Writing is no longer proof of work

Writing in tech has long functioned as "proof of work" - not mainly for communication but as evidence of compliance, effort, or learning. Examples include checkbox templates like risk assessments that exist to show process compliance rather than assess risk, long consultant reports whose bulk serves as proof that money was spent, and hierarchical status rollups that force managers to build minimal mental models and thereby demonstrate understanding. Organizations treated such writing like a cryptographic proof-of-work, relying on trust plus spot checks to prevent fraud because convincing text was historically expensive to produce.

Large language models have broken that guarantee: high-quality, tailored reports, templates, and status updates can now be produced cheaply, undoing writing’s evidentiary value. Organizations face an overhang where processes remain but are trivially exploitable. Responses will include requiring live presentations or video, banning AI-generated text (difficult to enforce), or using AI to audit engineers’ work, yet no clean solutions exist. Practically, the result is a shift away from producing writing purely as evidence: spend time only on documents intended to communicate to humans, not to satisfy performative checkboxes, while recognizing a new class of documents aimed at machine consumption.