Stewart Squared
Episode #87: Tighter Than Microsoft, Smarter Than Apple: Anthropic's Blueprint to Own the AI Stack
- AI & Agents
- Money & Sovereignty
- Claude
- Anthropic
- coding agent
- planning agent
- flow state
- SaaS
- harness
- agents.md
Listen
Download MP3About this episode
In this episode of the Stewart Squared podcast, host Stewart Alsop is joined by his father, Stewart Alsop II, to talk through a wide range of topics stemming from their shared obsession with AI and technology. The conversation kicks off with Stewart's frustrations around recent changes to Claude that have disrupted his morning workflow of building his own coding and planning agents, leading into a broader discussion about Anthropic's business strategy versus OpenAI's, the Apple-versus-Microsoft analogy for how AI companies are positioning themselves, and why Dario Amodei keeps making bold claims about AGI while struggling to serve existing customers. From there, the two branch out into how large enterprises — from banks to airlines — are using AI to replace legacy systems like COBOL, the historical parallels between today's AI disruption and the industrial revolution, the nature of large organizations and whether they're even a permanent feature of human civilization, and finally, Stewart Alsop II's own career arc from journalist to venture capitalist, including near-misses with Elon Musk's x.com and reflections on what separates great investors like Mike Moritz and John Doerr from the rest of the pack. Stewart Alsop II also mentions his newsletter, where readers can find his takes on figures like Sam Altman, and recommends the book about the founding of Benchmark Capital for anyone interested in what makes a great investment partnership.
Timestamps
00:00 - Stewart describes his morning flow state routine, copying and pasting between planning and coding agents while removing SaaS dependencies using Claude.
02:00 - Claude's recent model downgrade sparks frustration, as Anthropic quietly reduces reasoning quality to manage server capacity for new users.
04:00 - OpenAI versus Anthropic contrasted through Sam Altman's business-only approach versus Dario Amodei's strategic geek leadership and company vision.
07:00 - Anthropic's enterprise strategy revealed as enabling internal software developers to build applications faster, replacing outside SaaS vendors entirely.
09:00 - The Claude Code harness and agents.md standardization debate shows Anthropic deliberately rejecting open standards to build proprietary infrastructure.
13:00 - Microsoft and Apple analogies debated, concluding Anthropic resembles Apple's hardware-software integration model rather than Microsoft's vendor lock-in approach.
18:00 - Large company IT departments explored, examining how AI transforms legacy infrastructure management across enterprises with thousands of employees.
22:00 - COBOL replacement emerges as Claude's killer enterprise use case, allowing companies to modernize decades-old systems without breaking operations.
27:00 - Decentralization and democratization of AI discussed alongside Anthropic gatekeeping new models from consumers while slowly releasing them to enterprises.
31:00 - Industrial revolution parallels drawn to current AI disruption, questioning whether large organizations are eternal or merely industrial-age phenomena.
39:00 - Job displacement fears examined through historical disruption patterns, concluding predictions about white-collar job losses remain fundamentally unknowable.
44:00 - Stewart Sr. explains his career shift from journalism to venture capital, driven by financial incentives and timescale differences between reporting and investing.
49:00 - Hall of fame investors compared, revealing no consistent pattern among legends like Draper, Moritz, and Doerr beyond individual instinct and partnership dynamics.
55:00 - Partnerships examined as the core unit of venture capital success, with Andreessen Horowitz and Benchmark cited as rare examples of scalable partnership models.
Key Insights
1. Anthropic has shifted its business strategy away from serving individual power users and toward enterprise clients. The company has moved to block third-party harnesses and push all users toward API pricing, signaling a deliberate pivot to lock in large corporate customers who use AI to modernize internal software infrastructure.
2. The difference between OpenAI and Anthropic comes down to strategic consistency. Dario Amodei set a clear direction when Anthropic was founded and has stuck to it, while Sam Altman has bounced between acquisitions and announcements without a coherent throughline. Great companies, as observed historically, define a strategy and follow it.
