The deceptive nature of growth and the AI durability framework Growth is often a mask that hides fundamental failures within a business. In the current venture climate, hyper-growth can blind investors to underlying structural issues, leading to catastrophic miscalculations. Miles Clements, Partner at Accel, argues that the only way to avoid these traps is a rigorous evaluation of "time to value" versus the "durability of value." This framework separates transient novelties from generational platforms. Take the contrasting examples of legal or accounting AI versus "vibe coding" apps. AI tools for professional services often suffer from slow adoption cycles; selling to lawyers and accountants is a grind. However, once integrated, the durability of that value is transformational. Conversely, early vibe coding apps offer instant gratification—the weekend warrior can build a functional app overnight—but they often lack long-term retention because the value doesn't compound. The coding vertical has become the primary battleground in AI because it satisfies both ends of the spectrum: developers see productivity gains within hours, and the utility scales exponentially as teams integrate the tools into their core workflows. Cursor remains the engineering platform of record despite Twitter noise The "Cursor is dead" narrative, fueled by competitive moves from Claude Code and public criticism from high-profile figures like Chamath Palihapitiya, ignores the sheer scale of the company's market expansion. While Anthropic's Claude Code has captured the zeitgeist, the market for AI-assisted engineering is not a zero-sum game. Clements highlights that these tools are bringing entire cohorts of new developers online and radically increasing consumption per user. Critics often point to Cursor's reliance on external models or its perceived high cost, but the metrics tell a different story. The company recently surpassed $2 billion in Annual Recurring Revenue (ARR), a figure that signals unprecedented product-market fit. Furthermore, Cursor is successfully transitioning from a feature-based IDE to an agent-driven platform. Data indicates that 90% of Cursor users are daily active users of the agent product, which grew 15x over the last year. By the time most observers realized Cursor was a player, it had already begun building specialized coding models to ensure enterprise-grade performance that generalist models cannot match. Learning from the sting of missed generational founders No venture firm has a perfect track record, and Miles Clements is candid about the misses that haunt Accel. Two specific examples—ServiceTitan and Rippling—illustrate the dangers of rigid adherence to traditional valuation rules. In the case of ServiceTitan, Accel hesitated on a round in the $250 million range because of strict internal caps on vertical SaaS multiples. The company eventually reached a $9 billion valuation. The lesson was clear: if you understand the depth of market disruption, you must be willing to break the rules on price. Rippling represents a different kind of miss—one rooted in an underestimation of what Clements calls the "marginal ease of ARR accumulation." Parker Conrad demonstrated a unique ability to capture margin in areas others ignored, such as laptop provisioning and IT leasing, turning them into high-growth revenue lines. While Accel stuck to its "knitting" regarding ownership thresholds and valuation caps, the reality is that generational founders like Parker Conrad deserve a different playbook. In a hyper-competitive market, the "science" of valuation must occasionally yield to the "art" of recognizing when a founder is operating on an entirely different plane. The shifting mechanics of venture ownership and exits The traditional venture model—investing at the Series A to secure 20% ownership and riding it to an IPO—has been inverted. In today's market, firms must often "ladder" their way into a significant position. This involves sponsoring tenders, participating in growth rounds, and buying into the IPO. This multi-stage approach is essential when dealing with companies like Stripe or DataBricks that stay private longer. This delay in going public is a rational response to the current state of the public markets, which Clements describes as increasingly irrational and "casino-ized." When a single security report or a macro shift can wipe out billions in market cap regardless of company fundamentals, founders naturally prefer the relative stability of private secondary markets. For the top tier of private companies, the public market's traditional benefits—employee liquidity and M&A currency—are now readily available through private tenders. Consequently, the threshold for a successful IPO has moved; companies now need a clear line of sight to a $5 billion market cap and beyond to escape the "murky" middle ground where smaller public stocks are often ignored or oversold. Why founder conviction is the ultimate signal In a world where technology is transient, the founder remains the only durable asset. Miles Clements emphasizes that when a founder's conviction wanes, the investment thesis usually dies with it. This is why Accel continues to double down on companies where the original visionaries, like Mike Cannon-Brookes at Atlassian, remain at the helm. These leaders possess the resilience to navigate the "humbling" reality of market cycles and the emergence of disruptive technologies like AI. The venture business remains a human one. Success isn't just about picking the right sector; it’s about the professional standard maintained over decades. This involves a commitment to the "rituals" of the firm—the partner meetings and portfolio reviews—that keep investors grounded when the market gets euphoric. While the tools of the trade have shifted from issue tracking to autonomous agents, the core requirement remains the same: identify the outliers who are building for a ten-year horizon, not just the next fundraising cycle. The future belongs to those who can distinguish between a momentum-chasing fad and a platform with the structural integrity to compound for decades. Implications for the next decade of innovation The AI revolution is not over; it has barely entered its first true compounding phase. The belief that the "generational" bets in AI were all made years ago is a fallacy that Clements has since rejected. While early foundation model investments like OpenAI and Anthropic were massive, the application and agent layers are just beginning to take shape. We are moving toward a world where software replaces labor directly, rather than just augmenting it. This shift demands a change in investor mindset—looking for outcomes that aren't just $1 billion exits, but $100 billion platforms. The goal is no longer to find a company that hits a double; it is to find the ones capable of rewiring the entire global economy.
