# A rational conversation on where AI is actually going | Benedict Evans

Podcast: Lenny's Podcast: Product | Career | Growth
Published: May 31, 2026
Reading time: 24 min
Canonical: https://podbrew.app/briefs/lenny-s-podcast-product-career-growth-a-rational-conversation-on-where-ai-is-act

Benedict Evans, an independent analyst and former Andreessen Horowitz partner, joins the podcast to share his unique insights on the future of artificial intelligence. Known for his deeply researched presentations, Evans helps founders, investors, and operators navigate the complexities of the tech landscape.

The discussion dives into why AI is experiencing its '1997 moment'—early, exciting, yet deeply uncertain. Evans unpacks where value will truly accrue in the AI stack, the surprising boom in professional services within AI companies, and why distribution is fast becoming the ultimate competitive moat. He also addresses the burgeoning anti-AI backlash and offers a rational perspective on the technology's societal impact.

Evans posits that AI is as transformative as the internet or mobile, but not more, offering a grounded view amidst widespread hype. His analysis moves beyond panic to examine the nuanced transformation of jobs—distinguishing between tasks and entire roles—and provides practical advice on how individuals and businesses can prepare for an AI-driven future where things will likely be okay with the right approach.

## Key takeaways

- AI's impact is comparable to previous platform shifts like the internet or mobile, not a more radical 'industrial revolution.'

- The AI landscape currently mirrors the internet in 1997, characterized by rapid development, unproven applications, and highly varied user adoption.

- Cutting-edge AI labs are heavily investing in professional services, contradicting the belief that AI would eliminate the need for consultants.

- Companies rely on external professional services to integrate AI, redesign workflows, and conduct strategic analysis because they lack the internal capacity and specialized teams for such large-scale projects.

- Automating individual tasks often makes them cheaper, leading to an increase in the overall complexity and scope of the job rather than job elimination.

- The real value in many professional roles lies in defining 'what' needs to be done and 'why,' rather not merely executing 'how' to do it, which AI can often handle.

- Historical examples like Excel, software development tools, and the growth in the number of accountants despite automation show that making a task easier doesn't equate to fewer jobs.

- New technologies consistently automate old jobs and create new, often better ones, leading to long-term societal enrichment despite initial frictional pain.

- AI's rapid adoption leverages existing digital infrastructure like PCs and broadband, rather than requiring the development of entirely new foundational technologies.

- Enterprise-wide AI integration and significant job displacement will occur over years, not months, due to lengthy software sales cycles and the complexity of replacing established systems.

- AI is vastly expanding the Total Addressable Market (TAM) by automating new economic sectors, enabling companies to reach unprecedented valuations.

- Value accrual for foundational AI models may not directly correlate with their widespread adoption, akin to how electricity providers do not capture a percentage of revenue from products that use their utility.

- Foundational AI model companies will likely experience margin compression and less long-term success due to ongoing competition and the absence of a winner-takes-all dynamic.

- The application layer in AI presents the larger and more enduring opportunity, as foundational models are expected to become commoditized infrastructure, similar to cloud services rather than a single dominant platform.

- For commoditized technologies like AI models, distribution is the most valuable asset and competitive moat.

- Incumbent tech giants can leverage their extensive user bases to integrate and push "adequate" AI features, capitalizing on user inertia and the power of default.

- AI-first startups face an urgent need to establish proprietary distribution channels and user stickiness to avoid being overshadowed by incumbents' broad reach.

- The actual economic impact of AI on jobs is largely unknown due to a critical lack of transparent usage data from AI model developers.

- Understanding the difference between automatable tasks and the underlying "jobs" people are hired for is crucial for identifying valuable human contributions in an AI-driven economy.

- To succeed, individuals must proactively dive into AI, understand its capabilities, and learn how to leverage it to become a more valuable asset in the evolving professional landscape.

## 02:00 - 05:52 AI is as transformative as the internet or mobile, currently resembling the internet in 1997 with wide-ranging adoption and unknown outcomes.

