It Still Can't Do My Job: Four Years of Moving Goalposts (2022–2026)
The author reflects on four years (2022–2026) of AI development, arguing that despite rapid advancements, AI still cannot fully automate their professional job. They critique the shifting benchmarks and goalposts used to measure AI capability, maintaining that genuine human expertise and judgment remain irreplaceable.
Background
This essay is a running critique of claims that AI will soon replace knowledge workers, written by someone in a technical or creative field. The author tracks how AI boosters have repeatedly moved the goalposts since 2022: each time a benchmark is passed (e.g., passing the bar exam, coding a web app), they redefine "real work" as something harder. The piece argues that AI remains a narrow tool that requires constant human oversight and cannot reliably handle complex, ambiguous, or context-dependent tasks without breaking. It taps into a growing backlash against the hype cycle around large language models (LLMs) like ChatGPT, Gemini, and Claude — a sentiment particularly common among programmers, designers, writers, and others whose jobs were supposedly first in line for automation. Key background: since ChatGPT's launch in late 2022, a recurring pattern has been tech executives and pundits predicting mass job displacement within 1–3 years, followed by the actual experience of workers finding that AI tools produce plausible but often wrong results that take more effort to fix than to do from scratch.