CrowdStrike
Companies
- Mar 9, 2026
- Feb 28, 2026
- Jul 30, 2025
- Jun 28, 2025
- Jun 14, 2025
The artificial heartbeat of digital creativity YouTube is entering a transformative, if controversial, era. At the recent Made on YouTube event, the platform unveiled a suite of AI-driven features that aim to lower the barrier to entry for content creation. From Dream Screen, which utilizes DeepMind's Veo model to generate six-second video backgrounds, to automated comment replies, the message from Google is clear: AI is no longer a peripheral experiment; it is the new engine of the creator economy. However, the rollout raises fundamental questions about the value of human presence on the platform. While Linus Sebastian attended the Creator Summit and heard the official pitch—that these are "tools for enhancement, not replacement"—the reality on the ground feels more complex. As Luke Lafreniere pointed out, many of these capabilities already exist in the wild. Bringing them natively to the YouTube Studio dashboard doesn't just democratize high-end production; it potentially automates the very personality that viewers tune in to see. The tension between "efficiency" and "authenticity" is the defining conflict of this update. Generative shortcuts and the decay of visual standards The introduction of Dream Screen allows creators to generate still and moving backgrounds for YouTube Shorts using text-to-video prompts. While this sounds like a boon for small creators without budgets for expensive sets or location shoots, the practical application often results in what critics call "motivational junk" or AI-generated spam. There is a legitimate concern regarding the "photocopy of a photocopy" effect. As more creators use AI to generate backgrounds, and those videos are subsequently fed back into the training models of future AIs, the visual language of the platform risks entering a state of creative decay. When the Inspiration Tab begins suggesting ideas based on what AI previously suggested to other creators, the "human in the loop" becomes more of a rubber stamp than a visionary. For a platform built on the unique perspective of individuals, a move toward homogenized, algorithmic aesthetics feels like a step backward, even if the technology itself is impressive. The hollow intimacy of automated engagement Perhaps the most unsettling feature is the new AI-powered comment reply suggestion tool. This system analyzes a creator's past interactions to suggest replies that mimic their specific "voice" and tone. While YouTube argues this helps busy creators manage large communities, it fundamentally alters the social contract between the audience and the uploader. Linus Sebastian argued that a reply from a creator is valued precisely because it represents a moment of real human attention. Masking an AI behind a creator's handle—and potentially even having it sign off with personal initials like "-LS"—is arguably manipulative. It fosters a fake intimacy, encouraging "parasocial" relationships that are rooted in a lie. If a viewer believes they are interacting with their favorite personality but are actually chatting with a fine-tuned Large Language Model, the trust that forms the foundation of community-driven platforms begins to erode. Without clear "AI-generated" disclosures on every automated reply, YouTube is inviting a future of deceptive engagement. Fragmentation and the rise of the niche community While AI dominates the headlines, a broader shift in viewer behavior is occurring: the decline of the "Mega Channel." For years, the platform was dominated by generalists who covered a wide variety of topics. Now, YouTube's own data suggests a pivot toward hyper-specific, smaller communities. These "tiny fandoms"—ranging from enthusiasts of a specific bridge in North Carolina to creators who purely review Taylor Swift albums—are where the highest engagement now lives. This shift explains why YouTube is testing a "Hype" feature, effectively a "pay for algorithm" model where fans can spend money to boost smaller creators (under 500,000 subscribers) onto local leaderboards. It is a curious contradiction: while the platform builds AI tools to help everyone look like a big-budget production, the audience is fleeing toward smaller, more intimate corners of the site. The irony is that the automated tools might actually accelerate this flight, as users seek out the few remaining spaces where they can be certain a real human is on the other side of the screen. Anti-cheat escalation and the Linux casualty Beyond the creative tools, the tech side of the industry remains in a state of flux, particularly for Linux users. Rockstar Games recently integrated BattlEye anti-cheat into Grand Theft Auto V, effectively killing compatibility for Steam Deck and Linux gamers. This move is particularly baffling given that BattlEye is officially