Claim
When coding agents author most production code, teams select programming languages based on how well agents perform in them, not on developer ergonomics or personal preference.
Mechanism
Developer ergonomic qualities matter while the developer is the primary author: concise syntax, expressive type system, pleasant tooling. Once agents write most code and developers shift to review and direction, the author-enjoyment frame becomes irrelevant. The qualities that predict agent success are different: richness of training data for that language, consistency and predictability of its idioms, quality of compiler and runtime error messages. A language an agent produces accurate code in is more valuable than one a developer enjoys typing. The decision frame shifts from "I enjoy working with it" to "my agent gets great results with it."
Conditions
Holds when: coding agents write the majority of new code and developers review and direct rather than author directly. Fails when: agents remain unreliable for the language domain, or the team is primarily junior engineers still developing their own writing fluency and needs language ergonomics to learn effectively.
Evidence
Ball runs three daily production applications written entirely by agents without knowing which language they use, demonstrating that the developer-preference frame is already dissolving in practice for individuals who have adopted agents as primary authors.
"Agents are a hundred times bigger than any language, framework, or library choice."
The natural implication: teams now optimize language selection for agent performance rather than developer satisfaction.
Signals
- Engineering discussions shift from "what do our engineers prefer?" to "which language does our agent pipeline produce the best results in?"
- Teams adopt languages with large training corpora and consistent idioms even when engineers find them verbose or unpleasant to write by hand.
- New hire preferences carry less weight in stack decisions.
Counter-evidence
Teams with strong existing investments in a language and large existing codebases face significant switching costs. Agent performance gaps between mainstream languages are narrowing rapidly, which may make this criterion less decisive over time as frontier models become equally capable across most mainstream languages. Developer familiarity still matters for reviewing and correcting agent output; a language nobody on the team can read remains a risk.
Cross-references
- As AI handles coding reliably, most software bugs shift from implementation errors to specification errors: the same shift that moves bugs from implementation to specification also moves selection criteria from ergonomics to agent-legibility.
- When agents write code, the decisive technology selection criterion is whether failure modes are legible to the agent: the companion claim about which specific technical properties become decisive when agents are the primary authors.
- Coding agents erased the cost advantage of cross-platform mobile frameworks: a parallel case where agent capability reversed an earlier ergonomic tradeoff in mobile engineering.