2026 April & May - What I Have Learned

Posted on

Blogs and Articles

The Art of Asking Smarter Questions - Arnaud Chevallier, et al., Harvard Business Review [Link]

Using these five techniques helps us ask smarter questions by transforming our questioning from a reactive habit into a deliberate strategy. Instead of just asking whatever pops into our head, we can look at a situation and decide exactly what kind of thinking our team or project needs at that moment.

  1. Investigative: What’s Known?

    • Purpose: Digs deep to generate nonobvious information and clarify the core problem or opportunity.

    • Example prompts: What happened? What is and isn't working? What evidence supports our proposed plan?

    • The Goal: To avoid surface-level assumptions. It often utilizes successive "Why?" or "How?" questions to uncover overlooked data.

  2. Speculative: What If?

    • Purpose: Helps you consider a problem more broadly, reframe the issue, and explore creative, alternative solutions.

    • Example prompts: What other scenarios might exist? Could we do this differently? What potential solutions have we not considered?

    • The Goal: To overcome limiting assumptions and jump-start innovation (often using prompts like "How might we...?").

  3. Productive: Now What?

    • Purpose: Assesses the availability of talent, capabilities, time, and resources needed to execute a strategy.

    • Example prompts: What is the next step? Do we have the resources to move ahead? Are we ready to decide?

    • The Goal: To identify metrics, milestones, and potential capacity bottlenecks to keep plans on track.

  4. Interpretive: So, What…?

    • Purpose: Focuses on sensemaking and synthesis by pushing you to continually redefine the core issue and draw out the true implications of your data.

    • Example prompts: What did we learn from this new information? How does this fit with our overarching goal?

    • The Goal: To convert raw information into actionable insight and ensure everything ladders back up to the main mission.

  5. Subjective: What’s Unsaid?

    • Purpose: Deals with the human element—personal reservations, frustrations, hidden agendas, and emotional alignments that can derail a decision.

    • Example prompts: How do you really feel about this decision? Are all stakeholders genuinely aligned?

    • The Goal: To clear out "pluralistic ignorance" (where team members hide misgivings because no one else is speaking up) by creating a safe space for dissenting views.

Transformations That Work - Michael Mankins, Patrick Litre, Harvard Business Review [Link]

Summary:

  1. Treating transformation as a continuous process: Instead of viewing change as a discrete program with a strict beginning and end (the traditional "unfreeze-change-refreeze" model), successful companies recognize that they must operate in a state of constant transformation and manage an evergreen backlog of issues.

  2. Building it into the company's operating rhythm: Transformation shouldn't be isolated in a separate program-management office. It must be woven directly into the executive team's regular, weekly operating routines and plan reviews so that managing change becomes part of everyone's day job.

  3. Explicitly managing organizational energy: Because transformations fail when they exhaust employees, successful programs carefully sequence initiatives. They ensure no single function or group is forced to change too many primary routines at the same time, and they reward milestones to sustain morale.

  4. Using aspirations, not just targets, to stretch thinking: Relying strictly on external benchmarks often sets the bar too low and limits "the art of the possible." True transformations rely on bold ambitions and breakthrough thinking to fundamentally reshape business models.

  5. Driving change from the middle out: Top-down mandates can be superficial, and bottom-up efforts often just trim around the edges. Midlevel executives possess the unique combination of frontline experience and strategic context needed to design and execute meaningful, lasting structural changes.

  6. Accessing substantial external capital from the start: Transforming a business is expensive and cannot be reliably funded through internal cost-cutting alone. Successful efforts tap the capital markets early to ensure the initiative is fully funded to fuel growth.

Make Decisions with a VC Mindset - Ilya A. Strebulaev, Alex Dang, Harvard Business Review [Link]

To spur innovation within a traditional organization using the VC mindset, the article highlights that we don't need to apply these rules to routine decisions in predictable environments. Instead, we should deploy them strategically in times of high uncertainty, disruption, or when developing radically new products.

  1. A High Comfort Level with Failure

    The VC Skill: VCs accept that up to 80% of their investments will fail. Their business model relies on the understanding that home runs matter and strikeouts don't; one massive success can cover all other losses combined.

  2. Prioritizing the Individual over the Group

    The VC Skill: VCs seek out contrarian ideas, recognizing that breakthroughs often come from a single person in the room rather than a group. They actively structure environments to prevent individuals from being drowned out by group consensus. They achieve this by:

    • Keeping teams small (often 3 to 5 partners) to ensure clear communication and accountability.
    • Collecting independent feedback in advance (sometimes blindly) before meetings to avoid bias.
    • Allowing junior team members to speak first so they aren't swayed by senior executives.

  3. Valuing Disagreement over Consensus

    The VC Skill: Research cited in the article shows that VC firms requiring unanimous approval actually perform worse. High-performing VCs intentionally lean into friction and unresolved opposition. For instance, some use a "consensus minus x" rule, while others (like Venrock) allow the lead partner to make the final investment decision unilaterally, regardless of other partners' skepticism. They also intentionally assign a "devil's advocate" or "red team" to argue against deals.

  4. Championing Exceptions over Dogma

    The VC Skill: VCs allow for workarounds and give individuals the power to keep a controversial idea alive even if the broader team rejects it. This includes tools like the "anti-veto," where a leader can unilaterally move a project forward despite a negative team vote to protect unconventional opportunities (as seen with early investments in Zoom and Kahoot).

  5. Extreme Agility over Bureaucracy

    The VC Skill: VCs move at an incredibly fast pace because they know slow tracking means losing the best deals to competitors. They set aggressive timelines (e.g., funding decisions within days or weeks) and operate with a streamlined, single-tier decision model.

Advice for the Unmotivated - Robin Abrahams, Boris Groysberg, Harvard Business Review [Link]

Based on the provided article, the authors outline a four-step process called DEAR (detachment, empathy, action, and reframing) to help individuals interrupt the cycle of numbness and recover their motivation at work:

  1. Detachment

    Before making major career moves or reacting under stress, you need distance and perspective to clear cognitive distortions.

    • Reflect and break away: Review what went well at the end of the day, then completely disconnect from work using a physical ritual (like closing your laptop or signing out of email).

    • Meditate: Dedicate 10 to 20 minutes twice a day to simple focus or breathing exercises to lower your stress response.

    • Move your body: Use even brief stretches, walks, or exercise to replenish psychological energy and improve mood.

    • Think in the third person: Address yourself by name or use third-person pronouns in your inner monologue to look at your problems more objectively.

  2. Empathy

    Disengagement often causes "depersonalization." Reconnecting with your humanity and the humanity of others helps rebuild motivation.

    • Practice self-care: Acknowledge your values and treat yourself kindly with small daily rituals.

    • Treat people as people: Combat numbness by making eye contact, practicing social niceties, and appreciating colleagues' work.

    • Ask questions: Stay curious about your customers, bosses, and peers to gain fresh perspectives on your role.

    • Look for friends: Cultivate real workplace friendships to make a frustrating job more enjoyable.

    • Help others: Explaining systems to new hires, mentoring, or assisting teammates builds empowerment and has been shown to reduce personal burnout.

  3. Action

    Channel restless or frustrated energy into proactive, productive strategies.

    • Tackle the little stuff: Checking small, mundane tasks off your to-do list builds a sense of momentum and small wins.

    • Invest in outside activities: Hobbies, side hustles, or volunteering provide a sense of satisfaction that carries over into your primary job.

    • Job craft: Redefine your current responsibilities by strategically shifting your focus to tasks that leverage your strengths or interests.

    • Gamify: Turn boring tasks or meetings into mental puzzles, competitions, or games to trigger your competitive drive.

    • Pretend and dress the part: Imagine how a mentor or superhero would handle a situation, and wear clothing that makes you feel confident and professional to help you step into character.

  4. Reframing

    Alter your perspective regarding who you are at work and why your job matters.

    • Examine your work identity: Create an informal title for yourself (e.g., teacher, visionary, logistics expert) that reflects the unique value you naturally bring to a team.

    • Look at the big picture: Focus on the why rather than the how of your tedious daily tasks by connecting them to a higher-order purpose or larger organizational goal.

