Gianfranco's Best of July 2025 Reading List

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

This month, I've curated my favorite essays offering insights into the latest developments in AI, societal impacts of technology, strategic business thinking, financial trends, and personal growth.

If you only have a few minutes, these three posts were my favorite, and are included in the list below:

• AI Market Clarity - Elad Gil
  • Elad Gil argues that several first-wave generative-AI segments have “locked in” likely short-term champions: LLM foundations, coding assistants, legal, medical scribing, customer support and AI-search. Massive capital moats plus hyperscaler alliances make fresh LLM entrants unlikely. Next frontiers—accounting, compliance, finance, sales, security—remain open, with “agentic” workflows shifting revenue from seat-based SaaS to units of cognition and enabling AI-driven roll-ups.
  • Read More

• Tech Philosophy and AI Opportunity - Thompson
  • Thompson argues that AI’s riches still hinge on scarcity—GPUs and elite researchers—so strategy follows company creed: Apple’s “bicycle-for-the-mind” ethos keeps it frugal on chips and losing talent, while Meta’s do-it-for-you worldview justifies spending athlete salaries to build agentic models. Google chases its founding “I’m Feeling Lucky” AI quest despite search risk; Microsoft weds tooling Copilots to OpenAI, whereas Anthropic joins Meta in full-automation ambition. Hiring flows map precisely to these philosophical quadrants.
  • Read More

• If Writing is Thinking… - Steven Sinofsky
  • Steven Sinofsky warns that corporate writing already goes unread, and generative AI will turbo-charge the problem: authors may not deeply review AI-drafted memos, while recipients will rely on lossy AI summaries, eroding shared context and accountability. His career of 20-page strategy memos showed even “big boss” documents get skimmed; outside tech, peer-reviewed science, Wall St. reports, and org planning suffer the same fate. Without deliberate culture and human engagement, AI-authored/AI-read documents risk becoming a compiled codebase nobody understands, undermining decision quality across industries.
  • Read More

Artificial Intelligence

• AI Market Clarity - Elad Gil
  • Elad Gil argues that several first-wave generative-AI segments have “locked in” likely short-term champions: LLM foundations, coding assistants, legal, medical scribing, customer support and AI-search. Massive capital moats plus hyperscaler alliances make fresh LLM entrants unlikely. Next frontiers—accounting, compliance, finance, sales, security—remain open, with “agentic” workflows shifting revenue from seat-based SaaS to units of cognition and enabling AI-driven roll-ups.
  • Read More

• Meta Superintelligence - SemiAnalysis
  • SemiAnalysis details how Mark Zuckerberg has gone into “founder-mode” after Llama 4’s misfire, attacking Meta’s two weak spots—compute and talent—by: (i) ripping up its datacenter blueprint to erect tent-style 1-2 GW GPU campuses (Prometheus and Hyperion) with on-site natural-gas power and Arista/ Broadcom fabrics; (ii) spending >$200 M per star researcher and buying 49 % of Scale AI to fix data/eval gaps; and (iii) forming a small, incentive-loaded “Superintelligence” team poached from OpenAI/Anthropic.
  • Read More

• DeepSeek Debrief: >128 Days Later ($) - SemiAnalysis
  • DeepSeek’s bargain-bin pricing was never the story; throttled latency, 64 K context, and aggressive batching reveal a company hoarding scarce GPUs for R&D while letting third-party “inference clouds” carry the user load. Traffic to its own app flatlined, yet token consumption via open hosts is up 20×, proving price per token is an output of compute politics, not generosity. Anthropic faces a rhyming squeeze—slower Sonnet speeds as demand outstrips its AWS + TPU allotment.
  • Read More

• Your Prompt Is the Product: Working with o3 & o4-mini-high in the Agentic Era - Nate
  • Nate argues OpenAI’s new o-series rewires prompting from “persuasion” to “configuration.” MODE (reflection / action / agentic) and EFFORT (quick / deep) behave like program flags: set them and the model reliably toggles depth, tool use, and bias for action. Super-human vision, probability-native answers, and stubborn frame anchoring offer power and peril; a four-block prompt (Purpose, Instructions, Reference, Output) plus the two switches tames both, turning prompts into lightweight systems rather than clever requests.
  • Read More

• Tech Philosophy and AI Opportunity - Thompson
  • Thompson argues that AI’s riches still hinge on scarcity—GPUs and elite researchers—so strategy follows company creed: Apple’s “bicycle-for-the-mind” ethos keeps it frugal on chips and losing talent, while Meta’s do-it-for-you worldview justifies spending athlete salaries to build agentic models. Google chases its founding “I’m Feeling Lucky” AI quest despite search risk; Microsoft weds tooling Copilots to OpenAI, whereas Anthropic joins Meta in full-automation ambition. Hiring flows map precisely to these philosophical quadrants.
  • Read More

