AI coding models now match 98% of software engineers, prompts a shift to fewer but much more productive workers

The article is written as a first-person account by Nikolai Tschacher, founder of the IP API service ipapi.is, reflecting on how his own software work has transformed since adopting AI tools in November 2022. He traces the trajectory of AI capability: early ChatGPT (late 2022) could write creative poetry and humor; by 2023–2024 it became genuinely useful for partial code assistance; and by November 2025, models like Opus 4.5 reached a threshold where they produce code as reliable as an experienced engineer, making human review unnecessary. The concrete example of his Node.js-to-Go rewrite anchors this claim: what would have required a software consultancy's six-month, $50,000 project in 2021 now costs $500 in AI tokens and two hours of developer time. The author observes that this does not eliminate senior engineers but amplifies their productivity — they become orchestrators of AI systems rather than code writers. However, he notes an apparent slowdown in AI model releases: Mythos/Fable (February 2026) remains competitive seven months later in August 2026, possibly because the US government evaluated it and found hundreds of zero-day vulnerabilities before its release, suggesting that leading AI labs may be proceeding more cautiously. The article hints that if this cautious approach continues, open-source and international competitors may eventually catch up, commoditizing AI; if it does not, the next frontier could be recursive self-improvement and AGI, raising containment questions.