    • Consider how others benefit: Actively remind yourself of who your work serves—whether it is the clients you help directly or your family members who rely on your income.

Case Study: How Aggressively Should a Bank Pursue AI? - Thomas H. Davenport, George Westerman [Link]

Expert Recommendations

  • Noemie Ellezam-Danielo (Société Générale):
    • Fix the Alignment: The tech team shouldn't dictate strategy alone. Siti should form a cross-functional executive group to align AI with core business goals.
    • Segmented Approach: AI needs to adapt to varied customer behaviors across different demographics and geographies, rather than forcing a blanket 95% digital mandate.
    • Look for Back-Office Wins: Focus on high-impact, lower-risk areas first, such as automating resource-intensive customer identity verifications.
  • Sastry Durvasula (TIAA):
    • Embrace Disruption Safely: NVF needs "Michaels" to challenge the status quo, but the technology must supplement rather than entirely replace human service.
    • Culture & Upskilling Overhaul: The shift requires an AI-savvy leadership culture. Because modern generative AI tools are user-friendly, the focus should be on upskilling existing staff to handle higher-value roles rather than just slashing head count.

Establishing a Solid Digital Foundation for AI-Everywhere Webinar - Harvard Business Review [Link]

Takeaways:

  1. Organizations across all industries are experiencing an unprecedented wave of excitement surrounding generative AI. However, there is a stark gap between a company’s AI aspirations and its actual digital capabilities. While technology and business leaders are being pressured to deploy GenAI solutions rapidly, many lack the infrastructure to support them sustainably.

  2. Alex Clemente reviews data from a recent HBR-AS research report (sponsored by Kyndryl) focused on the current realities of GenAI adoption.

    • Many early-stage GenAI initiatives fail to scale successfully or move out of the proof-of-concept phase because organizations treat AI as an isolated tool rather than an integrated capability.
    • Data quality, governance, and organizational cultural resistance remain the top barriers to meaningful adoption.

  3. To pivot from "AI hype" to "AI success," Howard Miller and Firas Bouz highlight the strategic best practices required to build an enduring digital framework:

    • Data Readiness & Architecture: Before deploying AI models, organizations must clean, unify, and secure their data architecture. AI is only as good as the underlying data feeding it.

    • Modernized Infrastructure: Legacy tech stacks cannot keep up with the compute-heavy, dynamic processing needs of GenAI. Moving toward flexible, cloud-enabled or hybrid infrastructure is essential.

    • Culture and Training: Overcoming user reluctance and bridging the skills gap is critical. True ROI comes from training employees to trust and integrate these technologies into their daily workflows.

Preparing for a Future Powered by Generative AI - Harvard Business Review [Link]

Transforming Consulting Through Generative AI- Harvard Business Review [Link]

Selected takeaways:

  1. Consulting firms are seeing massive speed and efficiency benefits by positioning AI as a collaborative partner rather than a simple chatbot:

    • KPMG: Has deployed gen AI capabilities to over 45,000 employees. Using the technology has cut coding assistance by 30 minutes and competitive intelligence studies by 90 minutes. Surveys show that 60% of users spend more time on high-value work, and 58% report increased creativity.

    • IBM Consulting: Built a platform featuring hundreds of trained AI assistants. A task like mapping a consumer persona—which historically took a human consultant half a day—can now be generated as a baseline within 60 seconds.

    • Omni Business Intelligence Solutions: Used gen AI to craft personalized proposals and marketing material, which freed up enough executive time to drive a 80% increase in closed deals year-over-year.

  2. Succeeding with AI relies heavily on closing user intuition gaps and training knowledge workers broadly. KPMG consumer trust data details a strong generational split in technical literacy and sentiment:

    • 74% of Gen Z and Millennials consider themselves highly knowledgeable about gen AI, compared to only 42% of Gen X and 13% of Boomers/Silent generations.

    • To counteract this discrepancy, companies like KPMG have developed "persona-based learning journeys" led by peer "digital navigators", resulting in over 74,000 completed AI courses across their U.S. workforce.

    • Similarly, PwC has invested $1 billion in gen AI, training over 75,000 employees on responsible prompt design and AI ethics to reduce anxiety surrounding job replacement.

How to Marry Process Management and AI - Thomas H. Davenport and Thomas C. Redman, Business Review [Link]

Selected takeaways:

  1. AI typically supports narrow, specific tasks or subprocesses rather than complete, end-to-end workflows. To transform a whole operation, companies must string multiple AI use cases together. This requires heavy change management, breaking down departmental silos, and establishing common data standards across the organization.

  2. Companies looking to merge process management with AI should follow a structured approach:

    Step 1: Establish ownership. Appoint a dedicated "process owner" who can influence cross-functional teams and coordinate departments (e.g., sales, finance, operations) toward an end-to-end goal.

    Step 2: Identify process customers. Determine exactly who benefits from the process (internal or external). Use generative AI to analyze customer feedback from calls, emails, and social media to find gaps.

    Step 3: Map out the existing process. Use process mining and task mining software to pull real-time data from system logs. This highlights actual bottlenecks and hidden inefficiencies instead of relying on manual guesswork.

    Step 4: Establish performance measures. Define clear metrics (such as cycle times or error rates) and set realistic targets based on what the data shows is possible.

    Step 5: Consider process enablers. Match the right technology to the problem. Use Robotic Process Automation (RPA) for routine, repetitive tasks; use generative AI for unstructured tasks (like drafting contracts); and use traditional machine learning for predictive tasks (like fraud detection or pricing).

    Step 6: Redesign the process. Utilize AI-driven design tools (like digital twins or generative AI process templates) to quickly simulate, build, and optimize new workflows.

    Step 7: Implement and monitor. Continuously monitor the new process using process mining to eliminate variations, adjust to new business requirements, and maintain predictable control.

The Secret to Successful AI-Driven Process Redesign - H. James Wilson and Paul R. Daugherty, Harvard Business Review [Link]

Takeaways:

  1. Generative AI and natural-language interfaces have made powerful tools accessible to nontechnical employees. Business transformation is no longer a niche technical skill; it is now in the hands of frontline workers who can use everyday language to surface data-rich insights and streamline their workflows.
  2. Rather than displacing the workforce, AI acts as a tool that amplifies human creativity, experience, and intuition. Human oversight remains the critical linchpin for ensuring alignment with objectives and refining system design.
  3. The future of continuous improvement lies in autonomous AI agents that exhibit goal-oriented behavior, logical reasoning, planning, and long-term reflection. These agents can independently make decisions, execute complex multistep workflows, and continuously adapt their strategies.
  4. The future of continuous improvement lies in autonomous AI agents that exhibit goal-oriented behavior, logical reasoning, planning, and long-term reflection. These agents can independently make decisions, execute complex multistep workflows, and continuously adapt their strategies.
  5. While Robotic Process Automation (RPA) is easily tripped up by process variations, ecosystems of collaborative, multimodal AI agents can handle highly complex, knowledge-intensive, and manual workflows (such as hospital revenue cycles or enterprise invoice processing) with high accuracy by learning from human demonstrations.

What People Still Get Wrong about Negotiations - Max H. Bazerman, Harvard Business Review [Link]

Takeaways:

  1. Most negotiators mistakenly believe that a negotiation is a zero-sum game—meaning any gain for one side is an automatic loss for the other. This mindset causes people to focus strictly on claiming value rather than creating it, leaving significant profitability and mutual benefits on the table.
  2. To avoid leaving value behind, you must identify all potential issues (beyond just the price or main percentage) and determine their relative importance before the meeting. Creating a weighted score sheet—allocating dollar or point values to different trade-offs—allows you to evaluate complex package offers logically rather than relying on emotional, on-the-spot intuition.
  3. Resolving issues sequentially (whether starting with the easiest or hardest) is a trap. Finalizing one item before discussing the rest prevents you from discovering smart trades across different issues. Keep all items open until you can jointly map out a package that reaches the Pareto-efficient frontier (where neither side can improve without harming the other).
  4. When trust is low or counterparts hide their cards, use these tools to uncover relative priorities:
    • Build trust and share information: Cultivate an open, problem-solving dialogue (especially critical for internal corporate negotiations).
    • Ask diagnostic questions: Instead of asking what they want, ask how they value issue A relative to issue B.
    • Give away information first: Leverage the norm of reciprocity. Sharing your own priorities (not your walk-away bottom line) prompts the other party to share theirs.
    • Make Multiple Offers Simultaneously (MESOs): Propose 3 packages of equal value to you but with different structures. The counterpart's preference will reveal their hidden priorities without them having to tell you directly.
  5. The negotiation doesn't have to end when the deal is signed. Once an initial agreement is locked in, the adversarial tension usually drops. You can propose looking at the deal one more time to see if any new terms can be added to make it even better for both sides, with the strict ground rule that the original agreement stands unless a mutually superior alternative is found.

