When Code Gets Cheap: Value Shifts to Judgment and Systems

AI is making code cheap, creating both the possibility of fewer programmers (substitution) and more software demand (Jevons’ paradox). Because this deflation stems from real productivity gains, it can accelerate innovation, shifting value from typing code to choosing, integrating, and designing systems. Beck advises embracing commodity tools for routine work while investing in understanding, integration, and judgment that matter in any scenario.

Key Points

• AI lowers the cost, skill barriers, and time to produce software, creating both substitution effects and Jevons-like demand expansion.
• Programming deflation differs from macroeconomic deflation: it’s productivity-driven and can accelerate innovation via reinforcing tool improvement loops.
• As code becomes a commodity, value migrates to problem selection, systems thinking, integration, and navigating complexity.
• Quality bifurcates: a flood of cheap, mediocre code coexists with a premium on carefully crafted software; the middle shrinks.
• Practical strategy: use AI for obvious work, focus on integration, develop taste, and build skills (understanding, judgment) that win under any future.

Sentiment

Mixed but pragmatic: commenters broadly agree on AI’s current utility and the shift in value toward integration and judgment, while expressing strong skepticism about reliability, last-mile progress, and pure “deflation” claims.

In Agreement

• AI already boosts productivity for routine, boilerplate, and exploratory coding; many won’t return to pre-LLM workflows.
• Value is migrating from typing code to integration, architecture, problem analysis, and deciding what to build—classic senior responsibilities.
• As costs fall, experimentation and attempted complexity increase; integration becomes the new bottleneck.
• Jevons-like dynamics are plausible: cheaper code can expand demand and change what gets built, even as substitution reduces labor for some tasks.
• Pragmatic hedge: adopt commodity AI where it works; invest in human capabilities (judgment, systems thinking) that matter under multiple futures.

Opposed

• LLMs are unreliable and inconsistent; the last 10% may be intractable, so comparisons to compilers or assumptions of steady progress are flawed.
• Once you factor rework, debugging, and verification, code isn’t truly cheaper; deflation is an overstatement—commoditization is a better frame.
• Overuse of AI on unsuitable tasks destroys productivity (a Laffer-like curve); domain-constrained use works, general use does not.
• AI could trigger an overproduction crisis in software: far more code than organizations can absorb profitably, depressing wages and creating cleanup burdens.
• Risk of vendor lock-in and loss of code understanding if teams rely heavily on AI; proliferation of low-quality or “soul-less” code.
• AI threatens junior roles and outsourcing first; some argue seniors should resist normalizing AI as a replacement for junior developers.
• Regulatory/audit pressures may grow as code volume and divergence from norms increase; current SAST/DAST false positives show oversight challenges.