The End of Software Scarcity

Until the recent proliferation of AI coding agents, software has been scarce because it's been very hard to do well. Malte Ubl, CTO of Vercel, on The Room Podcast describes the problem as humans trying to coax electron systems into understanding and representing the way we see the world. I have horribly butchered the paraphrase and inserted a lot of my own words, but the idea stands as I don't think anyone would argue that coding bug-free software that scales is easy.

As AI increasingly writes more code, software itself is becoming abundant. This has led to two opposing viewpoints. The first is a glass-half-empty perspective: AI, as a platform shift, will erode SaaS margins, leaving only a few remaining years for meaningful value creation in software. This belief helps explain the current frenzy in venture capital funding.

I subscribe to the glass-half-full view. An abundance of software will be a net positive, unlocking entirely new greenfield opportunities for products and businesses.

A Software Plentiful World

I like to think of AI the way I think about irrigation during the Agricultural Revolution. Before irrigation, farming depended on rainfall or proximity to rivers, both of which were scarce constraints. Irrigation made water plentiful and enabled agriculture across vast stretches of dry land. As a result, grains and crops became commoditized. When economics of food production stopped being the constraint, creativity became the new differentiator, giving rise to experimentation in cooking and, ultimately, the restaurant and hospitality industry. This idea builts on top of Jevons paradox.

Likewise, I believe that as AI removes the economic constraints of building software, we will see a wave of genuinely new and exciting outcomes. Many of the top software engineers I know report that AI-driven productivity gains have not led to working less, but instead have enabled them to focus on problems they previously could not tackle. This is what I mean by the world becoming more greenfield. As a society, there are many problems we have long wanted to solve but could not because of economic constraints. As those constraints are lifted, we gain the ability to focus more of our energy on improving the world we live in.

So then, let's fast foward 5 years and assume that most workflows are codified and the logic that we need for our jobs is completely commoditized. Will there be any jobs left in technology?

Taste as a Statement

If we go back to the restaurant example, commoditized ingredients did not eliminate differentiation; they shifted it. When everyone has access to the same grains, proteins, and tools, the question stops being what can we cook? and becomes how should we cook it? Cuisine, atmosphere, service, and presentation become the primary driver of value. Taste becomes a statement.

The same dynamic will play out as software becomes abundant. When building software is no longer the primary constraint, improvement shifts from what should be built to how it should be built. The differentiator is no longer technical feasibility, but human preference. Products begin to reflect point of view: how they feel, how they sound, how they fit into daily life, and how they respect human attention.

This is where software starts to redefine how humans interact with the world around them. We should expect new interfaces and modalities to emerge that feel more natural and expressive: voice and audio-first experiences, richer visual and spatial software, and systems that adapt to individual workflows instead of forcing humans to adapt to them. Much like restaurants cater to different tastes, software will increasingly cater to different modes of thinking, communication, and creativity.

As promising as this sounds, it comes with less obvious but important second-order effects. As software becomes easier to create and more deeply embedded into everyday life, its failures become easier to amplify as well. Poor defaults, misaligned incentives, or subtle design mistakes can propagate at unprecedented scale. When everyone can ship, the cost of getting it wrong rises.

This sets the stage for the next shift: a world where guardrails, constraints, and policy are not obstacles to progress, but prerequisites for it.

Policy as Code

As software becomes easier to create and deploy, governance shifts from an external constraint to an internal system property. Guardrails can no longer be bolted on after the fact through manual review or slow regulatory cycles. Instead, they must be designed into the infrastructure itself, operating continuously and at scale.

In a software-plentiful world, this takes the form of policy-as-code. Permissions, identity, provenance, audit logs, rate limits, and kill switches are codified from business requirements and enforced automatically rather than procedurally. Regulation moves upstream into platforms and primitives, where constraints can be applied uniformly and adjusted dynamically based on context, risk, and intent. Compliance becomes executable, observable, and testable.

I had the privilege of meeting with the product team at Norm.ai a few months ago to discuss their LEAP platform. They have seen strong results by investing in a dedicated product function that enables their lawyers to codify law directly into logic. By using a domain-specific language, each lawyer can become a 10× legal-agent builder without leaving their domain expertise or learning the language of bits and bytes.

This raises a broader question: who will build these domain-specific abstractions for every other industry vertical?

I’m excited to share that I’ll be stepping into a new role at JPMorgan at the start of the new year, where I’ll be working on deploying LLM-based applications and agents to help Fortune 100 treasury teams move money with greater intention. The challenge goes far beyond building compelling demos; it’s about ensuring AI-powered workflows are durable, compliant, and reliable at massive scale, within a payments business that generates over $5B per quarter. What excites me most is gaining hands-on experience with AI-driven compliance in a G-SIB and learning how to encode regulatory requirements and decision logic into production AI workflows.