Leaders shouldn't Try to Do It All - A.G. Lafley and Roger L. Martin, Harvard Business Review [Link]

Takeaways:

  1. Shift from Absolute Importance to Comparative Advantage

    • The Trap: Most leaders prioritize tasks based solely on how important they are to the organization, taking on the top items until they run out of time.
    • The Fix: Leaders should only spend time on activities that nobody else in the organization can do nearly as well. If someone else can do it at parity or better, it should be offloaded—even if it is a critical task.

  2. The Four-Step Process to Manage Time

    The authors suggest a rigorous framework to restructure your calendar:

    • Remove absolute disadvantages: Identify "should do" tasks that you handle simply because of tradition or precedent. If a subordinate possesses equal or better skills for that specific task (e.g., routine faculty hiring or standard investor relations), hand it over completely.
    • Delegate minor comparative advantages: Even if you are highly skilled at something (like creative brand reviews or routine financial management), delegate it if your advantage is only modest. This empowers your team and frees up massive blocks of your time.
    • Take on strong comparative advantages: Reinvest your recovered hours into areas where you can make a unique, decisive difference. For example, A.G. Lafley focused heavily on driving a consumer-centric innovation process, while Roger Martin used his background to write books and articles that elevated his school's external profile.
    • Dedicate time to the irreplaceable: Protect your calendar for high-leverage tasks that literally only the leader can execute, such as establishing massive external strategic partnerships, setting long-term brand visions, and individually mentoring the next generation of leadership.

  3. An organization can expand its capacity, but a leader's time is fundamentally finite. Leading to win means accepting that you cannot do it all, and relentlessly focusing your energy where your personal value-add is greatest.

Want Your Company to Get Better at Experimentation? Learn Fast by Democratizing Testing - Iavor Bojinov, David Holtz, Ramesh Johari, Sven Schmit and Martin Tingley, Harvard Business Review [Link]

Takeaways:

  1. The Necessity of Scaling Up

    • High Failure Rates: Most new ideas do not yield positive results, and it is incredibly difficult to predict which ones will succeed.

    • The AI Imperative: As generative AI drastically lowers the cost of creating new digital user experiences, companies must vastly increase their testing volume—into hundreds or thousands of experiments per year—to remain competitive.

  2. Democratizing the Testing Process

    • The Bottleneck: Relying solely on data scientists to design, run, and analyze every test severely limits an organization’s capacity to scale.
    • The Self-Service Model: Companies must empower product managers, engineers, designers, and marketers to run their own experiments by building or buying platforms with automated guardrails, simple interfaces, and embedded statistical rigor.
    • Redefining the Data Scientist's Role: Data scientists should shift from execution to high-impact work: building platforms, training teams, developing new statistical methodologies, and uncovering overarching strategic patterns.

  3. Transitioning to "Experimentation Programs"

    • Look Beyond Individual Memos: Reviewing tests in isolation wastes time and fails to build institutional knowledge.
    • Learn Across Experiments: Organizations should group tests into programs (e.g., focusing entirely on "search functionality" or "product-detail pages") to identify macro-trends, catch diminishing returns, and determine where to reallocate corporate resources.
    • Build a Knowledge Repository: Implement a centralized, searchable system—potentially enhanced by a generative AI assistant—to store hypotheses, metrics, and long-term insights across the entire enterprise.

  4. Shifting Cultural and Financial Incentives

    • Reward Outcomes, Not Outputs: Evaluating employees on the number of "successful" experiments makes them risk-averse. Instead, incentives should be tied to the overall performance of the business unit.
    • Mitigate Risk Progressively: To encourage high-risk, high-reward testing without fear of breaking systems, companies should leverage automated rollbacks (trip wires based on guardrail metrics) and roll out features in gradual phases.

How Project Leaders Can Tame Unpredictability - Anton Skornyakov, Harvard Business Review [Link]

Takeaways:

Project leaders can tame unpredictability by using the agile technique of vertical slicing—breaking a project down into small, fully functional "slices" to run rapid tests, accelerate learning, and mitigate risks before scaling.

The author outlines how to apply this strategy across four key areas of unpredictability:

  1. Human Behavior

    • The Challenge: People's reactions to organizational change or new initiatives are highly unpredictable.

    • The Takeaway: Deploy a narrowly defined, representative test group to observe real-world behavior and uncover practical issues (e.g., logistical bottlenecks, material defects) before committing a massive budget.

  2. Interpersonal Dynamics

    • The Challenge: Changes affecting career paths, organizational structures, shared responsibilities, or network effects introduce complex emotional and political dynamics.

    • The Takeaway: Unlike behavioral tests, your slice here needs a large enough sample size to trigger and observe those group dynamics, but focused on a small, manageable behavioral shift.

  3. Technological Change and Interoperability

    • The Challenge: Integrating new technology with legacy systems or non-standardized external platforms is a major source of uncertainty.

    • The Takeaway: Create a "tracer bullet"—a minimum viable version of the end-to-end tech solution. Manually integrate just one stream of data (e.g., a single vendor or route) all the way to the consumer-facing frontend to catch integration friction early.

  4. Organizational Interdependencies

    • The Challenge: Relying on multiple internal teams and external stakeholders introduces risks outside of your direct control.

    • The Takeaway: Use an "organizational tracer bullet" to test a single operational scenario across all involved teams. This forces collaboration early and exposes contract or process hurdles before finalizing large-scale agreements.

Sam Altman May Control Our Future—Can He Be Trusted? - The New Yorker [Link]

Substack

Snowflake: AI Consumption Wins - App Economy Insights [Link]

Takeaways:

The enterprise software market is splitting between consumption-based models (which thrive as AI scales) and traditional seat-based SaaS models (which face headwinds as AI potentially replaces human roles).

Snowflake surged 38% after hours, completely shifting the narrative around its position in the AI race. Salesforce stock continued to slide as strong AI adoption metrics failed to translate into clear top-line acceleration.

The differences between Snowflake and Salesforce

  1. Core Product & Purpose

    • Salesforce is primarily a Customer Relationship Management (CRM) application layer. It is built to store and manage customer data, sales pipelines, and support workflows through an interface designed for human workers.

    • Snowflake is an enterprise Data Cloud and Analytics Platform. It is a backend data warehouse and data lake layer built to ingest, store, govern, and analyze massive volumes of diverse structured and unstructured data from all corporate systems (including Salesforce, ERPs, and web logs).

  2. Business & Pricing Model

    • Salesforce traditionally operates on a Seat-Based SaaS Model. Companies pay a fixed subscription price per user license.
      • The AI Challenge: As AI agents automate tasks, companies may need fewer human "seats," threatening Salesforce's core growth model unless they can pivot to charging for AI-delivered work units.
    • Snowflake operates entirely on a Consumption-Based Model. Customers only pay for the exact compute time and storage volume they use.
      • The AI Advantage: Because AI models and agents require massive amounts of compute and data ingestion to learn and execute tasks, Snowflake directly captures more revenue the harder an AI works.

  3. Data Integration and Moat

    • Salesforce operates a suite of fragmented applications (Sales, Service, Marketing, Commerce, Tableau, MuleSoft, Slack) and is trying to centralize them via its Data Cloud. It relies heavily on users inputting and managing data within its application ecosystem.

    • Snowflake acts as a "Clean Sheet" repository. With its newer tools like Cortex Code, it allows organizations to run AI applications and query data that securely sits outside its own databases—including data sitting directly inside Microsoft, SAP, or Salesforce applications.

