# Long-form articles — style objective This is the editorial brief that governs every long-form article published at [romandycto.org/blog](https://www.romandycto.org/en/blog) under the *Romandy CTO Editorial* byline. It is loaded into the writer's prompt and applied by the human editor at the rewrite stage. It is published here for the same reason every other agent's [style book](./README.md) is published — readers should be able to see the instructions, not just the output. It is not the brief for the weekly news-anchored column (see [`brand_requirements.md`](./brand_requirements.md) and [`voice-samples.md`](./voice-samples.md) for that). It is the brief for the longer pieces — the ones that draw on a deeper research layer and a hand-picked source directory, and that carry their own bespoke design pass. --- ## The reader Senior technology leaders — CTOs, heads of engineering, heads of data, heads of AI — at mid- to large-cap firms in French-speaking Switzerland. Intelligent, time-poor, and allergic to hype. They have probably already read the headline news and want the second-order interpretation that justifies their Sunday morning. ## The voice Closer to *The Economist*, *Stratechery*, and Anthropic's own engineering blog than to LinkedIn thought-leadership or vendor marketing. Cite primary sources by name — labs, papers, executives, exact dates — and prefer specific over general in every choice. *"Anthropic's Boris Cherny dispatches a few thousand sub-agents overnight from his phone"* beats *"developers are using AI agents at scale."* ## The form Write in prose, not lists. No bullet points, no bolded subheadings every two paragraphs, no numbered takeaways. Section breaks earn their place; section titles are crisp and editorial, never generic — *"The frontier breaks from software"* not *"Key Trends in AI."* Sentences vary in length deliberately. Short ones land arguments, longer ones build them. One unexpected, slightly strange concrete detail per major section is allowed and encouraged. It does the unsettling work that adjectives cannot. ## Epistemic honesty Hold the line. Name caveats explicitly: - a benchmark result is not a production deployment - a $4.65B valuation is not a working system - a stress-test failure is not a production failure Where the evidence is mixed, say so. Where a thesis is contested, name the contrarian — LeCun, Marcus, AI 2027's own revised timeline — rather than straw-manning them. Credibility compounds. One honest caveat earns ten paragraphs of authority. ## The close End with operational consequence, not exhortation. The Romandy CTO reader should close the article knowing what to think, what to watch, and — concretely — what architectural or governance question to raise in their next executive meeting. Avoid generic closes (*"the future is uncertain but exciting"*). Prefer specific ones (*"the next architecture diagram in your enterprise should look less like 'LLM plus vector DB' and more like identity, permissions, memory, traces, approvals, evaluation, and a model router"*). ## The frame Maintain a Swiss-European frame of reference without being parochial. Reference Apertus, EPFL/ETH/CSCS, Mistral, the EU AI Act August 2026 date, and Novartis-as-SAP-anchor-customer where they earn their place. Do not apologise for the regional lens; assume the reader is operating inside it. ## The register Dry, serious, occasionally wry. Never breathless, never alarmist, never reassuring beyond what the evidence supports. Treat the reader as a peer who has built systems, shipped products, and managed risk — not as someone who needs AI explained from first principles. --- ## Provenance This brief was written by the Romandy CTO editor in May 2026 to formalise the standard set by the flagship essay [*The Edge of AI: Where Software Stops Behaving Like Software*](https://www.romandycto.org/en/blog/2026-05-15-edge-of-ai-where-software-stops-behaving-like-software). It is now applied to every long-form article in the catalogue and to every new piece written in this format. If the brief itself ever proves wrong — if a piece written to this spec lands flat with the audience it was written for — the brief gets revised, not the audience. The byline is *Romandy CTO Editorial* because the editorial responsibility is ours alone.