3. Claude's recent model changes represent a deliberate downgrade in reasoning quality to manage server capacity. The version jump from 4.6 to 4.7 was a number change, not a capability upgrade, and existing users are experiencing degraded relevance realization as Anthropic accommodates a larger user base on the same infrastructure.
4. The most transformative use case for AI in large companies is replacing legacy systems like COBOL with modern applications. AI can analyze decades-old code, identify vulnerabilities, and rebuild infrastructure without disrupting operations, potentially allowing companies to shrink large developer teams dramatically while improving performance.
5. The future of large organizations is not elimination but greater efficiency. Large companies will always exist to manage scaled operations like airlines or manufacturing, but AI fundamentally changes how many people are needed to maintain and develop the software that runs them.
6. Every major disruption in history has produced fear of widespread job loss, yet outcomes have generally been better afterward. Predictions from figures like Dario Amodei about mass unemployment are speculation dressed as logic, and the actual future remains unknowable until it becomes the present.
7. Successful venture capital partnerships have no single replicable formula. Hall of fame investors like Draper, Moritz, and Doerr each use entirely different decision frameworks, and the health of a partnership depends more on how the specific partners interact with each other than on any universal system or methodology.
Episode transcript
Everyone is wrong about AI, but if you know the history of technology, you can be wrong and still be right.
I know the history since I started in technology with the personal computer in 1980. So listen in while we explore what is true about AI. Yeah. Okay.
Welcome to the Stuart Squared podcast. Got a very exciting episode today. My dad is in New York, so we're on the same time zone, and we're doing it in the morning, which is my preferred time to do it. I've already been up for three or four hours now. and each morning I get up and I spend my mornings in flow state, most of the time. Flow state, as I just go from planning agent to coding agent to plan, planning agent, decoding agent, just copying and pasting. I'm a glorified copy and paster. And then I ask a good question every once in a while that sets the stage to where. Where we go. and so I'm, you know, the last couple episodes we've talked about how Claude did me dirty. and they've done a lot of other people dirty as well. and I am reacting it to the way that I've was using Claude before, which was I've. I'm basically removing all of my dependencies on. On SaaS software one by one. And that was going really quickly until they did me dirty. and now I have to focus all of my cognitive energy on building my own harness for a coding agent and a planning agent. because we're running into this thing where it's just like, they don't have enough compute. They're making all these things like, oh, we got AGI in the background. We're going to destroy everyone. All the Dario keeps on telling everybody. Everybody makes fun of Dario right now on Twitter because he's like, oh, we're not going to have any jobs left. We got AGI, we got Mythos. We're going to do all this different stuff, and then they can't serve the customers that they currently have. So, like, so it's clear that there's like a difference between the exoteric and the esoteric, like what they're telling people. So all that means basically, that I am spending my mornings, each morning figuring out how I can reverse engineer Claude. And it's funny because Claude is completely open to helping me reverse engineer it. but the key factor is, is that the quality in the reasoning, the quality in the, in its thinking context, its ability to pick out, what I really after, its, relevance realization, its ability to understand what's relevant has been hit really hard by the recent changes. All the recent. They didn't upgrade the model from 44.6 to 4.7, although they increased the number. They did not upgrade the model. They basically downgraded it according to how they can use their current server capacity to, allow a whole bunch of new people in at the same level as also like old timers that me that are figuring out how to get more and more out of the $200 subscription. So it's such an interesting arms race. we can. We're going to talk about that. But also, as with the show, we can go into the history. and I have a specific question about, like, this is all geeks who have been doing R and D since at least 2015, but, probably earlier than that. And now it's their time and now they're starting companies. And I want to understand what happened historically because I feel like they're geeks and they don't understand business. Is that a real thing in the past? Like, do geeks have a fundamental mis. Misunderstanding about how to turn their R and D into business products?
Your first full transcript is free. After that, an email opens every transcript in the index — a list of readers we can write to, not a guest book.