How SpaceX Makes Money - App Economy Insights [Link]

  1. While SpaceX is seeking a massive \(\$1.5\) to \(\$2\) trillion valuation to raise up to \(\$75\) billion, the underlying segments have dramatically different economic profiles:

    • Connectivity (Starlink): The absolute profit engine. It generated \(\$11.4\) billion in FY25 revenue with an impressive 63% EBITDA margin (\(\$7.2\) billion). It is actively underwriting the company’s massive capital expenditures.

    • Space (Launch): The strategic moat. It reflects a technical monopoly, handling over 80% of global payload weight to orbit. It shows a paper operating loss (\(\$0.7\) billion) only because it completely absorbed \(\$3.0\) billion in Starship R&D.

    • AI (xAI, Grok, X): The capital sink. Folded into SpaceX via a February 2026 merger, this segment brought in \(\$3.2\) billion in revenue but suffered a \(\$6.4\) billion operating loss in FY25.

  2. SpaceX is burning cash at an unprecedented, hyperscaler pace to build out its AI and space infrastructure:

    • CapEx Ramp: Full-year FY25 CapEx hit \(\$20.7\) billion. In Q1 FY26 alone, CapEx reached \(\$10.1\) billion (with \(\$7.7\) billion swallowed by the AI segment for gigawatt-scale clusters like COLOSSUS).

    • Balance Sheet Pressure: Cash reserves dropped from \(\$24.7\) billion to \(\$15.9\) billion in just the first 90 days of 2026.

    • The IPO's Role: This historic \(\$75\) billion IPO isn't for early investor liquidity; it is strictly fuel to sustain this capital-intensive runway before Starship and AI integrations fully monetize.

Warren Buffett has long been skeptical of IPOs. His view is simple: IPOs come to market when sellers choose the timing, not when buyers are likely to get the best deal.

--- How to Invest in IPOs - App Economy Insights [Link]

Instead of falling for the fear of missing out (FOMO) on Day 1, the author suggests a "boring on purpose" strategy:

  1. Avoid the Day 1 Hype: Let the initial open-market volatility pass completely without touching the stock.
  2. Wait for the Second Earnings Call: The initial prospectus (S-1 filing) is just a snapshot. Waiting for two public quarters allows you to see how management handles public market scrutiny, tracks key metrics, and manages guidance.
  3. Nibble in Year One: If you must invest, start with a tiny position. A small starter position lets you follow the company closely without making the inflated IPO price your entire cost basis.
  4. Anchor to Valuation: Don't just look at whether a company is exceptional (like SpaceX or OpenAI)—look at whether its current stock price leaves any room for execution errors.
  5. Give it Time: Truly great businesses will compound for a decade or more. Missing a 20% move on Day 1 is not a disaster; buying at peak euphoria usually is.

The Bottom Line: A hot IPO wave is a sign of a highly optimistic market, not an automatic buy signal. The best strategy for hyped frontier companies is to add them to your watch list, let the insider selling pressure clear, and invest strictly on your own terms.

The Trillion-Dollar Off Switch - App Economy Insights [Link]

Takeaways:

  • On June 12, just days after launching its powerful new "Mythos-class" model Claude Fable 5, Anthropic received a US export-control directive restricting access. Because it couldn't screen users' nationalities in real time, Anthropic took both Fable 5 and Mythos 5 offline worldwide.
  • The government cited national security concerns over potential jailbreaks. While Anthropic claims the vulnerability is mundane, its own defensive program (Project Glasswing) had previously shown Mythos 5 was capable of finding flaws across major operating systems and browsers, highlighting the genuine "dual-use" risk of advanced AI.
  • The incident highlights a new regulatory risk for pure-play AI labs (like Anthropic and OpenAI) heading toward trillion-dollar IPOs. While "building" the frontier model captures the most economic upside, it introduces a single point of failure if a government pulls the plug. "Renting" a model offers a swappable, lower-risk alternative.
  • Apple’s recent WWDC 2026 announcement revealed that its new Siri AI is powered by renting Google's Gemini (paying a reported \(\$1\) billion/year). This strategy allows Apple to skip the massive CapEx arms race and partially insulate itself from direct model-regulation shocks, though it surrenders control over the core AI brain.

Bottom Line: For AI investors, the critical question has shifted from "Who has the best model?" to "Who can keep it online?".

How FIFA Makes Money - APP Economy Insights [Link]

Takeaways:

  • Driven by an expansion to 48 teams and hosting in the lucrative US market, FIFA expects to bring in a record \(\$13\) billion for the 2023–2026 cycle (a 72% jump from the Qatar cycle).

  • Broadcasting Rights: \(\$5.3\) billion (~40%) — FIFA's largest revenue engine.

    Hospitality & Ticketing: \(\$3.6\) billion (~28%) — Heavily boosted by 2026's new dynamic pricing model.

    Marketing & Sponsorship: \(\$3.3\) billion (~25%) — Major brand partnerships.

    Licensing: \(\$0.4\) billion (~3%) — Merchandise and video games.

  • As a non-profit, FIFA targets a near-breakeven budget. About 58% (\(\$7.6\)B) goes directly into staging competitions (including an \(\$871\)M team prize pool), 30% (\(\$3.9\)B) goes to development and education grants for its 211 member federations, and 7% (\(\$0.9\)B) covers governance and administrative costs.

  • For the first time, ticket prices float based on demand. While the cheapest seats remain around \(\$60\), premium seat listings have exploded past \(\$32,000\), raising concerns about pricing out everyday fans from the "people's game."

  • To smooth out its lumpy four-year financial heartbeat, FIFA is branching into new tournaments, notably launching the expanded 32-team 2025 Club World Cup and targeting a \(\$1\) billion revenue mark for the 2027 Women's World Cup in Brazil.

Wall Street's Top Stocks in Q1 - App Economy Insights [Link]

Palantir: Tokens Are the New Coal - App Economy Insights [Link]

Takeaways:

  • Hyper-growth driven by the AI Platform (AIP). Dropping inference costs (cheap tokens) are massively expanding agent workflows, boosting demand for Palantir's safety and governance layer.
  • Revenue surged +85% Y/Y to \(\$1.63\)B. Remaining Deal Value (RDV) nearly doubled to \(\$11.8\)B.
  • Outlook: Management raised FY26 revenue guidance to \(\$7.7\)B (+71% Y/Y).

Amazon: The Inference Era - App Economy Insights [Link]

Takeaways:

  • AWS revenue grew +28% Y/Y to \(\$37.6\) billion (its fastest growth in 15 quarters), signaling that AI monetization is hitting full stride.
  • Amazon is positioning AWS for the "Inference Era." While training is a GPU story, running persistent, multi-step Bedrock Managed Agents requires massive compute power, driving their custom silicon business (Graviton, Trainium) to a \(\$20\) billion annualized revenue run rate.
  • OpenAI frontier models and Codex are entering Bedrock. OpenAI has committed to spending \(\$100\) billion over 8 years on AWS and anchoring its workloads on Amazon's Trainium chips.
  • Amazon is maintaining a \(\$200\) billion CapEx plan for 2026 to build out AI data centers. This infrastructure spend collapsed free cash flow down to \(\$1.2\) billion, prioritizing long-term AI dominance over short-term cash.
  • Retail remains highly efficient, delivering over 1 billion same-day/overnight items so far in 2026, while Advertising surged +24% Y/Y to \(\$17.2\) billion, providing a high-margin cushion for their AI investments.

Google: The Anthropic Paradox - App Economy Insights [Link]

Takeaways:

  • Alphabet plans to invest up to \(\$40\) billion in Anthropic (\(\$10\) billion upfront, \(\$30\) billion in milestones) and provide 5 gigawatts of compute capacity over 5 years.
  • Struck at a \(\$350\) billion valuation, a massive discount compared to Anthropic's recent \(\$1\) trillion secondary-market price.
  • Even though Anthropic's Claude competes with Gemini, Google is prioritizing winning the cloud infrastructure and compute layer over model exclusivity.