Semiconductors

• The post about GPUs (An Approachable Primer, $) - read.technically.dev
  • GPUs thrive on AI’s “onion-mandoline” math: billions of trivial matrix ops processed in parallel, a workload CPUs fumble and NVIDIA’s CUDA stack orchestrates with near-monopoly lock-in. Scarcity isn’t just TSMC yield—it’s capital, coolant, and two decades of tribal software knowledge that gate newcomers. Cloud ASICs (TPU, Inferentia, Groq) prove faster in niches yet trade speed for vendor captivity, leaving renters—not owners—holding the silicon.
  • Read More

• How Oracle Is Winning the AI Compute Market ($) - SemiAnalysis
  • Ellison turned Oracle from lagging cloud vendor into “investment-grade neocloud” by front-loading risk: 15-year, multi-GW leases in Abilene and Johor secured OpenAI and ByteDance before hyperscalers could self-build. Pairing Foxconn ODM servers with in-house RoCE networking shaves ~20 % CapEx, yielding mid-40 % EBIT on bare-metal GPU deals even at $2.6/chip-hour. The margin is real but not durable; once capacity loosens, price wars will test Oracle’s six-year depreciation bet.
  • Read More

Finance and Economics

• Secondaries - The Diff
  • Venture's liquidity crisis birthed an entirely new asset class. When hold periods stretched from 3 to 10+ years post-2000s, Facebook's pre-IPO secondary trading created the infrastructure template—now companies go public incrementally through controlled employee buybacks and LP stake transfers. The public/private binary collapsed into a continuum where liquidity correlates inversely with strategic flexibility.
  • Read More

• #33 Episode – QSBS 2.0 is Here - Law of VC
  • The One Big Beautiful Bill Act slashes QSBS’s holding period to a 3-year “glide path” (50% tax-free at 3 yrs, 75% at 4, 100% at 5), lifts the lifetime exclusion to $15 M per taxpayer and raises the asset cap to $75 M (both CPI-indexed from 2027). New stock issued on/after 5 July 2025 qualifies, giving founders, employees and VCs faster exits and bigger shields—while late-joining LPs, older SAFEs and California residents still lose out. Critics note 74 % of gains will keep flowing to $1 M-plus earners, but supporters see a turbocharge for Series A/B capital and a “second-chance” window for companies that once exceeded the old $50 M limit.
  • Read More

• I Summarized Mary Meeker’s Incredible 340 Page 2025 AI Trends Deck - Nate’s Newsletter
  • Meeker’s deck hails an “unprecedented” AI curve—record user uptake, cap-ex splurges, plummeting inference costs—but Nate argues adoption masks friction: value sticks only where workflows, proprietary data, and human judgment intersect. He highlights eight gaps—consumer hype vs. B2B ROI, agent fragility, tacit knowledge, talent scarcity, policy lag—urging operators to own pixel-level use cases, distribution, and audit-ready instrumentation before the economics re-price again.
  • Read More

Startups

• The roadmap to Product/Market Fit (PMF)… maybe - Smart Bear (Longform)
  • Cohen distills WP Engine’s journey into an eight-step playbook: align your venture with a personal edge, vet the market with Fermi-style estimates, define and interview an ideal customer profile, ship a simple-lovable-complete (SLC) v1 fast, prioritize marketing and sales over code, drive retention by fixing the promise-delivery gap, use ruthless “rocks-pebbles-sand” prioritization, and manage founder psychology amid constant rejection. Skipping steps courts luck; following them compounds learning into PMF.
  • Read More

• Putting the Jobs-to-be-Done Interview to Practice - Commoncog
  • Chin’s field report warns that JTBD interviews look simple but break novices on two rocks: faded customer memory and self-justifying narratives. Mastery demands tight recency filters (< 6 months), scripted memory cues, two-person interview teams, and rehearsal reps before every project. Post-call, fill Moesta’s timeline immediately—gaps expose flawed questioning. Commoncog’s own run flipped its growth model: 57 % of memberships came from binge readers hitting a paywalled series, not newsletter converts, redirecting product and funnel experiments toward series discovery.
  • Read More

Science and Society

• Stop Pretending You Know What AI Does to the Economy ($) - Noahpinion
  • U.S. commentary is stuck on an “AI kills jobs” brainworm: headline scares about new-grad unemployment and consulting reports on automation crumble under broader data showing historically strong labor markets, rising tech hiring, and studies that robots/AI typically add employment by creating complementary tasks. High-profile pessimists often hard-code negative assumptions (e.g., no new tasks) rather than observe evidence. America’s unique cultural anxiety risks self-sabotage just as rivals press ahead — policymakers and media should ditch knee-jerk alarmism and track real-time outcomes instead.
  • Read More

