AI Amplifies Experience: Good Decisions Compound
After 20 years and 6,000 days of coding, ThePrimeagen feared AI devalued his skills—but realized experience prevents catastrophic choices like forking Chromium, making right decisions exponentially more valuable as code becomes cheap.
Experience Outweighs Raw Code Output in AI Era
ThePrimeagen shares his crisis after 20 years of intense programming—6,000 days honing skills across Go, JavaScript, C, Rust, Zig, and 14 years mastering Vim motions. He once cranked 15,000 lines of code weekly without aids, building tools like Pacman perturbers to test JS-to-HDMI latency. Yet AI hype, with figures like Gary Tan claiming 37,000 LOC daily, sparked doubt: Is value just 'taste' (pretty UIs) or lines of code? No—AI drops code cost to near-zero, skyrocketing the premium on right code. Bad choices compound exponentially (2^n problem, not linear), like forking Chromium, which takes 6 hours to compile on 2023-2025 hardware. AI's 'internal monologues' lead here without human guidance, ignoring alternatives like stdin/stdout over web servers or when to denormalize Boyce-Codd databases.
His 'two-by-four moments' crystallized this. First, walking into a literal 2x4 board while obsessing over a job switch to WebFilings (now Workiva), which shaped his engineering rigor—he fabricated fears until reality hit. Second, spotting an AI suggesting Chromium forks for a 'trivial' web issue, revealing how experience avoids such traps. These decisions, earned through 6-hour debug sessions over 5-minute manual reads, multiply AI's power: generate parsers instantly, but maintainability demands knowing why.
Gain 'Toxic Productivity' Through Failure, Not Speed
Forget constant shippable outputs—true productivity is 'toxic' in chasing endless experience over instant wins. ThePrimeagen's biggest lessons came from repeated failures, not friendly manuals. For juniors: AI lowers barriers (e.g., 'write an OCaml parser' works), but don't fear irrelevance. Even if you never type again, decision-making endures—why this data format? Why serialize? Pizza analogy: Infinite toppings (cheap code) yield garbage; restraint from experience crafts mastery.
He counters AGI panic (daily Twitter 'achievements') and predictions like Dario Amodei's 'coding gone in 12 months.' Good engineering judgment can't vanish; it guides AI multipliers. Ending with DHH: 'It's fun to be competent.' Ship with Neovim or VS Code themes via AI, but competence compounds over hype.