Tesla: \(\$25\) Billion AI Pivot - App Economy Insights [Link]

Takeaways - Tesla:

  • Tesla raised its FY26 capital expenditure guidance to \(\$25\) billion (up from \(\$9\) billion in FY25) to fund AI training clusters, "Terafab," and Optimus 3. This heavy spending is expected to turn free cash flow negative for the rest of 2026.
  • Q1 revenue grew 16% Y/Y to \(\$22.4\) billion, and gross margin hit 21% (though aided by one-time warranty and tariff benefits).
  • FSD subscriptions (\(\$99\)/month) and Supercharger usage drove a 42% surge in Services revenue. Unsupervised Robotaxi service expanded to Houston and Dallas.
  • Elon Musk noted that Hardware 3 lacks the memory bandwidth for future autonomy, requiring dedicated retrofit centers and complex split software paths moving forward.

Takeaways - SpaceX:

  • Following its merger with xAI, SpaceX entered a strategic partnership with AI coding platform Cursor. SpaceX has the option to acquire Cursor later this year for \(\$60\) billion (or pay a \(\$10\) billion partnership fee if they decline).
  • The deal aims to fix weaknesses in xAI's Grok chatbot, which Musk admitted was "behind in coding."
  • Cursor gets access to SpaceX’s Colossus supercomputer (200,000 NVIDIA GPUs) to solve its compute constraints, while SpaceX secures a highly popular developer app layer and elite engineering talent.

Apple Enters a New Era - App Economy Insights [Link]

Takeaways:

  • In September 2026, Tim Cook will step down as CEO after 15 years, transitioning to Executive Chairman. Hardware chief John Ternus will take over as CEO.
  • Cook's legacy focused on operational excellence, supply chain efficiency, and high-margin Services. Ternus (an engineer by trade) represents a pivot toward new product ambitions.
  • Ternus inherits a \(\$4\) trillion company with slowing growth (3% average since 2022). He faces high executive turnover, pressure to deliver a mass-market hardware reset (e.g., smart glasses, home robotics), and a costly supply chain diversification away from China.
  • Apple currently relies on Google's Gemini to power Siri. To own its AI future, Apple may need to sacrifice capital efficiency and massively ramp up CapEx, which was only \(\$13\) billion in 2025 (compared to rivals planning \(\$100\)B–\(\$200\)B in 2026).

Anthropic Leapfrogs OpenAI - App Economy Insights [Link]

Takeaways:

  • Anthropic announced an annual revenue run rate (ARR) of over \(\$30\) billion, surpassing OpenAI's reported \(\$24\) billion. However, this is partly due to accounting differences; Anthropic uses a gross method (including cloud provider cuts), while OpenAI uses a net method.
  • Anthropic launched Mythos, a highly restricted, high-premium (\(\$125\)/M tokens) autonomous cybersecurity model, and secured a 3.5 gigawatt compute partnership with Google and Broadcom.
  • Intel partnered with Tesla, SpaceX, and xAI for Elon Musk’s 100-million-square-foot Terafab project in Austin, bringing its 18A node IP and advanced packaging technology to keep the supply chain within the US.
  • OpenAI acquired tech talk show TBPN for hundreds of millions to manage its narrative amid a damaging New Yorker exposé alleging safety neglect and leadership manipulation. Additionally, several top executives (CFO, COO, CMO) are stepping down or shifting roles, with internal rifts over a potential 2026 IPO timeline.
  • Meta launched Muse Spark (internally "Avocado"), its first closed-source model developed via its $14.3 billion Scale AI partnership. The model relies heavily on Meta’s massive distribution network (3.5 billion users) to monetize AI through consumer apps.

The Great AI Rotation - App Economy Insights [Link]

Uber’s Robotaxi Endgame - App Economy Insights [Link]

Takeaways:

  • Uber isn’t trying to build the best self-driving car or technology stack. Instead, it is positioning itself to be the dominant marketplace and commercialization platform that robotaxis run through.
  • Uber has rapidly lined up 8 autonomous partnerships (including NVIDIA, Zoox, Wayve, and Motional). Its latest move is a \(\$1.25\) billion investment in Rivian to deploy up to 50,000 autonomous R2 vehicles by 2030, starting in San Francisco and Miami in 2028.

The Strategic Playbook

  • Through its new Uber Autonomous Solutions (UAS) branch, Uber handles the operational logistics AV developers lack: matching demand, fleet management, charging, maintenance, and financing.
  • Pure robotaxi fleets face severe inefficiencies during low-demand periods and unreliability during spikes. Uber layers AVs into its existing network of ~10 million human drivers to absorb volatility and ensure high asset utilization.
  • AV partners using Uber in early markets (Austin, Atlanta) see a 30% increase in trips per vehicle per day and 25% lower ETAs compared to launching standalone apps.

The Economic Shift

  • Transitioning to an AV model will likely lower Uber's traditional ~30% mobility take-rate. However, Uber expects to offset this by capturing a broader stack of autonomy service fees (ranging from 10–15% for pure distribution up to 15–25% when bundling UAS fleet services).

Key Long-Term Risk

  • Uber’s strategy depends on a fragmented market where tech winners still need a demand aggregator. If a competitor like Tesla (with its Cybercab) or Google's Waymo successfully scales a completely vertically integrated, direct-to-consumer network, they could bypass Uber entirely.

OpenAI Picks a Lane - App Economy Insights [Link]

Takeaways:

  • Shuttering the standalone Sora app and canceling a \(\$1\) billion Disney partnership to cut costs and focus on high-margin enterprise seats ahead of a potential IPO.
  • Merging fragmented products (like Atlas browser and Codex) into a single desktop super app.
  • Stepping back from direct in-app payment processing to act strictly as a discovery/referral layer.

Micron: Demand Goes Vertical - App Economy Insights [Link]

Takeaways:

  • Q2 FY26 revenue skyrocketed 196% Y/Y to \(\$23.9\) billion. Next-quarter revenue guidance (~\(\$33.5\) billion) exceeds any full-year revenue in the company's history prior to 2024.
  • Gross margins hit 74% and are guided to 81% next quarter, edging out Nvidia's current profitability levels.
  • To keep up, Micron raised its FY26 CapEx forecast to over \(\$25\) billion and warned that FY27 spending will step up by another \(\$10\)+ billion, causing some investor unease.
  • The memory deficit has expanded to standard PC and phone manufacturers. Industry leaders project that supply tightness could persist for four to five more years.

YouTube and Podcasts

OpenAI President Greg Brockman: AI Self-Improvement, The Superapp Bet, Path To AGI, Scaling Compute - Alex Kantrowitz [Link]

Takeaways:

  1. OpenAI is deprioritizing standalone video generation (Sora) to concentrate its limited compute resources on a single, unified Super App that integrates chat, browsing, and advanced coding tools.
  2. Coming later this year, OpenAI plans to deploy an automated AI researcher designed to execute the end-to-end tasks of a human research scientist in silicon, accelerating autonomous model development under human oversight.
  3. For individuals worried about job security, Brockman's core advice is to move past the "blank box" intimidation layer and intentionally develop a sense of agency.The future workforce will thrive by acting as "CEOs" who orchestrate, delegate to, and oversee fleets of specialized AI agents.

Artemis II, Jamie Dimon’s “American Dream,” Snap’s Crucible Moment | Diet TBPN [Link]

Takeaways:

  1. SpaceX officially filed for an IPO, immediately setting it up to enter the public markets as a trillion-dollar company.

  2. JP Morgan Chase announced its "American Dream Initiative," a massive commitment to support small businesses, homeownership, and healthcare access. The bank plans to add 3 million new small business customers and lend up to \(\$80\) billion over the next 10 years.

    In a rare senior outside hire, Dimon recruited Warren Buffett’s protégé and former Geico CEO, Todd Combs. Combs will head up a new \(\$10\) billion Strategic Investment Group focused on onshoring industries that America has outsourced (e.g., defense tech, US semiconductors, aerospace, and energy).

  3. Nestle confirmed that thieves in Italy pulled off a brazen heist, stealing a delivery truck packed with 12 metric tons (413,000 units) of Kit Kat bars bound for Poland. Rather than burying the potentially embarrassing security breach, Nestle leaned into the internet humor by joking that thieves took their "Have a break" slogan too literally. This prompted other major brands (like Domino’s UK and Ryanair) to join the viral meme bandwagon, turning a logistics crisis into a corporate PR victory.