AI represents a significant technological shift comparable to the internet or mobile revolutions, rather than a new industrial revolution. This perspective contrasts with some who view AI as even more foundational and others who underestimate its impact.

The current state of AI is likened to the internet in 1997: a period of immense excitement where much technology is still nascent, and the definitive applications and business models are yet to be established. Many early adopters believe everyone is already on board, but real-world adoption varies widely, with many outside tech using AI infrequently.

This early phase also sees intense competition among foundational model developers like OpenAI and Anthropic, reminiscent of early search engine battles like Excite. The long-term winners, value capture, and overall structure of the AI industry remain highly uncertain.

> I think that AI is as big a deal as the internet or mobile, and only as big a deal as the internet or mobile, 'cause clearly there's a bunch of people in tech who think, no, this is more like the industrial revolution or something.

## 09:58 - 14:02 AI Labs Unexpectedly Drive Demand for Professional Services

Leading AI labs are surprisingly investing heavily in professional services, defying the common expectation that AI would automate consulting roles out of existence. This counter-intuitive trend highlights a significant need for human expertise in the age of advanced artificial intelligence.

The primary driver for this demand is that companies generally do not have internal teams sitting idle, ready to tackle large-scale projects like completely reimagining internal workflows, integrating AI into existing systems, or performing complex strategic analysis. These are the types of projects traditionally outsourced to firms such as Bain, BCG, McKinsey, Accenture, Infosys, branding agencies, or architecture firms, which provide the specialized personnel and focus needed.

The situation underscores a critical distinction between "tasks" and "jobs." While AI can automate individual tasks like writing code or creating a PowerPoint, the 'hard part of the job' often involves higher-level strategic thinking, problem-solving, project orchestration, and navigating complex organizational changes. These complex, non-coding challenges still necessitate human guidance and intervention.

Consequently, the rise of AI is not eliminating the need for professional services but rather reshaping it, creating new demands for external expertise to help businesses understand, integrate, and optimize AI technologies effectively. This shift suggests that human problem-solving and strategic implementation skills remain paramount, even as specific tasks become automated.

> You would think AI is going-- like, consultants were gonna be gone. No, we don't need- All these people anymore, AI is gonna do their work. Instead, like the most cutting edge AI labs are the ones most investing in these folks. It's, I think it's pretty surprising.

## 14:02 - 18:02 Automating tasks often leads to more complex work rather than job elimination

Making individual tasks cheaper or easier through automation does not necessarily reduce the demand for human labor or eliminate entire jobs. Instead, it frequently leads to an expansion of the job's scope and complexity. For instance, while Excel made financial calculations significantly faster, junior investment bankers still work long hours, as the overall complexity of their strategic tasks has increased. Similarly, advanced development tools automate much of software coding, but engineers now build more sophisticated applications, maintaining or even increasing the need for their skills.

The true value of many professional services lies not in executing specific tasks, but in defining what needs to be done and why. For example, clients don't hire McKinsey or Bain simply for a 75-slide deck, which AI can generate; they pay for deep analysis, understanding internal politics, assessing customer needs, and strategic guidance on *what* actions to take. This higher-level conceptual work, which involves navigating ambiguity and human factors, remains difficult to automate.

This pattern is evident across industries and historical periods. Just as Amazon efficiently delivers a known product (a SKU) but doesn't help you decide *what* product to buy, AI can perform discrete actions but often struggles with the strategic foresight and critical judgment required for complex roles. Even in fields like accounting, despite decades of automation tools from adding machines to ERP systems, the number of accountants has actually increased, indicating that making tasks easier often expands the possibilities for human work.

> What you actually pay Bain to do is to go and walk all over your, your company and work out, yes, but why is it that you didn't do that? And how do the politics of this work? And what do you actually need to do?

