delayed payouts into an ever-changing global fintech offering pay-later credit, bank deposits, and even ad sales. This repeated shape-shifting highlights the tension between chasing scale—by rebranding, pivoting products, and switching from merchant focus to consumer focus—and preserving a cohesive identity that fosters lasting credibility. As Klarna now eyes a Wall Street IPO, investors wonder whether its chameleon instincts reflect unstoppable adaptability or ongoing search for true form.
  • Read More

Semi Conductors

• AI Promise and Chip Precariousness ($) - Ben Thompson
  • Ben Thompson highlights rapid AI gains from labs like Anthropic, DeepSeek, and OpenAI while noting that U.S. chip leadership remains tied to TSMC in Taiwan. He warns of potential fragility as China advances its homegrown chips, leading to a global dynamic where the U.S. relies more on Taiwan even as China’s dependence diminishes.
  • Read More

• Nvidia GTC and ASICs, The Power Constraint, The Pareto Frontier - Ben Thompson
  • Ben Thompson recounts Jensen Huang’s argument that reasoning-driven inference has skyrocketed AI compute needs and how Nvidia’s software-hardware synergy—via Dynamo and scalable GPUs—pushes efficiency to the “Pareto Frontier.” Huang highlights “iso power” constraints to demonstrate that every new Blackwell architecture iteration outperforms competing ASICs in tokens-per-second throughput without choking on user response times. By dynamically balancing prefill and decode tasks, Nvidia aims to lock in leadership even as specialized chips threaten its market.
  • Read More

Technology and Society

• The Trump Administration Group Chat, Signal: The Coke of Secure Messaging, Transparency Versus Security - Ben Thompson
  • Ben Thompson recounts how top U.S. officials inadvertently leaked classified plans for bombing Yemen in a Signal group chat that included The Atlantic’s editor, illustrating the friction between end-to-end encryption and mandated record-keeping. He spotlights the deeper question of whether absolute transparency stifles candid debate, especially when every digital exchange may become a permanent public record. Leaders face tough trade-offs between robust security, personal accountability, and the discipline needed to make pivotal decisions under scrutiny.
  • Read More

• When and How Will AI Offer People Search? - The Diff
  • The Diff suggests that next-gen AI could enable “reverse name search,” letting users discover relevant strangers for hiring or networking, yet raises privacy concerns akin to facial recognition. Meanwhile, industries with scarce online resources stand to benefit by publishing niche knowledge for LLMs to index, effectively becoming go-to references in specialized domains.
  • Read More

• Microsoft EOLs Skype, Skype’s Founding, Microsoft’s Skype Charity - Ben Thompson
  • Ben Thompson recounts Skype’s early peer-to-peer magic and how Microsoft’s belated attempts to modernize it were doomed by the rise of mobile-first messaging like WhatsApp. He argues Microsoft overpaid for an already-declining service, effectively subsidizing Skype’s survival until now, when it finally ends in favor of Teams.
  • Read More

• America Is Missing The New Labor Economy – Robotics Part 1 ($) - ¥ Dylan Patel and co-authors
  • ¥ Dylan Patel and co-authors argue that advanced robotics — especially “general purpose” machines able to adapt and self-improve — will revolutionize manufacturing, but warn the U.S. is lagging as China aggressively scales up robotics, magnets, batteries, and high-volume hardware production. They detail how China’s industrial base, policy support, and fast iteration cycles give it a critical edge in developing fully automated, lights-out factories and next-gen humanoids, putting Western manufacturing competitiveness at existential risk.
  • Read More

• Who Wins Nobel Prizes? - Brian Potter
  • Brian Potter details how most Nobel Prize-winning work—often done in the U.S., U.K., and Germany at elite institutions—contrasts sharply with a long tail of lesser-known contributors worldwide. He shows that winners are receiving the award later in life, even though their groundbreaking work generally happens at a younger age, prompting questions about whether delayed recognition signals fewer radical discoveries or simply a backlog of deserving candidates. Leaders and aspirants alike must weigh the personal commitment needed for excellence against the reality that truly significant achievements may remain uncelebrated for decades.
  • Read More

Finance and Economics

• The Intelligent Investor — Jason Zweig - Jason Zweig
  • Zweig uses a childhood memory of almost blowing off his fingers with a firecracker to warn against the thrill-seeking edge of high-risk speculation, then dissects how “credit” literally means belief and underscores the faith investors place in opaque private-credit valuations. It’s a reminder of the illusions and leaps of faith in investing—whether it’s one-day option trades or private credit bets that hinge on trust rather than transparent markets.
  • Read More

• Massive AI spending: a red flag? - Owen Lamont
  • Owen Lamont explores how historically, large waves of investment—especially fueled by exuberant narratives—often foreshadow disappointment. While today’s “massive AI spending” may look bold, he warns that overbuilding and collective overoptimism can lead to subdued or even negative future returns, highlighting that fervor alone doesn’t guarantee genuine long-term gains.
  • Read More

• Buy-and-Hold is More Active than it Looks - Byrne Hobart
  • Byrne Hobart explains that holding a position for years involves repeatedly rejecting fresh reasons to sell, requiring consistent scrutiny of both fundamentals and management credibility. He argues that a concentrated, low-turnover strategy can internally diversify over time as successful companies evolve, yet it also demands a steady hand through deep drawdowns. Investors, he concludes, are effectively paid to maintain strong conviction while continuously re-evaluating whether their core theses still hold.
  • Read More

• KKR’s Berkshire Dreams ($) - Marc Rubinstein
  • Marc Rubinstein shows how KKR, once reliant on traditional private equity cycles with institutional investors like Washington State’s pension fund, is reinventing itself by courting retail capital and holding long-term stakes in a “mini Berkshire” strategy. He notes that slowing distributions and limited exit options have prompted KKR to diversify beyond buyout funds, leveraging permanent capital structures and a growing retail base to sustain growth. Their pivot highlights the tension between classic private equity’s close-ended model and the aspiration to compound investments over decades in a more Berkshire Hathaway-like fashion.
  • Read More

Personal Development

• Outcome Orientation as a Cure for Information Overload - Cedric Chin
  • Cedric Chin’s “outcome orientation” is a discipline of persistently asking “What am I trying to achieve here?”—it surpasses standard “info diets,” letting you glean crucial insights from chaotic or adversarial sources while staying mentally clear and purposeful about data consumption.
  • Read More

• Reality has a Surprising Amount of Detail - John Salvatier
  • John Salvatier recounts how a seemingly straightforward task—building basement stairs—becomes unexpectedly intricate when materials warp, angles can’t be assumed, and small misalignments cascade into bigger issues. Through these vivid moments of frustration and discovery, he argues that what first appears simple nearly always hides critical nuances; people often get “stuck” precisely because they don’t realize what’s invisibly shaping their assumptions. By consciously seeking out overlooked details in everything from home projects to conversations, we can avoid hidden traps, expand our perspective, and stay open to the subtle factors that truly drive how reality unfolds.
  • Read More