• How to take our country back ($) - Noahpinion
  • Smith argues that America’s social unraveling isn’t ideological but technological: the smartphone-social-media complex radicalized elites and sidelined the tolerant majority. Since the 2010s, algorithm-driven feeds empowered extremist influencers, foreign meddlers, and teenage trolls, eclipsing the stabilizing gatekeepers of old mass media. He frames social media as a runaway “shoggoth” we built and proposes counter-tech tactics—school phone bans, sentiment-based feed dampers, and a forced sale of TikTok—to reclaim national cohesion.
  • Read More

• Should the Federal Government Sell Land? - Construction Physics
  • Potter’s GIS mash-up shows that of 350 M federally owned acres in the western states, 70 % lies more than 20 miles from any town and 60 % is in parks, forests or refuges. After removing protected and mountainous tracts, only 26 M acres—under 8 %—are flat, unprotected and near people. These pockets cluster mostly around Las Vegas, with smaller buildable rings near Phoenix, Boise and Tucson. Opening them could add some housing land, but it would barely dent statewide shortages.
  • Read More

• If Writing is Thinking… - Hardcore Software
  • Steven Sinofsky warns that corporate writing already goes unread, and generative AI will turbo-charge the problem: authors may not deeply review AI-drafted memos, while recipients will rely on lossy AI summaries, eroding shared context and accountability. His career of 20-page strategy memos showed even “big boss” documents get skimmed; outside tech, peer-reviewed science, Wall St. reports, and org planning suffer the same fate. Without deliberate culture and human engagement, AI-authored/AI-read documents risk becoming a compiled codebase nobody understands, undermining decision quality across industries.
  • Read More

Life

• Encore Anxiety - Working Theorys
  • Anu Atluru describes the paralysis that follows early acclaim: once people expect brilliance, creators fixate on preserving reputation instead of pursuing curiosity, leading to “encore anxiety”—a cruel twin of impostor syndrome. Social-media algorithms and always-on audiences amplify the fear, pushing artists toward audience capture and risk-aversion. The cure is to treat each project as independent, prioritize truth-seeking over perception management, and accept that disappointing some past fans is the price of future originality.
  • Read More

• Flounder Mode - Colossus
  • Kevin Kelly’s “Hollywood-style” career shows founders that serial, curiosity-driven projects can compound into cultural leverage without unicorn tunnel-vision. By treating each interest as a limited-run production, he’s built WIRED, foresaw quantified-self devices, and coined “1,000 True Fans” — all while keeping joy, breadth, and family intact. The essay implies that optimizing for sustained enthusiasm, not brutal singularity, is a viable strategy for long-term creative impact.
  • Read More

• New-Dad Reality Check - Good Anger
  • In a playful “pros & cons” inventory written five months into fatherhood, Sam Parker counters viral gloom-and-doom takes about parenting. Cons? Basically stalled video-game progress. Pros? An avalanche of visceral joys: snot-free newborn sneezes, frog-leg chest naps, scalp scent euphoria, diaper-table eye contact, pure baby giggles, nightly reunion highs, turbo-charged purpose (“lock the fuck in”), pram masculinity, public kindness, rediscovered fun, effortless pride, awe for a partner’s performance, and clarifying life perspective—one mission: help this tiny human thrive.
  • Read More

• Niche > “Well-rounded” - Sivers
  • Derek Sivers argues that to pierce the world’s thick “squishy pile of apathy,” creators must present a razor-sharp concept, not an all-things-to-all-people blur. Generic “My Songs” artistry is unmemorable; a hyper-specific hook (e.g., Sushi, Soufflé, and Seven Other Songs About Food) gives media and audiences something to grasp, share, and recall. Focus on one vivid angle per project, then pivot over time—Bowie, Miles, Madonna, Prince, Joni, and Paul Simon all built expansive legacies through successive, clearly defined phases. Be a knife, not a sphere: cut through, make a point, repeat.
  • Read More

• Are you stuck in movie logic? - Useful Fictions
  • Hall argues that many real-life conflicts persist because we imitate film tropes in which characters avoid plain speech; relief and progress come from naming the tension early, even before we fully grasp it. She proposes three moves—ask what deeper issue the “plot” obscures, voice your fear of raising it, and share half-formed intuitions—to replace silent strain with collaborative clarity. Ignoring problems, she warns, corrodes agency and honest self-assessment.
  • Read More