Jul 21, 2025 · 47m · a16z
The Future of Software Development - Vibe Coding, Prompt Engineering & AI Assistants
Martin Casado · 21m spoken Jennifer Li · 10m spoken Matt Bornstein · 9m spoken Erik Torenberg · 3m spoken
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this a16z podcast roundtable, partners Jennifer Li, Martine Casado, and Matt Bornstein join host Erik Torenberg to explore how artificial intelligence is transforming software engineering, modern infrastructure paradigms, developer productivity, and tech investment strategies.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. The host holds 7.2% of the talking time here. How this is scored →
The host as informed peer 2.5 Guest teaching 3.6 Guest disagreement 1.5 The host pushing back 1.1
speaking balance: gold is the host, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 28:48 Martin refutes Sam Altman's startup heuristic
Martin Casado explicitly rejects the premise of Sam Altman's framework regarding general model dominance, arguing that RL trade-offs make specialized models necessary.
Hardest push from the host ▶ 18:23 Host challenges dev tools viability narrativeThe host directly introduces the conventional VC skepticism that dev tools suffer from small market sizes, prompting the guests to defend infrastructure TAM expansion.
Biggest teaching moment ▶ 3:25 Martin explains application logic abdicationMartin Casado educates the room on a foundational computer science shift, explaining how AI differs from prior compute models by delegating decision logic rather than abstracting hardware resources.
The host holds their own ▶ 18:23 Host cites historical VC anti-pattern on dev toolsThe host demonstrates sharp industry knowledge by identifying how venture capitalists historically misjudged developer tools as having small TAMs before massive exits like GitHub.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The host as informed peer | Guest teaching | Guest disagreement | The host pushing back | Why |
|---|---|---|---|---|---|---|
| Defining Modern Infrastructure and AI as the Fourth Pillar | 2 | 4 | 1 | 1 | The host opens with standard exploratory prompts about defining infrastructure versus enterprise and asking if AI models represent a fourth layer. Martin Casado and Jennifer Li educate the host by defining technical buyers and explaining how AI models differ from historical abstractions by abdicating logic rather than just resources. | |
| Software Being Disrupted and the Layering Effect | 2 | 3 | 1 | 1 | The host asks a broad question on super cycles and what can be learned from past transitions. The guests collaboratively reframe software disruption, detailing how natural language prompt interfaces fulfill the long-standing promise of low-code tools. | |
| The Evolution of a16z Infra Practice and the Shift in Technical Buyers | 3 | 4 | 1 | 1 | The host brings specific context regarding Martin Casado being an early portfolio founder at a16z to prompt a historical overview. The guests detail the subtle differences between selling to centralized technical buyers versus vertical SaaS markets. | |
| Historical Waves of Infrastructure: On-Prem, Cloud, Mobile, and COVID | 2 | 3 | 1 | 1 | The host asks the guests to trace historical infrastructure waves since 2009. Martin Casado and Jennifer Li provide a timeline covering pre-cloud, SaaS recurring revenue metrics, COVID-driven remote adoption, and the current AI transformation. | |
| Infrastructure Subcategories: Dev Tools, Foundation Models, and Data Systems | 4 | 3 | 2 | 1 | The host demonstrates domain knowledge by pointing out that VCs historically wrote off developer tools due to small TAM assumptions. The guests build on this, explaining how infrastructure inherently expands TAM rather than occupying static markets. | |
| Defensibility, Value Accumulation, and Switching Costs in AI | 2 | 4 | 2 | 1 | The host asks how defensibility applies across app and model layers in AI. Matt Bornstein and Martin Casado refute naive commoditization arguments, explaining market expansion dynamics and why infrastructure switching costs remain exceptionally high. | |
| Specialized Models, Context Engineering, and Karpathy's Software 3.0 | 3 | 4 | 2 | 2 | The host introduces external quotes from Sam Altman and references Andrej Karpathy's framing to prompt debate. Martin Casado pushes back on Altman's premise, arguing that specialized reinforcement learning trade-offs prevent single general models from dominating every task. | |
| Human Centricity in Software, Agent Capabilities, and Strategic Integration | 2 | 4 | 2 | 1 | The host prompts discussion around current internal debates, AI agents, and market structure. Matt Bornstein offers a contrarian framing against agent marketing hype, highlighting how uncorrected LLM error loops degrade agent performance. |
Statements from this episode (23)
Martin Casado: AI is disrupting software itself
“Software was always the disruptor. One of the most exciting thing about the AI wave is, like, software's being disrupted.”