What Happens When Every CEO Becomes Omnipresent? | This Week in AI [Link]

Takeaways:

  1. Large Tabular Models (LTMs): LLMs rely on positional encoding where the order of words matters. For deterministic database tasks (like cancer diagnosis or fraud tracking), changing column orders shouldn't change the output. LTMs solve this by focusing entirely on processing raw, structured structural data without lossy human rollups.
  2. Nick Harris notes that chip performance no longer simply doubles every 18 months. Progress now relies on building larger chip packages and networking them together. Traditional copper cables can't travel far and are creating massive, dangerously heavy, megawatt-powered datacenter racks. Lightmatter uses glass fiber optics (photonics) to shoot 1.6 terabits per second over a single fiber, letting thousands of GPUs collaborate over a distance of a kilometer as if they were a single brain. Using optics instead of copper can triple the time it takes to train foundation models, aggressively accelerating the rate of AI takeoff.
  3. Victor Riparbelli explains that video is currently a broadcast medium—one version shipped to everyone. Synthesia is shifting toward real-time, interactive video workflows where canvas models, synthetic diagrams, and conversational avatars roleplay and adapt to users dynamically. The current bottleneck for these immersive, custom-generated experiences is compute and bandwidth. Personalizing a one-hour custom movie right now would roughly cost \(\$700\) in compute tokens.
  4. The panel fiercely debates whether small companies should "vibe code" their own systems. Jeremy shares that his 15-person team entirely vibe-coded their own custom CRM (Fetch) inside Slack instead of buying Salesforce. Countering this, Victor and Nick argue that the hidden cost of vibe coding is the "focus cost" and token spend required for code verification. If open-source agents break or delete codebases, the developer hours spent fixing them often eclipse a simple SaaS subscription.
  5. The fundamental difference between this wave and past industrial revolutions is that society is automating cognition rather than physical work. AGI feel hidden to the average populace because most people—including many engineers—are simply not trained to manage AI or ask deep, scientific questions.

Eric Schmidt on the Robotics Race, Singularity Timeline, and Energy Shortage | 241 - Peter H. Diamandis and Eric Schmidt [Link]

Takeaways:

  1. Many in Silicon Valley believe that human-like AI agents and recursive self-improvement will lead to a Superintelligence moment within the next 2 to 3 years.
  2. The ultimate resource constraint for AI scaling in America is electricity, not capital or data. There is an estimated 92-gigawatt shortage of power in the U.S. through 2030 (the equivalent of roughly 60 nuclear power plants).
  3. Modern AI data centers have become massive airflow and liquid-cooled machines stretching half a mile long. It is projected that data centers will eventually consume 10% of the entire U.S. electricity supply.
  4. While the U.S. is heavily focused on central AGI/ASI models, China is dominating the physical AI and low-end robotic hardware landscape, driven by their existing EV supply chain, vertical integration, and aggressive work culture.
  5. The global landscape will likely accommodate around 10 massive-scale AI companies, primarily divided between the U.S. and China. China's strategy relies heavily on open-source edge computing.
  6. Schmidt reiterates that a minor biological, nuclear, or catastrophic event triggered by AI may be descriptively required to force global governments (like the U.S. and China) to pause their brutal competition and cooperatively establish hard guardrails. To win the race while preserving societal health, the U.S. needs to accelerate energy permitting, actively attract high-skilled immigrants and involve experts in history, ethics, and psychology to align Superintelligence with democratic values.

SpaceX IPO, Iran War Fallout, Quantum Bitcoin Hack, The Space Opportunity - All-In Podcast [Link]

Takeaways:

  1. Chamath predicts a 99.99% chance that Tesla and SpaceX will eventually merge into a single entity (ticker "E"). This would solve governance issues, create external market validation, and capitalize on cross-disciplinary compounding (e.g., Tesla’s autonomous robotics deployed on the Moon).
  2. Returning to the moon via the Artemis II mission highlights a new frontier. Due to low gravity and a lack of atmosphere, manufacturing materials on the Moon and using "mass drivers" (electric rails) to shoot them back to Earth will theoretically be cheaper than traditional terrestrial shipping.
  3. If AGI becomes a reality, the durability and valuation of software companies will erode rapidly. Investors are beginning to rotate into defensive, cash-flowing "HALO" (High Asset, Low Obsolescence) businesses.
  4. Recent computing algorithms and theory improvements have shifted the timeline for industrial-scale quantum computing from decades away to the next 5 to 7 years.

Marc Andreessen on The Future of VC: Will a16z Go Public & Why Introspection is Dangerous? - 20VC with Harry Stebbings and a16z [Link]

Takeaways:

  1. In venture capital, learning too much from failures can be dangerous. It is easy to completely write off a category because a previous investment failed, causing you to pass on the pattern-matched winner later on (e.g., passing on internet search in the 90s or AI before 2017). Adopting the mindset that everything is your fault drains away resentment and keeps your focus productively anchored on intrinsic self-improvement. However, as a management blanket rule for already stressed-out founders, a16z avoids pushing it.
  2. Andreessen quotes pioneer Arthur Rock, agreeing that firms would likely perform better by shredding business plans and focusing entirely on the team/resumé. Great teams lap mediocre teams executing perfect business plans.
  3. The Andreessen Core Formula:
    • IQ: Extreme intelligence is table stakes.
    • Courage: The absolute determination to run straight through problems like a cartoon character leaving a hole in a brick wall.
    • Drive: An intrinsic, primal ambition to build something of your own, distinct from external markers of success like net worth.
  4. VCs shouldn't chase "diamonds in the rough" out of investor ego. True diamonds have high visibility; if a deal has been ignored by the mainstream, it usually points to a structural defect or a hyper-disagreeable founder who has alienated everyone else. Echoing Don Valentine, more companies die from indigestion than starvation. Overfunding warps discipline, while high valuations create artificial hurdle rates that scare away future investors from down rounds.
  5. While the remote-work era from 2020 to 2023 promised to decentralize the tech industry, AI has whiplash-reversed this trendd. The tech ecosystem is now more geographically centralized in Silicon Valley than ever before in its history.
  6. AI is the ultimate small-D democratic technology. The highest-performing models in the world are consumerized and instantly accessible to anyone with a smartphone.
  7. The narrative that AI causes net labor displacement is fundamentally wrong and relies on zero-sum Marxist economics. Historically and moving forward, core technology boosts the marginal productivity of the individual worker, freeing them from grunt work to perform higher-value tasks. Current tech layoffs are driven by interest rate shocks and massive, undisciplined over-hiring during COVID—not AI. Many corporate entities simply use AI as a convenient "silver bullet" excuse to downsize.

How to Reorg After AI Changes Everything | Block's Owen Jennings on the a16z Show [Link]

Takeaways:

  • Block eliminated the standard mid-level management layer required to coordinate traditional, heavy feature teams (which typically consisted of 14+ people, including dedicated project managers, product managers, and numerous localized engineering leads). They replaced these with highly fluid squads of 1 to 6 people. Because these squads leverage automated AI tooling, they require significantly fewer coordinators and middle managers to align their work.
  • Product and engineering management layers were cut by 50% to 60%, widening leadership spans and drastically accelerating the internal velocity of data and decision-making.
  • Workflows transitioned from linear code production and manual peer reviews to asynchronous oversight. Managers and engineers run dozens of parallel background agents, context-switching between them to review, nudge, and commit agent-generated pull requests.

Why Juries Are Turning Against Meta and YouTube - Hard Fork [Link]

Takeaways:

  1. Separate juries in Los Angeles and New Mexico recently found social media giants liable for harming young users
  2. A recent leak exposed the agentic framework surrounding Anthropic's coding harness (Claude Code), enabling developers to rapidly clone and "Frankenstein" the tool over open-source models.
  3. Citing research from Anthropic's Nicholas Carlini, AI tools have become so proficient at finding software vulnerabilities that nearly all legacy codebases (like the Linux kernel) will eventually need to be hardened to resist AI-driven cyber attacks.

How Bots, Deepfakes and AI Agents Are Forcing a New Internet Identity Layer | Alex Blania on a16z [Link]

Takeaways:

The exponential rise of highly persuasive AI agents and deepfakes is making it nearly impossible to digitally distinguish real humans from bots, rendering standard one-to-one biometric verification obsolete.

