I Remain a Skeptic
Skepticism over large language models is colliding with enthusiastic reports of personal productivity gains, as developers argue over whether AI tools meaningfully improve software quality or merely create the illusion of working faster. Many commenters warn that current gains look modest and uneven, raise concerns about skill atrophy, hallucinations, code slop, and the use of AI to homogenize and cheapen intellectual labor, while others cite concrete wins in debugging, refactoring, and enabling ambitious solo projects. Looming over the debate is the $1.5T poured into AI infrastructure, with some predicting transformative “AI employees” and others fearing a bubble that may burst before broad, durable benefits appear.
Perceived Productivity and Workflow Changes
- Many developers report feeling significantly faster: more code, quicker debugging, and willingness to tackle more ambitious projects (e.g., full app rewrites, shipping multiple apps in weeks).
- Others emphasize modest gains: perhaps 1.5–2x on suitable tasks, far from “10x” or “1000x” claims.
- Some argue there’s a big gap between subjective productivity and observable organizational or macro-level gains.
Measurement, Evidence, and Economic Impact
- Several commenters note empirical studies suggesting small (~10%) productivity gains, not orders of magnitude.
- Others counter that programmer productivity is hard to measure and that we’re still early; effects may be lagging.
- Some say industry has “almost nothing to show” at scale; others point to large sets of bugs found/fixed with AI.
Software Quality and Security
- LLMs are praised for debugging and security auditing: finding long-standing bugs and vulnerabilities, generating detailed repros and tests.
- Skeptics question net benefit: risk of new bugs, shallow understanding, and “vibe-coded” slop that’s hard to maintain.
- Some highlight that bug-finding improves theoretical security, even if issues weren’t yet exploited.
Labor, Skills, and Industrialization
- Strong concern that LLMs homogenize intellectual labor, making developers more interchangeable and weakening bargaining power.
- Others argue software was industrialized long ago; LLMs are just another tool.
- Many worry about skill atrophy, over-reliance, and a future where most programmers can’t work without an LLM.
Usage Patterns and Best Practices
- Productive users treat LLMs like junior employees: they delegate grunt work, review diffs carefully, keep changes small, and ensure personal understanding.
- Bad patterns: blindly accepting large PRs, using AI to bypass real comprehension, and flooding codebases/docs with verbose noise.
Ethics, Risk, and Investment Bubble
- Some highlight hallucinations as a permanent limitation, unacceptable in life-or-death contexts.
- Others focus on upstream harms: energy/ecological costs, data “theft,” and brain-deadening effects on users.
- There is debate over the $1.5T AI investment: some see inevitable transformative payoff; others predict a boom–bust cycle with stranded data centers and serious financial fallout.
New Possibilities and “Vibecoding”
- Several note a qualitative shift: people shipping apps or games they’d never have finished before, or staying productive despite injury/cognitive decline.
- Some argue this “vibecoding” itself is a valued hobby and consumer benefit, even if business impact is less clear.