The Future of Software Development - Vibe Coding, Prompt Engineering & AI Assistants
A discussion on the state of technical infrastructure, focusing on how AI and Large Language Models represent a new, fourth foundational pillar alongside compute, network, and storage. The talk covers how AI is disrupting software itself, the investment landscape, and the future of the developer profession.
Infrastructure never goes away; it just gets layered. The current AI wave represents the most dramatic transformation in the last 30 years, creating a new, fourth pillar of infrastructure that is fundamentally disrupting software itself.
AI as the Fourth Pillar of Infrastructure
Traditionally, infrastructure consists of three pillars: compute, networking, and storage. AI and models have emerged as a fourth pillar, not just leveraging the other three but fundamentally changing the requirements for them. Building and programming for AI requires different data centers, different chips, and different programming models.
The most significant change is the abdication of logic. Historically, software abstracted resources like compute or storage, but the core logic—the “if-then” decisions—always came from the programmer. With modern AI, we are asking the model to “come up with the answer for me.” This is a fundamental shift in computer science, forcing a rethink of what it means to be a programmer and what software is.
People are trying to reason about this new pillar by analogy—is it a database because it answers queries, or a network because it’s non-deterministic? In reality, we’re starting from a blank sheet of paper, which is what makes this moment so exciting.
Software is Disrupting Itself
For decades, software has been the disruptor of other industries. In a historic turn, software is now disrupting itself. This follows a classic pattern for technology super cycles:
- A new technology dramatically lowers the marginal cost of a core function (e.g., the internet lowered the cost of distribution).
- This leads to a massive expansion of the Total Addressable Market (TAM).
- New user behaviors emerge that incumbents are not equipped to handle, creating whitespace for startups.
The promise of “low-code” is finally being realized, but the new programming language is natural language. This gives creators and developers unprecedented leverage to prototype and build applications.
The Modern Infra Landscape
The definition of infrastructure is anything used by a technical buyer—a developer, data scientist, analyst, or DevOps professional—to build software. This focus on a horizontal, technical buyer is distinct from vertical SaaS, which targets specific industries.
Key areas of focus include:
- Developer Tools: Anything that makes developers’ lives easier and more efficient (e.g., Cursor). The TAM for developer tools, once considered small, is now massive.
- Data Systems: The backbone for data processing and analysis, from large-scale engines like Spark to analyst-focused tools like Snowflake and dbt.
- Foundation Models: The line between infrastructure and application is blurry here. Companies like OpenAI or 11 Labs provide core model infrastructure via APIs while also building consumer-facing applications (ChatGPT, voice creation studios).
A key trend is that the developer go-to-market now resembles consumer marketing. With over 50 million developers and growing, they are the new consumers, making individual adoption decisions before they become enterprise-wide buying decisions.
Defensibility and the Path Forward
During an expansionary phase like the one we are in, zero-sum thinking is a fallacy. Value is accruing at every layer of the stack, from chips (Nvidia) to models and applications. Switching costs for infrastructure remain high due to the complexity of integrating systems. Defensibility comes from deep domain expertise and the earnest secrets learned from building complex systems over time.
The key challenge in building with AI is not “prompt engineering” but “context engineering.” The real work is in building the systems and data pipelines to feed the right context into a model to get the best performance. This is a classic infrastructure problem that creates new patterns for building software.
The Future of the Developer
AI is a massive productivity boost for developers, not a replacement. We will likely have more developers, not fewer, as the barrier to creating software is lowered.
A model is a tool; programming is a fundamentally creative job of bringing something new into existence. The most difficult part of software development isn’t writing the code—it’s understanding user needs, defining requirements, and designing a workflow. That human-centric task does not go away. AI tools will handle more of the rote coding, allowing developers to focus on these higher-level design problems.