compatible with Valve's Proton layer; the developer simply chose not to enable it. However, a potential silver lining emerged from the wreckage of the CrowdStrike disaster. Microsoft appears to be moving toward pushing security and anti-cheat software out of the Windows Kernel. If anti-cheat is forced to operate in user-mode rather than at the highest privilege level of the operating system, the technical hurdles for running Windows-based games on Linux could vanish. This would be a massive win for open-source gaming, turning a restrictive security mandate into a gateway for platform independence. Accountability in the hardware ecosystem The hardware market currently faces its own challenges, notably regarding ASUS and its ongoing struggle with customer service reliability. A recent report from a high-end visual effects director detailed how an authorized ASUS service center shattered a $12,000 ProArt laptop and then refused a replacement because the device was out of warranty—ignoring the fact that the damage was caused by the technician, not a manufacturing defect. This highlights a persistent issue within the consumer electronics world: the "broken process." Even when a company's leadership expresses a desire to improve, the gap between corporate policy and the local service center remains vast. For power users and professionals, these service failures are not just inconveniences; they are threats to their livelihoods. It serves as a reminder that the most advanced hardware in the world is only as good as the support infrastructure behind it. Conclusion The current state of consumer technology is one of high-velocity experimentation. YouTube's pivot to AI seeks to solve the "blank page" problem for creators, yet it risks creating a "blank soul" problem for the platform. As we navigate the end of the year, the success of these initiatives will be measured not by how much content is generated, but by how much of that content actually matters to a human audience. Whether it's Nintendo suing Palworld over gameplay patents or AMD seeing sluggish sales for its Ryzen 9000 series as gamers wait for X3D variants, the market remains focused on utility and authenticity. Technology can mimic the style of a creator or the mechanics of a game, but it cannot yet replace the intentionality of a human decision.
Sep 20, 2024The Seniority Illusion and the Groovy Catalyst When The Primeagen walked into Netflix at 27, he felt like a fraud. Surrounded by seasoned engineers at least five years his senior, he accepted the assignment no one else wanted: building a backend with Groovy. This wasn't a choice born of expertise, but of survival and a willingness to tackle the "worst job" in the room. This highlights a critical principle in software development: growth often hides behind the tasks others avoid. By embracing the messy, unglamorous work of a legacy-adjacent language, he built the foundation for a career that eventually spanned a decade at one of the world's top tech firms. Success didn't come from knowing the stack perfectly on day one; it came from the decision to figure it out regardless. The Calculus of Determination Transitioning from a "wayward youth" to an elite engineer requires a specific kind of mental friction. Reflecting on a time he failed pre-calculus twice before barely scraping by with a 'C', he found himself at a crossroads. The shift happened during a grueling five-week summer session where he committed ten to twelve hours a day to a single subject. This wasn't about innate genius or being "math-brained." It was about raw volume and persistence. This experience serves as a blueprint for any developer struggling with complex concepts like recursion or system design. If you apply consistent, concentrated pressure to a problem, your background ceases to be a limiting factor. Actionable Practices for Personal Growth To achieve true excellence, you must commit to two core behaviors. First, **bet on your ability to solve the unknown**. Whether it is the rise of LLMs or a shifting job market, the only stable currency is your capacity to learn. Second, **protect your non-negotiables**. Early in his career, he worked 100-hour weeks, nearly sacrificing his marriage for a startup. Wisdom is realizing that an extra ten hours at the office rarely changes the trajectory of a product, but it can destroy the things that actually matter. Concluding Empowerment: Solve the Right Problems Intelligence might help you solve a hard bug, but wisdom dictates whether that bug was worth your life's energy. Stop chasing metrics or trying to please an imaginary audience. Whether you are coding or creating content, the most sustainable path is doing what you love in a way that feels authentic. Take the chance. Bet on your ability to adapt, but never cut corners on your values. The dividends of that bet will pay out for decades.
