LLMs Drive Performative Productivity, Question Real Gains

Josh Collinsworth published a blog post on June 5, 2026 titled "LLMs and performative productivity," describing his firsthand experience using AI agents to accelerate day to day developer tasks. Per Collinsworth, the agents enabled faster onboarding into unfamiliar codebases, rapid refactors including a long-delayed Nuxt upgrade completed in about an hour, additional feature work, scaffolding of greenfield projects, faster test creation, and numerous bug fixes. Collinsworth reports that many of those outputs delivered little user-visible value, that he did not develop sufficient system context to defend some pull requests, and that several new projects or features were later abandoned.

Companies and teams that adopt LLMs and agent tooling often see immediate gains in throughput for discrete tasks. Industry-pattern observations note that higher output can coexist with lower tacit knowledge transfer when the tools do heavy lifting. For practitioners, this can manifest as faster syntactic fixes, generated tests of unclear coverage, and features that pass CI but lack integration into product workflows.

Observed patterns in similar transitions show a recurring tension between activity metrics and impact metrics. Organizations that measure productivity by commits, PR velocity, or test counts may observe short-term improvements that do not map to customer usage or system comprehension. Metrics and review practices that prioritized surface-level indicators can therefore produce a performative impression of productivity.

For teams adopting agents, observers will likely watch adoption signals such as change in code review behavior, incidence of reverted or flaky PRs, usage metrics for newly shipped features, and retention of institutional knowledge during onboarding. Industry context: dashboards that combine usage telemetry with developer-output measures tend to reveal whether increased activity translates into product value.