Issue 013: Writing, Communication and AI Co-Authors

AI co-authorship is not edge behaviour in 2026. It is the default. KPMG's commitment to Claude across 276,000 employees, the first AI Natives cohort arriving at desks with frontier models already baked into their workflow, Google's pivot to AI-summarised search, and Pope Leo XIV's encyclical calling for robust external regulation are four stones on the same scale. The first draft is increasingly a machine output. The edit is human. The governance and disclosure questions follow behind, unsettled.

Three positions on disclosure are now in play. The maximalist says disclose any AI use, every time, in footnotes or front matter. The minimalist says disclose only when AI generated most of the substance. The pragmatic position says disclose when the reader's understanding of source materially changes. Each has trade-offs. Maximalism builds trust but flattens ordinary editing into a confession. Minimalism is workable but leaves too much room for convenient silence. The pragmatic position is honest but requires judgement on every piece, which is where most teams currently default to nothing.

The disclosure debate is a proxy for a deeper one. When the cursor does not type the first draft, what does authorship mean? The answer lies one step further back, in voice.

The most common failure in AI writing in 2026 is not the colourful hallucination. It is the quiet flattening of voice. Give most models a blank page and you get competent, structured, balanced, slightly bland prose. For many users that feels like an upgrade, because their own first draft never sounded that polished. The trade looks attractive in the moment, a small quality lift delivered fast.

For professionals whose writing is part of their reputation, that trade is dangerous. Consultants, executives, researchers, founders, none of them are followed for generic competence. Readers follow them for rhythm, vocabulary, judgement, point of view. The voice is the asset. Surrendering it for a small efficiency gain trades a long-term reputational asset for a short-term throughput win. Factual accuracy still matters; bad facts cannot ship. The point here is that voice flattening is the under-discussed risk.

The counter-argument is fair. Some writing should be plain. A compliance note, a help article, an internal process update does not need literary personality. In those cases, clarity beats style. But reputation-building writing is different, and the fix for it is mechanical rather than aesthetic. Three to five lines of your own writing, pasted into the prompt before the draft request, gives the model the steering wheel it needs to produce a draft that sounds like you on a good day. Five minutes of setup per session. The voice survives. The reader never has to do the work of detecting which paragraph the machine wrote, because the voice is consistent throughout.

The recommendation: build a small voice-sample library this week. Three to five lines per context, in a notes file you can paste from. The next time you reach for AI to draft something that carries your reputation, lead with the samples, then add the request. The Prompt Pocket below gives you the exact text.

Lex is a purpose-built writing environment where AI is invoked through specific commands rather than always-on suggestions. You draft in a calm editor, then call /rewrite, /continue, or paragraph-level prompts when you want help. Lex can be steered with samples of your own writing rather than a generic "professional" preset, which pulls the output closer to your register than a default chat tool would manage. Document-level edits, title generation, and continuation prompts all work without leaving the page.

The on-demand model rewards intent. Lex stays out of the way until you ask for help, which suits deep work and longer pieces. Pasting voice samples into a command produces drafts that sit closer to your register than the corporate-default register that always-on tools tend toward. The editor itself is clean, fast, and built for writers rather than for generic productivity.

Collaboration is lighter than Google Docs or Notion. Real-time multi-cursor team work is not there yet, and the commenting and track-changes parity that mature enterprise editors offer is not present. The on-demand model also rewards users who know what they want. Beginners who expect inline help every keystroke may prefer an always-on tool. For solo professional writing, Lex is excellent. For team workflows, it sits alongside Word or Docs rather than replacing them.

Some links are affiliate. Zymbos AI may earn a small commission at no extra cost to you.

Try Lex →

The professionals who keep their voice in 2026 will not be the ones who refuse AI. They will be the ones who use it carefully, with samples, with intent, and with an edit pass that knows the difference between competent and recognisable.

Three predictions for the next two quarters. First, at least one major professional services firm beyond the Big Four will announce a frontier-model sole-source commitment before September, because KPMG's announcement has set the procurement precedent and the laggards now look unstaffed for the change. Second, "AI fluency" will be replaced in serious hiring conversations with "judgement under AI," because the AI Natives cohort will rapidly expose the limits of fluency alone. Third, the first reputable industry survey on voice-preservation tooling will appear inside six months, because the gap between AI adoption and AI-assisted brand consistency is now visible to anyone paying attention.

The question for your week: pick one piece of writing you owe someone, your most reputation-loaded piece, and try the Prompt Pocket above on it. If the output sounds more like you than your usual AI draft, hit reply and tell me. I read every response.

If the output still sounds like a polished stranger, the samples were not specific enough. Try again with three lines that sound the most like you on your best day. Voice samples are the steering wheel; specificity is the grip.