Sep 10, 2024Audit Your Supply Chain Dependency management is no longer just about running an update command. Modern software relies on a sprawling web of open-source packages from managers like npm and pip. Blind updates invite disaster, as seen when the colors package maintainer intentionally introduced an infinite loop. You must implement a **Software Bill of Materials (SBOM)** to track every library and catch vulnerabilities before they reach your users. Treat third-party code as a potential back door, not just a convenience. Validate with Canary Releases Never push code to your entire user base at once. Even robust testing environments can fail to mimic real-world complexity, a lesson Microsoft learned during its 2018 Windows 10 update. Instead, use a **Canary Release** to deploy changes to a tiny subset of users. This strategy isolates potential failures, such as data loss or crashes, to a controlled group, providing the telemetry needed to halt a rollout before it becomes a global headline. Limit the Blast Radius High-level authorization is a liability. The recent CrowdStrike outage highlights the danger of granting tools Kernel-level access. If a product doesn't strictly require deep system permissions, revoke them. Apple has already moved toward restricting legacy kernel extensions in macOS. By strictly enforcing the principle of least privilege, you ensure that a single bug cannot trigger a system-wide Blue Screen of Death. Shift to Memory Safety Legacy languages like C++ are prone to manual memory errors, including the null pointer exception that crippled systems worldwide. Transitioning to memory-safe languages like Rust eliminates entire classes of bugs at compile-time. While Python remains excellent for high-level logic, system-critical components demand the strict safety guarantees that modern low-level languages provide.
Jul 26, 2024The Kernel-Level Collapse When CrowdStrike released a faulty update to its Falcon platform, it didn't just crash a few apps; it triggered a global cascade of failure. To understand why, we have to look at the Windows kernel. Security software requires the highest level of system permission to intercept threats. This lack of guardrails means that a single bug in a driver can bypass standard OS protections, leading to the infinite boot loops that grounded flights and halted hospitals. The Speed vs. Safety Dichotomy Software engineering usually thrives on rigid processes: better testing, canary releases, and staggered deployments. However, the Falcon incident exposes a fundamental dichotomy. In the face of an imminent zero-day threat, security teams must move at lightning speed. Waiting weeks for a staged rollout could leave millions of machines vulnerable to a massive attack. We are trapped between the need for slow, methodical quality assurance and the necessity of rapid response. This isn't just a technical hurdle; it is a structural contradiction in how we maintain critical infrastructure. A Future of Increasing Complexity As we integrate more AI and robotics into the physical world, the stakes for software quality escalate. We are currently moving toward a reality where code reviews and cloud optimizations are managed by algorithms that we understand less by the day. When software takes physical form—such as self-driving cars or industrial robotics—a simple camera tracking bug becomes a life-threatening event. Rethinking Development Culture To prevent the next disaster, businesses must stop treating maintenance as a mere cost center. We need to empower engineers to advocate for robust devops practices and observability, even when under pressure to ship features. Relying on AI to write and deploy our systems without deep human oversight is a recipe for catastrophic failure. We must prioritize system stability over short-term velocity before our digital dependencies become our greatest liability.
Jul 21, 2024The Friday the digital world stopped turning On July 19, 2024, the tech world witnessed what can only be described as a digital cardiac arrest. A single faulty software update from CrowdStrike, a titan in the cybersecurity industry, triggered an endless cycle of Blue Screens of Death (BSOD) on millions of Windows machines globally. This wasn't a sophisticated state-sponsored cyberattack or a catastrophic hardware failure in a data center; it was a self-inflicted wound. The Falcon sensor update, designed to protect systems, instead rendered them completely unbootable, halting operations at major airlines, banks, hospitals, and even emergency 911 dispatch centers across multiple US states. As a reviewer, I often talk about the importance of software stability, but we rarely see the consequences of failure on this scale. The incident highlighted a terrifying reality: our global IT infrastructure is hyper-centralized. When a handful of companies control the security and operating systems for the entire planet's critical infrastructure, a single line of bad code becomes a global liability. While Microsoft escaped direct blame for the fault, the reliance on their kernel-level access for security tools like Falcon meant that when CrowdStrike failed, it took the entire Windows ecosystem down with it. Linux systems and sanitized regions like Russia—where sanctions prevent the use of such software—remained ironically unaffected, standing as silent observers to the Western world's digital meltdown. The manual labor of a digital recovery The most agonizing aspect of the CrowdStrike outage isn't just the downtime; it's the recovery process. Unlike a typical cloud glitch that can be patched server-side, this update crippled the machine's ability to even reach the internet. To fix the affected systems, IT managers have been forced into a grueling manual workflow: physically accessing each machine, booting into Safe Mode, and manually deleting specific system files. This is a nightmare scenario for any enterprise with a distributed workforce or offshore IT support. Compounding the misery is BitLocker. Many affected corporate laptops use Microsoft's encryption tool, which requires a recovery key to access the drive in Safe Mode. The cruel irony? Many companies store those recovery keys on internal servers