# AI Coding Agents Excel at Mechanical Bulk Work ## Claim AI coding assistance earns its keep on repetitive, mechanical, high-volume tasks — bulk refactors, boilerplate generation — not on the deliberate, judgment-driven work of matching an existing design or crafting something genuinely new. <!-- atomicity: 1 idea ✓ --> ## Scope & conditions - Holds for tasks with clear before/after transformation rules: CSS extraction, format conversion, boilerplate scaffolding. - Breaks down as a heuristic once the "bulk work" itself requires novel judgment calls — the same livestream that surfaces this claim also shows the agent needing multiple human-driven correction rounds to match an existing page's width and styling, which is visual-consistency judgment, not mechanical transformation. - Doesn't claim AI agents can't do creative work at all — only that the reliable, low-supervision win is on the mechanical end. ## Evidence Responding to a chat comment questioning whether LLMs actually save time, the streamer distinguished mechanical busy-work (bulk CSS refactors, boilerplate generation for a from-scratch video-animation tool) — where AI assistance was "super powerful" — from the "craftsmanship" of deliberately hand-writing something new, where he didn't experience the same gain. The same stream demonstrated the boundary in practice: the AI agent's first-pass single-post template broke navigation, didn't match the homepage's width, and styled a comments section with low-contrast (dark-on-dark) text the host had to catch and remove himself as an accessibility failure. > [!quote] > "The stuff that I find it find LOM super super powerful for is like all that busy work like oh my god we got to like let's now abstract out all our CSS." ## Links - [[3. Resources/Permanent Notes/Automated Guard Rails Should Match Mechanical Verifiability]] — because extends; both notes draw the same line between mechanically-checkable work (safe to automate) and judgment calls (still need a human), from two independent sources in the same batch. - [[3. Resources/Permanent Notes/Developer-Tool Adoption Tracks Cognitive-Load Reduction Not Feature Count]] — because applies-in; the tasks AI agents actually reduce cognitive load on are the mechanical ones, consistent with this claim. - [[3. Resources/MOCs/AI & the Future of Knowledge Work MOC]] — because applies-in the vault's ongoing thread on where AI assistance is and isn't trustworthy. ## Tests / falsifiers - If a documented case shows an AI agent independently producing a novel, judgment-heavy design match (not just mechanical transformation) without multiple human correction rounds, that's a direct counter-instance. - Track correction-round counts across a sample of AI-agent-generated templates, split into "layout/visual judgment" vs. "functional/mechanical" categories — if the two categories don't differ in correction rounds needed, this claim's distinction doesn't hold up. ## Maintenance - Last tended: 2026-08-16 - Next review: 2026-11-16 ## Source - [[3. Resources/Literature Notes/Using WordPress Studio Code to Build a Hockey Team Site]] — the Studio Code agent's demonstrated strengths and correction-needing gaps in this same session.