## 18:02 - 22:03 AI's Impact on Jobs Will Mirror Past Technological Shifts

Historically, new technologies automate existing jobs while simultaneously creating new ones, often leading to better roles and overall societal enrichment. This pattern has repeated for centuries; for example, people in 1800 couldn't conceive of a 'railway engineer' job. While these shifts cause "frictional pain" and job dislocation, the long-term outcome has consistently been economic growth and improved living standards, moving past concerns like crop failures.

Despite claims that AI is different, its rapid adoption is primarily due to existing digital infrastructure. Technologies like ChatGPT can quickly reach hundreds of millions of users because people already own PCs and have broadband access, unlike past innovations that required building entirely new foundational networks or hardware. This leveraging of existing platforms creates a compounding effect, making adoption faster, but it is not a fundamental break from historical technological progress.

The notion of an immediate AI-driven job apocalypse is further tempered by the realities of enterprise adoption. Integrating new AI solutions into large companies is not an overnight process; typical enterprise software sales cycles are around 18 months. Organizations won't instantly rip out established systems like SAP to replace them with new AI tools, meaning widespread, transformative job changes in the enterprise will unfold over several years, not months.

> You can always see the job that's gonna go away, and you don't know the new job because it doesn't exist yet, and it's something that sounds dumb anyway, like, 'What's a railway? Why would that be a thing?'

## 22:03 - 26:03 Technological transformations quickly become normal despite their profound impact

New technologies, including current AI advancements, follow a common pattern: their full transformative power isn't immediately obvious and takes years to unfold. Initially, people need time to discover all possible applications, even when the underlying technology has existed for a while. This delay isn't always about the tech itself, but about someone realizing a problem exists and identifying a way to solve it with available tools.

History offers clear parallels. An IBM ad from the 1950s described an electronic calculator, the size of a fridge, as being equivalent to "150 extra engineers," a claim strikingly similar to the value propositions of today's AI coding assistants. This highlights a recurring pattern in how foundational technologies are introduced and their potential articulated.

Massive shifts, once revolutionary, quickly become normalized and forgotten. For instance, barcodes allowed supermarkets to stock vastly more products by enabling efficient tracking, fundamentally changing retail. Before the internet, compiling data for a simple chart on supermarket product growth could take weeks of library research and phone calls. Today, the same task takes only hours with Google, demonstrating how the internet, a once-huge deal, is now an invisible enabler.

These examples illustrate that while new technologies like AI are indeed huge deals, humanity has navigated similar transformations before. We tend to forget the scale of past changes because they've become woven into the fabric of daily life, making us prone to underestimating the long-term, systemic impact of current innovations.

> An IBM electronic calculator, it's the size of a fridge, is like having a hundred and fifty extra engineers.

## 26:03 - 30:04 The Definitions of AGI and Superintelligence Are Constantly Changing

The definition of "AI" is a moving target, often described as whatever machines can't yet do. As AI capabilities advance, tasks once considered cutting-edge AI, like image recognition or sentiment analysis, become commonplace and are no longer categorized as "AI." This mirrors how other technologies, such as jet airliners, were once new and considered "technology" but are now standard.

Similarly, the definitions of AGI (Artificial General Intelligence) and superintelligence are undergoing constant redefinition. AGI is increasingly being redefined to mean what current AI systems can accomplish, such as performing a certain percentage of economically valuable work. This is a significant shift from earlier, more philosophical definitions that focused on consciousness or having a "soul." The relationship between AGI and superintelligence is also unclear, with no stable consensus on whether one encompasses the other or if they are distinct levels of capability.

Despite the fluidity of these terms and the uncertainty surrounding whether AI will ever achieve human-level intelligence, its utility is undeniable. Regardless of future breakthroughs or plateaus, current AI is already a profoundly transformative technology that will continue to change the world and be widely adopted across various sectors.

keyTakeaways": ["The definition of 'AI' is a moving target, consistently referring to tasks machines can't do yet, leading to constant reclassification as capabilities improve.", "Definitions of AGI and superintelligence are fluid and often redefined to align with current technological progress, rather than fixed, abstract concepts.", "Regardless of whether AI reaches human-level intelligence, it is already a profoundly transformative and useful technology with significant real-world impact."], "quote": "AI is whatever machines can't do yet."}```

## 30:04 - 34:05 AI expands the total addressable market and creates trillion-dollar opportunities, but value accrual is complex.

