Everyone writes their own
Ten people ask the same AI for the same report in ten different ways and get ten different standards back. Nothing is shared, nothing improves and nobody knows which version worked.
← Home / Why Onuris
Why Onuris
Most firms already pay for a capable AI. Most of them get ordinary results from it. The model is rarely the problem. The brief is the problem: a single line typed in a hurry, different for every person and every attempt. Onuris sells the brief, written once, properly, for every department.
Fig. 01 — The problem
The evidence is starting to show it. In a 2026 Harvard working paper, Fiona Chen and James Stratton studied about 300 million work events at 718 software companies. After teams adopted AI coding agents, code written rose about 30%, but the time to review and merge a change rose 49%, and completed work did not rise in step. AI made writing faster. Checking became the bottleneck.
Ten people ask the same AI for the same report in ten different ways and get ten different standards back. Nothing is shared, nothing improves and nobody knows which version worked.
A one-line request leaves the model to guess the audience, the format and the standard. It guesses fluently and does not tell you that it guessed. Errors then read just as confidently as facts.
Without stated sources, shown calculations or a rule for when to stop, a reviewer has to redo the work to trust it. At that point the time saved is gone.
Fig. 02 — What you are buying
Each pack is sixty briefs for one department, built from one shared engine. Every brief carries the same discipline, so the standard holds from the first prompt to the sixtieth.
The structure follows published findings on how models reason, and the guidance that Anthropic and OpenAI publish for their own models. Each choice is explained, with sources, on the method page.
Every sentence follows the writing rules of ASD-STE100, the controlled English that aerospace engineers use for maintenance manuals. One instruction per sentence, nothing vague, nothing to misread.
Figures are tied to named inputs, calculations are shown, assumptions are stated and missing information is marked as missing. Solid prompts finish with a review list of what a person must check first.
Each department’s briefs name the UK rules that apply to its work, from FRS 102 to the Equality Act and UK GDPR. The AI applies them where relevant, marks what a qualified person must check, and stops to hand a decision to a person at the thresholds you set.
Every brief is formatted for the model you use: Claude, ChatGPT, Gemini, Copilot or an open model. Each one has a slot for an example of your own best work, so the AI learns your house style.
Four tiers for each task, from a single precise instruction to a staged workflow with self-critique. Use the light tier for routine work and the full brief when the stakes are high.
Fig. 03 — Plain, controlled English
ASD-STE100 Simplified Technical English was created so that a technician, often working in a second language, reads an instruction once and does it correctly. An AI benefits for the same reasons. Short sentences carry one instruction each. The condition comes before the action. Words such as may, should and could, which leave the reader to decide what was meant, are not used. A checker runs over every pack, and a pack that breaks the rules is not released.
Before
Stop and flag to the Finance Director immediately if cash falls below zero in any week, if any action involves deferring PAYE, VAT or other statutory payments, or if the forecast suggests the company may be unable to pay its debts as they fall due, which calls for professional advice on directors’ duties.
After
Stop and flag to the Finance Director immediately if cash falls below zero in any week. Do the same if any action involves deferring PAYE, VAT or other statutory payments. Do the same if the forecast shows a risk that the company cannot pay its debts as they fall due. In that case, say that the directors need professional advice on their duties.
Every word does a job
An Onuris brief is longer than a one-line request, because it carries the context, rules, checks and format that a one-line request leaves the AI to guess. What it does not carry is filler. Each sentence gives one instruction, and each brief fixes the shape of the answer, so the AI does not pad its reply. The aim is usable work on the first answer, not the fourth, and fewer rounds of back-and-forth to get there.
Source for Fig. 01: Chen, F. and Stratton, J. (2026), Artificial Intelligence in the Firm: Bottlenecks in Software Production, Harvard University working paper, as summarised by InfoQ. The study measured software teams. Onuris is not affiliated with the authors.
We apply the ASD-STE100 writing rules. We do not restrict the briefs to its controlled dictionary, because finance, legal and technical terms are what the work is about. ASD-STE100 is a trademark of ASD, the AeroSpace, Security and Defence Industries Association of Europe. Onuris is not affiliated with ASD.
Fig. 04 — The arithmetic
Writing sixty good briefs for one department is real work. Someone who knows the department has to write them, test them on your AI, fix them, format them for each model and keep them current as the guidance changes. A single pack costs about £14 per prompt, once, with all four tiers and every AI format included.
Doing it yourself
With Onuris
Your numbers
Fig. 05 — Straight answers