Why Jevons Paradox, Not the Silicon‑Carbon Ratio, Explains the Future of AI‑Powered Software

Jevons Paradox describes a simple phenomenon: when something becomes sufficiently cheap, people tend to use more of it rather than less.

Applied to AI‑assisted coding, the paradox means that once writing software becomes as inexpensive as utilities, firms that could not afford software services before will become viable customers.

This shift suggests AI coding may not immediately reduce the number of programmers; instead, it could free a large pool of previously untapped, cost‑constrained demand, creating a need for more developers who understand industry contexts and can translate business problems into software.

Capace, in a No Priors interview, highlighted three insights that reinforce this view:

First, AI cannot produce truly valuable code. While AI can write code and is getting better at it, valuable software requires business judgment, system abstraction, engineering trade‑offs, and many implicit rules that rarely appear in formal requirements.

Second, the Jevons Paradox in practice. Historically many enterprises could not afford software; they relied on Excel because SaaS was too expensive, custom development was prohibitive, and self‑built teams were unsustainable. When AI coding drives development costs down, the decision shifts from “let’s handle it manually” to “let’s try a small system”. This change is not a substitution of existing stock but an expansion of new, previously suppressed demand.

Third, software lifecycles will become increasingly short and even one‑off. When software was costly, teams aimed for reuse, platformization, and standardization, targeting thousands of customers per feature. With cheap software, we will see many temporary or short‑cycle solutions: a two‑month internal tool, an ad‑hoc workflow system built on‑site, or an automation dashboard for a single event. These are analogous to how cheap paper gave rise to sticky notes.

Summarizing the three points: AI lowers the cost of writing code, but valuable software still depends on deep business understanding; the cost decline releases a massive long‑tail of demand; and that demand spawns numerous short‑cycle, highly contextual, non‑standard software projects.

Consequently, programmers are likely to split into two groups:

1. Industry‑knowledgeable Full‑stack Developers (FDEs). Their value lies not in raw coding speed but in translating hidden industry rules into functional software, understanding why customers struggle to articulate requirements, why certain Excel sheets persist, and how systems behave after deployment.

2. Freelancers amplified by AI. AI magnifies an individual’s delivery capacity, enabling a one‑person company to take on many small, fragmented enterprise projects that previously required a team.

Both groups share a common prerequisite: deep industry know‑how. Simply letting AI write code without understanding the client’s context is like increasing typing speed from 80 to 800 characters per minute—fast but not valuable.

These dynamics reveal a sizable market opportunity: a platform that matches enterprises with industry‑savvy developers, defines clear delivery criteria, and ensures post‑deployment acceptance and basic operations. Existing platforms focus only on posting requirements, quoting, and matching, which is insufficient because enterprises fear buying the wrong software, unreliable developers, project failure, and lack of maintenance. The truly valuable services would be: helping firms find industry‑knowledgeable talent, co‑defining what constitutes completed delivery, and ensuring the small software runs reliably after launch.

Observations from a recent tech‑community meetup illustrate differing CTO attitudes: consumer‑facing (ToC) CTOs see rapid AI impact and layoffs; government (ToG) CTOs are more measured; business‑facing (ToB) CTOs are conflicted because they face both efficiency pressures and the newly released long‑tail market. ToB customers are highly non‑standard, on‑site, and price‑sensitive, making them prone to abandon software when it seems too costly. With lower costs, these demands will re‑emerge.

In conclusion, AI coding represents a market revolution rather than merely an efficiency boost. Code becomes cheaper, but software demand is unlikely to shrink; instead, it may grow, become more fragmented, and stay closer to the operational front line.