To eliminate dystopian privacy risks, World uses Multi-Party Computation to fragment iris codes across decentralized computers and relies on zero-knowledge proofs so users can verify their uniqueness anonymously.

Without a cryptographically strong infrastructure to verify humanity, AI-driven automation will scale massive fraud across government social programs and structurally collapse the ad-based creator economy through bot-driven content farms.

World is shifting 90% of its resources to an aggressive U.S. expansion, aiming to deploy 50,000 biometric "Orbs" in major retail spaces and pilot on-demand motorbike couriers to ensure users are within 15 minutes of a device.

OpenClaw, Claude Code, and the Future of Software | Peter Yang on The a16z Show [Link]

Takeaways:

  1. Peter explains that 70–80% of the value of OpenClaw for him is its personal, conversational aspect. Using voice interactions via messaging apps like Telegram feels much more direct and personal than traditional chat interfaces.
  2. Functional apps used primarily to complete specific tasks (like Mercury banking or Calendly) are seeing reduced manual usage. Users prefer instructing an agent to run analytics, manage documents, or interact with APIs on their behalf.
  3. AI code generation tools currently feel like slot machines or casinos due to variable wait times and unexpected results. However, once customized with personal hooks and skills, they become deeply integrated workflows.
  4. Functional apps used primarily to complete specific tasks (like Mercury banking or Calendly) are seeing reduced manual usag. Users prefer instructing an agent to run analytics, manage documents, or interact with APIs on their behalf.
  5. AI is highly capable of tackling the first 80% of a creative or technical project (such as writing code or blog drafting), while human input is strictly needed to manually tweak and polish the final 20%.
  6. Instead of scaling corporate headcount, future companies will prioritize staying as lean as possible. A traditional 10-person product team may shrink down to two or three humans paired with a network of specialized AI agents.
  7. Entrusting objective negotiations and tedious operational functions (like long OKR alignment meetings) to AI agents eliminates high-emotion corporate back-and-forth.
  8. While large corporations may continue trimming staff, the cost to build software or launch a business is dropping close to zero. This enables a rise in independent solopreneurs, bootstrapped micro-businesses, and creators who can build their own products without traditional engineering backgrounds.

The State of Modern War: Palantir & Anduril Execs on Drones, AI, and the End of Traditional Warfare - All-In Podcast [Link]

Takeaways:

  1. America has dangerously hollowed out its commercial manufacturing ecosystem, leaving a hyper-consolidated defense industrial base that lacks the raw volume and agility required to rapidly regenerate weapon stockpiles during a major conflict.
  2. Modern tech startups are bypassing the government's sclerotic, "cost-plus" procurement bureaucracy by using private venture capital to build advanced, merit-based products before pitching them to the state.
  3. Integrating AI into the command center introduces unprecedented targeting precision that dramatically minimizes civilian casualties compared to legacy "dumb bombs," though human operators must always remain accountable for the system's actions.
  4. Failing to aggressively re-industrialize the West risks yielding a "Chinese Century" where the CCP explicitly dictates the global terms of international trade and engagement.

OpenAI's TBPN Mistake, SpaceX’s $2 Trillion IPO?, Iran Disables Amazon Infrastructure - Alex Kantrowitz [Link]

Takeaways:

  1. OpenAI acquired the daily tech talk show TBPN (The Better Podcasting Network)
  2. SpaceX has filed for an IPO, aiming to publically raise between $40 billion and $80 billion—potentially making it three times larger than any historical IPO
  3. In internal memos, Amazon Web Services (AWS) confirmed that Iranian-backed strikes in Bahrain and Dubai physically took down critical "availability zones,". Due to the severe volatility, the cost of specialized war and terrorism insurance for data center physical assets in the Gulf region has skyrocketed by 1,900% in just a few weeks.

Why Balaji Srinivasan Thinks the SaaS Apocalypse Is Overhyped | The a16z Show [Link]

Takeaways:

  1. The AI economy will shift toward distillation, decentralization, and localized "trusted tribes" resembling the low-trust Chinese internet model rather than central tech monopolies.

  2. AI text and slides are frequently lazy, generic, or deceptive ("Lauram AIPSM"). AI drastically reduces the cost of generation but scales up the hidden cost of validation.

  3. Humans will remain the essential "sensors" of taste and agency, while AI serves strictly as the "actuator" under a leash. AI turns employees into CEOs rather than replacing them.

  4. The wholesale destruction of SaaS is overhyped. Intelligent incumbents will survive by leveraging their existing distribution loops.

    While vulnerable incumbents that fail to execute may be cloned or disrupted, dominant developer platforms like Notion or Figma can use AI to build features and ship value to their established networks significantly faster. A raw technical clone of an app like Facebook yields zero value without its underlying user graph and distribution.

  5. Bitcoin has evolved strictly into transparent institutional collateral rather than digital cash for individuals. Zcash (and wallets like Zodal) will fulfill the actual multi-decade vision of private, scalable digital cash.

Meta Tokenmaxxing, Intel Joins Terafab, Frontier AI vs. China | Diet TBPN [Link]

We Have to Talk About Anthropic's Mythos - Hard Fork [Link]

Takeaways:

  1. Anthropic developed an incredibly powerful new model but has held it back from the public due to extreme cybersecurity risks. Instead of a commercial launch, Anthropic initiated Project Glasswing, giving exclusive, restricted access to the "blue teams" (defensive cybersecurity units) of major tech infrastructure and platform companies—such as Cisco, Broadcom, Microsoft, Apple, and Amazon.
  2. The hosts note that the sheer volume of undiscovered zero-day bugs this model can chain together autonomously means almost every major piece of software globally may need to be systematically patched or rewritten, creating a massive human bottleneck for software maintainers.
  3. This marks the first time since GPT-2 in 2019 that a massive gap has opened between the frontier capabilities hidden inside top AI labs and what the general public can actually use
  4. While the systemic threat is institutional, the hosts emphasize that everyday users should protect themselves against localized fallout by locking down basic digital hygiene: using password managers for unique, randomly generated passwords and enforcing multi-factor authentication (MFA) via authenticator apps.

The Mistake That Could Break America - David Friedberg - Chris Williamson [Link]

Takeaways:

  1. Friedberg highlights that a massive portion of leaders in the technology sector are actively leaving or preparing to leave California.
  2. Friedberg argues that politicians get elected by promising benefits they ultimately cannot fund. California faces a massive pension and healthcare liability crisis for public employees, estimated to be between \(\$600\) billion and \(\$1\) trillion in the hole.
  3. The state's massive tax revenues have been funneled into poorly managed projects. Friedberg points to massive budget waste, such as a \(\$30\) billion high-speed rail project with rampant executive turnover and a \(\$220\) million homeless program that only successfully transitioned six individuals out of poverty.
  4. The core philosophical threat discussed is California's proposed "Billionaire Tax Act". Friedberg emphasizes that taxing post-tax assets creates a dangerous precedent where private property rights are systematically degraded.
  5. While pitched at a 5% rate for billionaires, Friedberg warns that historical precedents—like the original US income tax, which started in 1913 as a mere 1% tax on high earners —prove that thresholds inevitably drop until a tax applies to the broader middle class. He fears the end state is one where 51% of the population votes to seize everything from the other 49%.
  6. Friedberg notes a tragic irony: humanity is right on the cusp of an era of unprecedented technological abundance—including near-free energy, radical life extension, and automated labor. Despite this potential, public sentiment remains deeply pessimistic. Friedberg cites a poll showing that AI is currently the most unfavorable concept in the United States, even ranking below highly polarizing political figures. He warns that if Western regulatory and tax regimes stifle this technological path, global competitors like China will glean the benefits instead.

What Really Happened Behind Closed Doors at OpenAI - Hard Fork [Link]

Takeaways:

The New Yorker's Profile on Sam Altman

  • Journalists Ronan Farrow and Andrew Marantz discussed their massive 16,000-word profile examining whether OpenAI CEO Sam Altman can be trusted.

  • The piece portrays Altman not through a single "smoking gun" incident, but rather a long, subtle accumulation of complaints from colleagues claiming he repeatedly tells different audiences different things.