AI is significantly expanding the total addressable market (TAM), enabling companies to reach valuations that were previously unimaginable. This phenomenon extends the

software is eating the world

thesis, where each technological platform shift, from mainframes to PCs to mobile, progressively increased the market size. AI continues this trend by automating vast new sectors of the global economy, allowing businesses to address larger portions of the economy than ever before.

Similar to how electricity became fundamental to virtually everything, AI is expected to integrate into all industries, generating new value and jobs over time, despite potential short-term disruptions. However, the path for foundational AI providers to capture a proportionate share of this created value is not straightforward.

> We're gonna be selling AI intelligence on a meter like water or electricity...my dear sweet child, you need me to explain the modern structure of the utility industry to you.

## 34:05 - 38:06 Foundation Models May Become Commodities, Shifting Value to Applications

Just as telcos provide amazing global technology infrastructure but their stocks went nowhere because all the cool stuff is made by app developers, foundation models could face a similar commodity trap. The telcos once thought they'd build all the services users wanted, but instead, value accrued further up the stack, with companies like Apple and other app developers.

A key question for foundation models is whether the model itself does the whole thing, like a universal chatbot, or if it primarily serves as infrastructure for a vast ecosystem of applications. If model companies continue to build specific tools like 'Claude for X' that resemble Excel templates, these could become billion-dollar companies in themselves, or the value could shift entirely to independent apps.

The leverage of foundation models is crucial. Will they be like Windows, where standardization on the OS drove immense power, or more like AWS? With AWS, software buyers typically don't standardize on the cloud provider; the underlying infrastructure is less relevant to the end-user. If foundation models resemble AWS more than Windows, their direct pricing power will be limited.

If the chatbot isn't the primary user experience and applications are necessary, and if model companies aren't building all these applications, then the models themselves could become perceived as commodities. In such a scenario, where three to six companies sell a similar offering, pricing power for the model providers would diminish, and the majority of the value would accrue further up the application stack.

> if the models themselves are basically commodities... Then why would the model companies have pricing power? And wouldn't all the value be further up the stack?

## 38:06 - 42:08 The Application Layer Offers Greater Opportunity Than Foundational AI Models

Foundational AI model companies like OpenAI and Anthropic are unlikely to achieve winner-takes-all dominance or maintain high margins long-term. The lack of a clear differentiation and the presence of indefinite competition will prevent them from exercising significant pricing power, making their offerings more akin to a commodity.

The more significant and lasting opportunity in AI will be found in the application layer. This mirrors the evolution of cloud services, where diverse applications are built on top of underlying infrastructure, rather than a single dominant platform like Windows controlling the entire ecosystem.

While predicting the future of technology is challenging, and past predictions about the internet in 1997 or mobile in 2000 were often incorrect, the basic economic principle holds: undifferentiated commodity infrastructure providers will struggle to maintain high margins. Foundational models, despite their complexity, are viewed through this lens.

This perspective suggests that the scientific advancements in foundational models, much like the intricate technology behind flat-panel screens, will not inherently lead to high profits for the providers. Instead, the value will accrue to those building diverse applications that utilize these commoditized underlying services.

> They're undifferentiated commodity infrastructure providers.

## 42:08 - 46:08 Distribution Becomes the Primary Moat for Commoditized AI Products

Major technological shifts, like the transition to mobile internet or the rise of AI, often don't disrupt all incumbents equally. Companies like Google and Meta successfully navigated the mobile era by leveraging their existing platforms, while others struggled. A similar dynamic is emerging with AI, where established players are well-positioned.