that are currently trapped in the same boot loop. It's a recursive failure that effectively bricks hardware until keys can be recovered from backup systems that may also be offline. We are looking at billions of dollars in lost productivity and a massive reputational stain on CrowdStrike, whose stock price plummeted as the scale of the "oopsy-doodle" became clear. If this is what a mistake looks like, it serves as a chilling preview of what true digital warfare would look like if targeted at our core infrastructure. Intel's silicon rot and the server exodus While CrowdStrike dominates the news cycle, a slower, more insidious crisis is unfolding for Intel. Reports from MMO publisher Alderon Games and technical deep-dives from Level1Techs suggest that Intel's 13th and 14th Generation Core i9 and i7 processors are suffering from a 100% failure rate in certain server environments. This isn't a software bug; it's hardware deterioration. These high-end chips appear to be literally rotting over time, losing stability until they can no longer maintain basic operation. Historically, the CPU has been the most reliable component in a PC build. You expect the GPU to sag, the RAM to throw errors, or the SSD to wear out, but the silicon at the heart of the system is usually rock-solid. This failure breaks that trust. Alderon Games has publicly announced they are migrating their entire server fleet to AMD Ryzen chips because they simply cannot rely on Intel anymore. This hits Intel where it hurts most: reputation among enterprise clients. For decades, the mantra was "nobody ever got fired for buying Intel." Today, that no longer holds true. If the 14900K and 13900K require a massive recall, the logistical and financial burden could be monumental, potentially requiring Intel to reallocate R&D teams from future products just to fix a legacy disaster. Netflix and the death of the ad-free middle class In the world of consumer software, Netflix is proving that they don't care about your boycott threats. Despite the initial outrage over password-sharing crackdowns, the streaming giant has reached a record 277 million subscribers. Their latest move is the final nail in the coffin for the "middle class" subscriber: the discontinuation of the $12 Basic ad-free plan in the US and France. Users are now being corralled into two extremes: the $7 plan with ads or the significantly more expensive $15.50 Standard plan. This is a classic "bait and switch" business model. You attract users with a sustainable, ad-free experience at a reasonable price, wait for the competition to decay or consolidate, and then unilaterally change the deal. There is something fundamentally "un-cool" about this. While technically legal under monthly contracts, it feels predatory. If we want to change this, we need a regulatory shift toward "grandfathering" as a legal requirement. If you sign up for a service at $12, the company shouldn't be allowed to keep your credit card on file while changing the terms of what you're buying. They should have to cancel the subscription and force you to re-sign. But Netflix knows the power of friction; they know most people will just grumble and pay the extra $3.50 because they don't want to lose their watch list. The "Enshittification" of gaming and the 82 percent We often complain about the state of modern gaming—the microtransactions, the battle passes, and the predatory "freemium" models. However, new data reveals that 82% of American adults who play games have made an in-game purchase in a free-to-play game in the last year. We are our own worst enemy. The reason companies like Activision and Microsoft continue to push these models is that they work. Even the most vocal critics are often contributing to the very systems they despise. There is a massive disparity in how these transactions are viewed. Some games, like League of Legends or Halo Infinite, largely limit purchases to cosmetics. You're a "sponge" for the whales, playing a free game because someone else decided to spend $200 on a shiny skin. That's a deal many are willing to take. But then there's "Pay to Win" territory, like Diablo Immortal, where you aren't a player; you're content for the paying whales to slaughter. When 82% of the population is actively feeding this beast, we lose the right to act surprised when the next big franchise is gutted for parts and sold back to us in $10 increments. Defending the critical voice against corporate bullying In a rare win for the little guy, the audio brand dCS recently apologized to reviewer GoldenSound after threatening him with a lawsuit over a three-year-old review. This is a crucial moment for tech media. dCS alleged inaccuracies but failed to provide specifics for months, eventually resorting to legal intimidation. It was only after a united front from the tech community and GoldenSound's employer, Headphones.com, that the company backed down, blaming a rogue employee for "unacceptable" behavior. This highlights the importance of a "unionized" mindset among independent reviewers. Whether it's Hardware Unboxed facing Nvidia's wrath in the past or this recent dCS debacle, we must stand together. Manufacturers need to understand that the reviewer's job is to protect the consumer, not to be a free marketing arm for the brand. An apology is a start, but the industry needs to move toward a culture of transparency and respect for critique, rather than litigious defense of the bottom line. Conclusion: Navigating a fragile digital landscape Between the CrowdStrike meltdown and Intel's silicon failures, this week has been a masterclass in the fragility of modern technology. We've built a world on top of complex layers of software and hardware that we assume will just work, only to find that a single mistake or a manufacturing flaw can bring global industry to its knees. Moving forward, the tech industry must prioritize resilience over centralization and honesty over litigation. Whether it's how we build servers or how we interact with our subscribers, the "move fast and break things" era needs to mature into an era of stability and accountability. If we don't learn these lessons now, the next "Blue Screen Day" might not have a manual fix.
Jul 20, 2024