# Cheap Token-Based Pricing Makes Wholesale Regeneration Beat Incremental Fixing ## Claim When AI generation is priced cheaply enough (token-based, near-zero marginal cost), it becomes rational to delete and wholesale regenerate an entire body of work from scratch instead of patching it piece by piece — inverting the usual assumption that incremental fixes are cheaper than starting over. <!-- atomicity: 1 idea ✓ --> ## Evidence Dan Pelato (Standex Electronics), WordCamp US 2026, on switching a DeepL-translated site to OpenAI after a tone-drift issue: "instead of going page by page and retroactively fixing each page when there's hundreds of pages, I said to Side, I said, 'Can we just delete it?... can we re-translate the whole site?'... It was less than 30 bucks for the whole... two sites." DeepL's character-based subscription pricing scaled expensively across 48 sites; OpenAI's token-based pricing ran under $1/day in production, making a full-site redo cheaper than the engineering time a page-by-page fix would have cost. ## Scope & conditions - Holds when the per-unit cost of AI generation is low enough that engineering/review time to identify and patch individual failures exceeds the cost of regenerating everything. - Breaks when regeneration itself carries hidden costs the source doesn't have to pay — e.g., hand-tuned content that would be lost, or a review burden that scales with total volume rather than with the number of changes (regenerating 500 pages still requires re-reviewing all 500, even if only 20 needed a fix). - The team still needed a human reviewer per language after wholesale regeneration — this claim is about the economics of *generation*, not a claim that regeneration removes the need for review (see the Human Review Is the Actual Authenticity Control note). ## Links - [[3. Resources/Permanent Notes/AI Amplifies Existing Bottlenecks Rather Than Creating Capability]] — because contrasts: that note warns cheap AI throughput amplifies existing bottlenecks without adding value; this note names a case where cheap throughput genuinely changed the optimal strategy (regenerate vs. patch), not just the speed of the old one. - [[3. Resources/Permanent Notes/Cheap Generation Raises the Total Maintenance Bill]] — because contrasts: that note is about cheap generation increasing long-run maintenance debt; this note is about cheap generation making a specific one-time regeneration decision economically rational — worth holding both in tension. - [[3. Resources/MOCs/AI & the Future of Knowledge Work MOC]] — because applies-in: a general economic pattern for any AI-assisted content pipeline (translation, code migration, bulk content refresh) priced per-token rather than per-subscription. <!-- atomicity: 1 idea ✓ --> - [[3. Resources/MOCs/AI Cost & Tokenomics MOC]] — applies-in: Joins the model pricing and cost-curve cluster.