When a core product or technology becomes a commodity, such as AI models, the value shifts significantly towards distribution. Much like web browsers, where the basic product is a thin wrapper over a rendering engine and design innovation is rare, the underlying AI model might not be the primary differentiating factor.

Incumbents possess a huge advantage by having vast existing distribution networks. Google can push Gemini through its services, and Meta can integrate Llama across its platforms, even if these models are merely "adequate." The power of default and user inertia means people are less likely to switch once a solution is integrated everywhere they already are.

This scenario poses a challenge for pure-play AI labs like OpenAI. They must aggressively pursue strategies to build strong flywheels, achieve user stickiness, and secure distribution before incumbents leverage their massive reach to make their own AI offerings the default for billions of users.

> if the product is a commodity, then the distribution is what matters.

## 46:08 - 48:10 Apple and Google Compete in On-Device AI Distribution

Apple's 2024 WWDC unveiled "Apple Intelligence," presenting a highly compelling vision for a personal AI assistant. This ambitious concept includes on-device, tool-using AIGC, designed to operate without prompt injection or hallucinations, and integrate seamlessly across ten thousand apps through a standardized API system. Despite the powerful vision, shipping such a complex system remains a significant challenge, with the product yet to be released.

Google is similarly pushing "Android Intelligence," powered by Gemini. The underlying AI models themselves are becoming commoditized, effectively serving as the "dumb thing underneath" that powers various features. The real competition is shifting from model performance to how these models are distributed and what unique product features are built upon them.

This creates a distinct battle for AI distribution at the device level. Apple's "Apple Intelligence," even if partially powered by Gemini, will offer a different user experience than Google's Android version. Google's marketing suggests a rollout primarily to its "most powerful devices," implying a more limited initial distribution, potentially reaching a smaller user base, such as Pixel phone owners.

> The models are commodity that powers different decisions about what the feature should be and what different distribution.

## 48:08 - 52:11 Analyzing the multifaceted anti-AI sentiment

Growing anti-AI sentiment is a complex mix of misinformation, unclear economic data, and cultural conflicts. For example, concerns about data centers' water usage are often exaggerated; they consume about 0.017% of US water, and local issues are typically planning problems, not an industry-wide crisis. Similarly, claims about data centers using 5% of US energy need to be contextualized as overall energy use.

The economic impact of AI, particularly concerning job displacement, remains largely unquantified. Economists struggle to determine if AI is taking jobs because there is a severe lack of transparent usage data from AI model developers. Without clear metrics like daily active users, researchers rely on academic studies and surveys to estimate AI's real-world adoption and effects.

A significant part of the backlash comes from cultural clashes, notably among creators. Artists who draw book covers or novelists are upset about AI-generated content, dubbed 'AI slop,' which can mimic their work without compensation. This extends to the increasing number of AI-generated podcasts, leading to broader debates over the ethical use and proliferation of AI-produced media.

The absence of reliable data on AI adoption and its actual effects on productivity and employment makes it difficult to assess its true societal impact. This data vacuum fuels speculation and allows misinformation to thrive, complicating public and policy discussions about AI's future.

> It's a big sort of fuzzy mess of different stuff, I think.

## 52:10 - 54:11 Parenting Children in the AI Era Requires Adaptability Based on Their Age

Public understanding of AI, much like early social media, involves a wide range of ideas, some true, some false, and many in a fuzzy middle. Concrete concerns from experts often don't resonate with the general public, where more abstract fears or misconceptions can dominate the conversation.

When considering how to raise children in an AI-shaped future, the approach heavily depends on the child's age. For children still in their early teens, the future impact of AI on their career path or life trajectory is highly uncertain, making specific, rigid preparations difficult.

However, for young adults nearing the job market, the landscape might be more settled, though still in unpredictable ways. This age-dependent variability means parents should prioritize flexibility rather than attempting to predict or over-engineer specific skills or paths for their children related to AI.

> I could be a lot, a lot more worried if I had a twenty-one-year-old, you know, I don't, I've got, you know, a kid in his sort of early teens. So it's a diff- those, those questions vary.