  • The hosts reveal that the law firm investigation commissioned after Altman's brief 2023 firing was never actually put into writing; it was kept entirely verbal to limit liability, resulting only in a vague 800-word press release.

  • While the profile is quite damning, it also highlights that Altman is the target of intense, sometimes fabricated smear campaigns orchestrated by tech rivals like Elon Musk.

More OpenAI C-Suite Drama, Is Siri Seriously Broken?, Meta’s Elusive Next Hit - Alex Kantrowitz [Link]

What You Can Learn From Our \(\$20\)B Groq Deal - Chamath Palihapitiya [Link]

Takeaways:

  1. Traditional corporate org charts are often "bullsh*t" built for internal politicians; real value is driven by building the company around elite technical talent
  2. Startups require teams to find emotional reward in technical milestones and quiet progress during years of darkness before hitting exponential growth
  3. True "A-level" performance comes from a culture of self-directed, high-agency individuals pushing each other—"iron sharpens iron".

Box CEO on the AI Adoption Gap | The a16z Show [Link]

Takeaways:

  1. Instead of relying on predefined IT connections, agents will seamlessly orchestrate data across complex, multi-system enterprise software (like SAP or Workday) at runtime
  2. You cannot simply "vibe code" your way through massive legacy software architectures. Systems of record aren't just neat data sets; they carry decades of complex business logic, meaning legacy systems are not disappearing anytime soon
  3. Wall Street models are missing the exponential growth curve by viewing the market through a zero-sum lens. Just like mainframes, bandwidth, and cloud compute before it, algorithmic or hardware shifts will drastically collapse the cost of intelligence, making current token-metering anxieties temporary.

Anthropic’s $30B Ramp, Mythos Doomsday, OpenClaw Ankled, Iran War Ceasefire, Israel's Influence - All-In Podcast [Link]

Josh Shapiro on Trump, Iran War Chaos, Israel's Failure, the Economy, and 2028 Race - All-In Podcast [Link]

Marc Andreessen introspects on Death of the Browser, Pi + OpenClaw, and Why "This Time Is Different" - Latent Space and a16z [Link]

SpaceX Goes Public, Claude’s Mythos Release, and the US Data Center Delay | EP #246 - Peter H. Diamandis [Link]

Ben Horowitz on AI Anxiety, Big Tech Transitions & The Future of Startups | a16z [Link]

Elon Musk vs. Sam Altman, AI Job Loss, and OpenAI’s $852B Valuation | EP #247 [Link]

Jensen Huang – Will Nvidia’s moat persist? - Dwarkesh Patel [Link]

Everything You Know Is About to Collapse - David Friedberg - Chris Williamson [Link]

The Pentagon's AI Plan + Behind the Anthropic Fight — With Under Secretary of War Emil Michael - Alex Kantrowitz [Link]

Building Agents at Home: Homeschooling, Parenting and More | The a16z Show [Link]

Allbirds’ AI Pivot, Snap Cuts 16% of Workforce, Amazon’s GlobalStar Deal | Diet TBPN [Link]

OpenAI's Identity Crisis, Datacenter Wars, Market Up on Iran News, Mamdani's First Tax, Swalwell Out - All-In Podcast [Link]

Why the A.I. Backlash Turned Violent in America - Hard Fork [Link]

I also feel like I had a reasonably prepared mind so that when that door opened, I was able to walk through it with my own two feet.

― Why Reading Most Books Is A Waste Of Time - Chamath Palihapitiya [Link]

Why Washington Suddenly Wants A.I. Regulation - Hard Fork [Link]

Takeaways:

  1. Anthropic's unreleased preview model, Claude Mythos, is the primary reason Washington is suddenly panicking. The model has proven to be incredibly powerful at finding novel code vulnerabilities and daisy-chaining them together.
    1. Palo Alto Networks CEO Nikesh Arora noted that during a concerted auditing effort using these advanced models, his company discovered seven times the volume of exploits they would typically find in a normal period.
    2. While proprietary SaaS can be patched quickly, older legacy code and open-source software remain high-risk vectors. Cyber security firms are using A.I. to build perimeter firewall "scaffolding" to block automated attacks before software can be manually updated.
    3. The Pentagon is legally battling Anthropic in court—designating it a "supply chain risk" because the lab refused a contract allowing "any lawful use"—while simultaneously installing and implementing Claude Mythos to scan their own systems for security vulnerabilities.
    4. The sudden consensus that "something must be done" has triggered a fierce bureaucratic scramble within Washington over who gets to hold the keys to A.I.
  2. The hosts highlight that while Washington is scrambling toward safety regulations, its current execution is deeply fractured and contradictory:
    1. At the exact same time national security officials are warning that these models are too dangerous to fall into adversary hands, Trump invited Nvidia CEO Jensen Huang onto Air Force One to join a high-profile trade delegation to China. The trip's goal includes negotiating chip trade policies—essentially trying to sell China the raw hardware compute power needed to build the very caliber of models (like Mythos) that Washington is trying to restrict.
    2. This institutional confusion is perfectly mirrored in the Pentagon, which has been using Claude Mythos to scan and secure its defensive infrastructure while simultaneously fighting Anthropic in court to label the company a "supply chain risk" because the lab refuses to allow unrestricted military use of its commercial tech.
  3. Venmo is moving away from its famous public-by-default feed, shifting onboarding for new users to "friends only".
  4. Frugal Amazon employees are allegedly abusing the company's internal A.I. tool (Meshclaw) to generate unnecessary background activity simply to boost their token consumption and look more productive to leadership.
  5. A University of Central Florida commencement speaker was loudly booed by humanities and arts graduates after proclaiming A.I. as the "next industrial revolution".

How the 1% Will Own Compute (and What It Means for You) - This Week in AI and This Week in Startups [Link]

Anthropic CEO on Safety, Job Displacement and Anthropic's $350B Valuation | WSJ [Link]

Takeaways:

  1. Amodei emphasizes that while public sentiment fluctuates wildly between AI hype and a burst bubble every 3–6 months, the actual capability of AI models follows a smooth, exponential line upward—acting essentially as a "Moore's Law for intelligence".
  2. The signature of this AI transition will likely be a world experiencing very high GDP growth (5% to 10%) occurring simultaneously with high unemployment (potentially 10%) and increased inequality. Anthropic launched the Anthropic Economic Index to track in real-time exactly how Claude is automating versus augmenting tasks across industries so that policymakers aren't acting blindly.
  3. Anthropic intentionally prioritizes business clients over consumer spaces to avoid the toxic incentives of traditional tech, like maximizing user engagement, serving ads, or generating algorithmic "slop". Developers and non-technical staff alike have shifted heavily toward agentic workflows rather than simple text chat, allowing tools to build entirely new apps end-to-end or organize projects within minutes. Anthropic is explicitly choosing not to focus on photo or video generation, viewing much of consumer short-form video AI content as addictive "slop" with minimal value to enterprise operations.
  4. One of Amodei's major worries is a dystopian outcome where a highly contained "zeroth world country" of ~10 million tech elites (mostly centered in Silicon Valley) accelerates to 50% GDP growth while decoupling entirely from the rest of the slower-moving global economy.
  5. When asked about the missing piece to make frontier AI safe, Amodei pointed to mechanistic interpretability—the science of physically looking inside neural networks (similar to an MRI for the human brain) to confirm if a model is being deceptive or lying, rather than just relying on text testing.

Watts, Wafers, and the Future of AI Infra | Gavin Baker - Invest Like The Best [Link]

Google I/O 2026 keynote in 35 minutes - The Verge [Link]

Joe Rogan Experience #2501 - Marc Andreessen - PowerfulJRE [Link]

Is It Time to Break the Two-Party System? | The Ezra Klein Show [Link]

Why MKBHD is glad he never went viral | EP 53 - Hard Fork [Link]

Claude Code Head Boris Cherny: Insane Growth, Tokenmaxxing, AI Agents' Next Frontier - Alex Kantrowitz [Link]

Sundar Pichai on Whether Google Is Falling Behind in A.I. - Hard Fork [Link]

SpaceX’s $2T Case, Nvidia’s Shock Selloff, America Turns on AI, Trump Pulls AI Order, Bond Crisis? - All-In Podcast [Link]