Jennifer Li: AI models are the fourth pillar of tech infrastructure
“I certainly think of as a fourth layer of infrastructure. It's you know, it certainly leverage and build on top of all the three pillars we're talking about. It has a lot of demand of compute, and of course it's trained and also producing a large amount of dat…”
Casado: AI is the first computing era where developers abdicate logic
“I don't remember ever in, like, the history of computer science where we've, like, from an application standpoint, we've abdicated logic. Like, actual, like, application. Like, in the past, we've abdicated resources. Like, you were like, give me compute. Give …”
Jennifer Li: Software infrastructure never disappears, it gets layered
“Infrastructure never goes away. It just gets layered.”
Casado: Lowering marginal technology costs expands TAM and creates startup opportunities
“So one of them is often when you bring the marginal cost of something down, like with compute, you know, we did it for computation and for the internet, we did it with distribution. It increases the TAM a whole bunch. So, so for one, you almost always see this…”
Bornstein: Horseshoe theory applies to software buyers as dev tools mirror consumer
“Horseshoe theory of software buyers happening now, where you have, like, consumers over here, and then you kind of work up to, like, apps, and, like, you would think that, like, infra is way over here, but it's actually kind of bending back around. Like, devel…”
Li: Global developer count has grown beyond 50 million
“When I first joined, you know, venture, or maybe even just started thinking about infrastructure developers deal, like, in the, Low tens of millions and they're becoming like the next generation of consumers. Now they're definitely the next generation of consu…”
Casado: AI wave is most dramatic tech shift in 30 years
“The AI transformation of the last three years has been the most dramatic I've seen in, you know, the last 30 years of being in this industry.”
Bornstein: Cursor is a16z's top developer tool company right now
“Cursor is probably our top developer tool company right now.”
Casado: Early in tech cycles, infrastructure and apps are indistinguishable
“Early in super cycles, it's very hard to distinguish between an infra company and, like, the application companies, and the reason is because the TAM is so small and so new, the new technology becomes the app, right?”
Bornstein: Simple AI wrapper startups no longer exist
“And like, we're like pretty clear past the sort of rapper phase. Like, I don't think there are any rappers anymore. Like, Building good products with AI is really hard, and the founders doing it now have, like, really good kind of intuition for how to do it.”
Casado: Consolidated AI infrastructure companies will maintain high profit margins
“For the AI wave, yeah, for sure, like, this is extreme, but it will slow down, and then the consolidation will happen, but I guarantee you'll just end up with these great companies that maintain margin, like, AWS still has great margins, you know, Google still…”
Li: API and infrastructure products have much higher switching costs than SaaS
“It just turns out, in general, the switching cost of infrastructure piece is so much higher. Even with like API business, people tend to think you can just like switch over to another API. There's so much logic embedded in Calling the API in the software itsel…”
Jennifer Li: Production AI apps require composing multiple models, not one monolith
“When we talk about complex systems, you cannot just like use one model that drives everything, at least not today. But you can compose like, you know, very capable and powerful models to like, you know, take certain tasks and also chain together, you know, pro…”
Casado: Software development will have formal AI engineering methods in five years
“And I truly believe in five years, we'll look back, we'll come up with a whole new, you know, set of formal ways to build software and they will have strong guarantees and we'll understand them and, you know, there'll be all the tools for it, et cetera.”
Casado: AI software development will still require professional engineers
“And I think where, where we've landed is, this is a real disruption. It is changing all of software. It'll look something, place different. But it is still gonna write, require professionals. And I do think that, like, the statement, it'll require professional…”
Bornstein: AI developer tools will increase, not shrink, software engineering headcount
“I think the best way to think about this is simply that we're going to have more developers. I think it's very unlikely that like we're going to shrink development teams because we have amazing new tools. Like that's just kind of not how these markets have wor…”
Casado: The median software PR changes only two lines of code
“Well, I think that's the median. It's the median is two.”
Bornstein: Synthetic data will not produce self-improving AI models
“The question is, like, does this lead to sort of, like, a self-improving utopia of models or not? And I think we have some pretty strong opinions on the not side of that.”
Casado: Companionship and therapy are ChatGPT's top use cases, not coding
“Like, if you look at the most common use cases of something like ChatGPT, I think the top one is like companionship and therapy, and then it's like managing my schedule, and it's like the top of the pyramid of need stuff. And then number five is professional d…”
Bornstein: AI coding agents succeed because code environments provide error-correction loops
“If you take the simplest definition that basically an agent is a, is an LLM running in a loop a very simple way to think about this is errors propagate throughout the loop, right? So if you have a small error, it gets worse and worse. And this is why a lot of …”