## 54:10 - 58:11 Every Wave of Technology Introduces New Ways to Cause Harm

New technologies consistently bring novel challenges and ways to negatively impact people's lives. For instance, sophisticated AI deepfakes can now create and disseminate damaging content rapidly, far exceeding previous tools like Photoshop. This pattern extends to social media, which, while connecting people, also inadvertently connected undesirable groups, amplifying societal problems.

The anxiety surrounding new tech is not unprecedented. The 1970s, for example, saw widespread panic over databases and their implications. Even older "1970s technology" can have devastating consequences, as evidenced by the UK Post Office scandal, where a buggy Fujitsu system falsely implicated hundreds of subpostmasters, leading to imprisonment, bankruptcy, and suicides.

While the scale and nature of threats evolve with each innovation, the underlying challenge remains. Society has always grappled with mitigating the downsides of technological progress. Instead of panic, a conscious and adaptive approach is necessary to navigate the inevitable disruptions and harms that accompany advancements.

> Every wave of technology comes with ways that you can ruin people's lives.

## 58:11 - 1:00:11 Navigating Unpredictable Career Paths and Overlooked AI Questions

Benedict Evans observes the evolving nature of careers, contrasting his own varied professional journey from equity analyst to consultant with traditional linear paths. He notes that while some professions, like software engineering, still offer clear trajectories, many people will experience diverse roles, making it challenging to predict long-term career paths.

This unpredictability makes offering specific career guidance, especially to younger generations, increasingly difficult. Instead of focusing on a singular job title, individuals must be prepared for continuous adaptation and a diverse set of professional experiences.

Evans also highlights critical, often unasked, questions regarding the AI landscape. One key insight is to differentiate between automated "tasks" and the broader "jobs" for which people are truly hired. This distinction clarifies what human value remains essential as technology advances.

Another significant question he raises concerns the economic dynamics of the AI industry, specifically whether AI model labs will maintain pricing power in the long term. This suggests a need for deeper analysis beyond current assumptions about the industry's profitability.

> What's the task and what's the job? What is just the thing that becomes a button or make the SKU versus what are people actually hiring you for?

## 1:00:11 - 1:02:11 AI's true potential lies in enabling new products and services, not just automating old ones.

The discussion highlights that technology's most transformative impact comes not from merely automating existing tasks, but from enabling entirely new possibilities. For instance, the evolution of recorded music revenue illustrates this shift: initially, digital music offered the 'old thing but more' by making individual tracks cheaper than CDs.

However, the real change came with services like Spotify, which redefined music consumption by offering unlimited access for a monthly fee. This wasn't just an online music store; it was a completely different model, fundamentally altering how people interact with music.

Similar transformations are seen with companies like Uber and Airbnb, which created new services that weren't feasible before the advent of their underlying technologies. These innovations didn't just improve existing services but unlocked entirely new markets and user experiences.

The crucial question for new technologies like AI isn't 'How do we do the old stuff but more?' but rather, 'What does this change? What wasn't possible before? What gets unlocked?' Often, these truly novel ideas initially seem 'crazy' until they become widely adopted, like Spotify, Uber, and Airbnb once did.

> What wasn't possible before? What gets unlocked, as opposed to just doing the old thing but more of it?

## 1:02:11 - 1:06:12 Predicting AI's impact on professions is a fundamentally flawed exercise

Quantifying how much a profession is "exposed" to AI by breaking down tasks into percentages is a deeply flawed approach. This mirrors the "expert systems problem" from early AI development, where attempts to recognize objects like cats by building rigid logical steps (e.g., edge detectors, fur detectors) ultimately failed. Similarly, one cannot accurately state that "seventeen percent of a senior partner's work at a law firm could be automated" because complex roles cannot be described or reduced to such discrete, automatable components.

