Scale and speed: why AI ghostwriting changes the sociology of authorship

AI makes drafting and re-drafting almost instantaneous. That sheer velocity changes who does the work, where effort accumulates, and how organisations should manage quality, provenance and long-term cost.

Why speed matters: more than faster drafts

When we say an AI can draft, re-draft and iterate in seconds, that is not only a comment about convenience. It reframes the economics and sociology of writing. Where human labour imposes cadence—time for research, reflection, edits, and coordination—AI lowers the marginal cost of producing another draft to almost zero. That shift amplifies production and distributes the work differently across people, tools and institutions.

The core dynamics at play

Velocity creates volume. Organisations that adopt rapid generation begin to produce orders of magnitude more text, code and assets. Unilever’s Closeup example shows how an AI-powered model helped produce over 100 assets across multiple markets in three days, illustrating what this velocity looks like in the wild.
Speed shifts responsibility. As Mark Boccia points out, writing by AI is fast, but the work of maintaining, curating and fixing what’s produced remains human. That faster output can therefore be a loan against future maintenance effort: features, copy or snippets created quickly still need editing, testing and upkeep.
Iteration cycles change attention and workflows. Developer communities report that low latency and fast response times materially increase experimentation: people test more variations, iterate rapidly and are less likely to pause and reflect. That can accelerate discovery, but it can also amplify poor choices if quality gates aren’t inserted.

Practitioner guidance: manage the scale, preserve quality

Below are concrete practices for teams using AI to draft or ghostwrite content. They aim to keep human judgement central while letting speed do what it does best.

Design explicit review budgets. Treat AI drafts like produced artefacts that require a fixed review allocation. If generation time is cheap, a sensible policy is to reserve 20–40% of total task time for human review, editing and testing. This aligns with the idea that velocity creates future maintenance needs and avoids assuming generated output is final.
Use human-in-the-loop patterns for quality and context. Practical HitL workflows—AI drafts, human checks for accuracy and tone, then human signs off—work well in customer-facing contexts. Salesforce’s write-ups show that AI can summarise histories and draft replies quickly; the human step ensures the response fits policy and nuance before it goes live.
Instrument provenance and audit trails. Add metadata to generated content: model version, prompt template, timestamp, reviewer identity and rationale for changes. When production is ubiquitous, provenance helps you trace why a phrase or decision exists and who is accountable for its correctness.
Insert quality gates and sampling. You cannot manually check every AI output at scale. Instead implement probabilistic sampling plus targeted audits for high-risk classes of content. Automate linting, factual checks and brand-style enforcement where possible to catch common issues quickly.
Budget for technical debt and maintenance. When AI accelerates prototyping and content production, you must explicitly account for the follow-on costs. As noted by practitioners, what feels like superpower speed can become a loan against future engineering or editorial time if you don’t allocate resources for upkeep.
Control iteration speed where necessary. Fast iteration is valuable, but unbounded iteration can cause noise and fragmentation. Introduce rate limits for automated regenerations or require checkpoints after a set number of iterations—this preserves developer or editor focus and prevents endless churning.
Design for human attention. Reduce cognitive load by presenting AI suggestions in context, with clear confidence scores or provenance. Developers in the OpenAI community emphasise that responsiveness is only half the benefit—how you present options matters hugely for whether people can make quick, reliable decisions.

Organisational implications

At scale, AI-generated content changes roles more than it eliminates them. Organisations often move from creators to curators: people become editors, validators and maintainers. That shift requires training, role clarity and policy updates so teams know how to accept, reject or remediate generated outputs.

Summary

AI’s ability to produce content at speed is a sociotechnical phenomenon. It changes incentives, allocates labour differently and creates downstream costs if not managed. Use practical controls—review budgets, HitL processes, provenance, sampling and explicit maintenance allocation—to harness speed without letting scale erode quality or accountability.

Expanded from

This post expands on a point from Sociotechnical systems in software and AI: Ghostwriting has always existed — why does AI ghostwriting feel different?.