Jul 21, 2025, 9:00 AM ETPodcast

The Future of Software Development - Vibe Coding, Prompt Engineering & AI Assistants

Infrastructure is defined internally as "technical buyer" products used by developers, data scientists, analysts, and administrators to build software, distinct from vertical SaaS used by non-technical roles.
A new infrastructure layer fundamentally changes programming methods and the surrounding software stack, necessitating rethinking of resource, memory, and latency requirements.
The current landscape comprises four pillars: compute, networking, storage, and the emerging "fourth pillar" of AI models, which leverages the other three while introducing intelligence to software.
AI models are categorized as infrastructure because they require specialized hardware (data centers/chips) and non-deterministic programming models that differ from traditional logic-driven software.

The most significant disruption in computer science history is the shift from programs executing explicit logic to AI models "abdicated logic," where systems generate answers rather than following rigid if/else programming structures.
AI is described as "software eating software," a unique super cycle where the disruptor (AI) is also the technology being disrupted, unlike previous waves where software merely disrupted external industries.
This shift moves the focus from abstracting resources (compute/storage) to abstracting decision-making processes, creating a "blank sheet of paper" scenario for developers and engineers.

TAM Expansion: Infrastructure investments historically expand Total Addressable Markets (TAM) by lowering marginal costs, creating new user behaviors and white space for startups, similar to the internet and microchip eras.
The Developer as Consumer: Developers are transitioning into "next-generation consumers," with the average infrastructure decision value rising from the low tens of millions to approximately $50 million, blending B2B and consumer-like adoption patterns.
Investment Shift: The firm split its practice into distinct "Apps" and "Infra" funds after realizing that evaluating technical buyers (centralized IT/developers) requires different due diligence than evaluating vertical market buyers (e.g., flooring companies).
Zero-Sum vs. Expansion: The current AI market is in an expansion phase where zero-sum thinking is detrimental; historical trends suggest layers eventually consolidate into oligopolies or monopolies while maintaining healthy margins.

Context Engineering: Prompt engineering is evolving into "context engineering," focusing on how to feed the correct data and tools into models to drive performance, requiring formalized data pipelines and indexation.
Agents and Error Correction: Coding agents show high efficacy when integrated with error-correction loops (linting, compilation), but general "web browsing" agents struggle due to error propagation without such safeguards.
Specialization vs. Generalization: The future likely involves a mix of highly general models and smaller, specialized models tailored to specific tasks (e.g., RL-trained for code), rather than a single model dominating all use cases.
Synthetic Data Debates: There is ongoing internal debate regarding the limits of synthetic data, with the consensus that it cannot lead to a self-improving utopia without the introduction of new real-world information.
Low-Code Realization: AI is considered the catalyst that finally realizes the low-code promise, replacing Python scripting with natural language for non-technical users to prototype and build applications.

Defensibility Evolution: While traditional infrastructure defensibility relied on deep domain expertise and long development cycles (e.g., creating a new database), AI defensibility now depends heavily on distribution, developer attention, and specific data context.
Switching Costs: Infrastructure switching costs remain significantly higher than SaaS due to deep system integration and embedded logic, preventing easy migration even with API-based access.
Vertical Integration Trends: The market is seeing a simultaneous push toward vertical integration (e.g., OpenAI building end-to-end consumer experiences) and horizontal specialization (e.g., API providers serving diverse sectors), with no clear winner in either model.
Role of Professionals: The consensus is that AI will not eliminate programmers but will require more of them; the role shifts toward defining formal systems and requirements, as natural language lacks the precision needed for complex system specification.