The other major fallacy in predicting AI's impact is the inability to foresee how new technologies will transform seemingly unrelated professions. For instance, in 1997, no one would have predicted that the internet would profoundly reshape the taxi industry, as it seemed to have nothing to do with physical transportation. Yet, the internet-powered Uber completely changed the business. This highlights how professions not initially thought to be "exposed" are often the most dramatically affected.

The personal trainer profession offers another current example, where an AI on a phone could potentially build routines and correct form, obviating the need for a human trainer. This unpredictable exposure is common. While Uber "demolished" the taxi business in many cities while also making the overall market larger, Airbnb's impact on traditional hotels, particularly for business travelers who value services like room service and gyms, is less disruptive. Predicting these nuances is exceedingly difficult.

> I think this is just the most ridiculous bunch of deluded horseshit.

## 1:06:12 - 1:08:12 Dive Into AI to Succeed in a Changing Job Market

In an era of radical uncertainty driven by AI, many people are worried about their jobs and careers. While the broader long-term outlook might appear stable, individual prospects can face significant disruption. Just as broad statistics can obscure individual risks, the general future of work doesn't guarantee security for everyone.

For those in professions facing major shifts, particularly roles like associates in professional services, simply resisting or complaining about AI is unproductive. While expressing moral superiority might feel good, it offers no practical advantage in navigating the evolving job landscape.

The only actionable advice is to proactively engage with AI. This means completely submerging yourself in understanding what the technology can do, how it changes existing processes, and how you can leverage it. By doing so, individuals can adapt their skills and become a valuable hire, even if it means competing in a market where firms might hire 50 associates instead of 100.

> What helps is you diving into this, completely submerging yourself in it and understanding what you can do with it, how this changes things, how you can be a great hire.

## 1:08:12 - 1:12:12 Benedict Evans's AI Use Cases for Proofreading, Imagery, and Automation

Benedict Evans struggles to find direct applications for AI in his core analytical work, which involves precise information retrieval and synthesizing new ideas. He notes that the tasks he wants a machine to do are precisely what current AI models struggle with regarding accuracy, likening his situation to a lawyer or an accountant finding AI impressive but not directly useful for their unique, precision-based roles.

Despite these challenges, Evans has identified practical uses for AI in his daily life and work. He uses it for basic proofreading tasks to ensure clarity and correctness in his writing.

AI also proves useful for generating images, particularly for conceptual tasks. For instance, Evans utilized AI to visualize redecorating his apartment, experimenting with different paint colors, tables, and rugs in a room to see how they would look.

He values AI for automating what he calls the 'boring stuff,' rather than creative tasks he enjoys. An example is using voice transcription for dictating memos, which he considers a form of AI-powered automation, specifically using Apple Notes for this purpose.

> I don't want the AI to do the stuff I do for fun, I want it to do the stuff, the, the boring stuff that I don't do for fun.

## 1:14:13 - 1:18:13 Benedict Evans's Lightning Round: Classic Films, AI Apps, and Old Phones

Benedict Evans recommends engaging with classic cinema, citing Ingmar Bergman's "The Seventh Seal" as a brilliant and surprisingly short film worth watching. He also notes a current lull in new, "cool" iPhone applications, suggesting the initial "white space" for innovation has largely been filled.

Evans observes a notable absence of breakout consumer AI apps, attributing this to the economic challenge of scaling to 50 million free users before establishing a viable revenue model. He implies that the marginal cost structure of AI applications makes this difficult compared to previous app paradigms.

Sharing personal life philosophies, Evans states his mottos are "It depends" and "It'll probably be okay." He also maintains a collection of old phones from his past as a telecom analyst, specifically from the era before the iPhone.

He highlights the creative explosion in phone design prior to the iPhone's dominance, likening it to pre-wind tunnel car designs where varied shapes were used to differentiate. He possesses devices like an Ericsson "Shark fin" flip phone from 1998 and a 2001 J-Phone from Japan, notable for its color screen and camera, which was a cutting-edge feature at the time.

> It'll probably be okay.

---

Get podcast briefs for shows you follow: https://podbrew.app/