Built for us
We run WYN on what we build
Before we build anything for a client, we build it for ourselves and use it every day. This is our own content newsroom: how it works, where people stay in charge, and what we hold it to. The thinking, not the recipe.
01 · The newsroom
An editorial process where AI agents do the legwork and people make the calls
Our insights, social posts and newsletter come out of a newsroom we built for ourselves. AI agents each handle one step of the process, from finding the news to preparing the visuals. Every step is checked, and nothing goes out until a person has read it and said yes.
Find the stories that matter
Reads a short list of trusted sources every morning and follows each story back to where it started: the announcement, the report, the study. A newsletter is a pointer, never a source.
Guardrail: if the original can't be found, the story is dropped.
Choose what's worth saying
Picks the stories for each slot and sets the angle: the point of view we'll argue, who it's for and what a reader should do next.
Guardrail: every piece has to argue a position we actually hold.
Challenge the plan
A separate check asks the awkward questions before anything is made. Would someone forward this to their boss? Does it connect to what we do? Does it land for someone who doesn't follow AI news?
Guardrail: weak ideas are replaced, not polished.
Write it
Drafts the article, the posts and the captions to our editorial guide: our voice, British English, a list of phrases we never use, and a clear structure from news to take to action.
Guardrail: every agent reviews its own work before handing it on, and says when it isn't good enough.
Check every claim
A separate step checks every figure, quote and claim against the original source, and flags anything misleading in how it's framed, even when each fact is true.
Guardrail: anything that can't be traced to a source is cut.
Make it look like us
Sources images from official press material, credits them, and lays out slides and short videos in our design system, then checks them at phone size.
Guardrail: no stock photos of people, no images lifted from other people's articles.
A person decides
Everything lands on a review desk with its sources, the fact-check and the strategy notes alongside it. A person approves it, asks for changes or rejects it.
Guardrail: nothing is published without approval. If an approved piece no longer passes our checks, it comes back to the desk.
Publish on schedule
Approved pieces go live on the website and are scheduled across our channels at the right time, with a summary each morning of what's going out.
Guardrail: it can only ship what was approved, exactly as it was approved.
What the agents do
- Read sources and trace stories back to the original
- Draft, check and lay out the work
- Review their own output and flag what's weak
- Keep a log of every step, visible on the review desk
What people do
- Decide what we stand for and what we cover
- Approve, change or reject every piece
- Plan the week and judge what's working
- Own every relationship: no agent speaks to a client
02 · One story, many formats
One idea, made once, published four ways
The expensive part of content is the thinking: finding the story, checking it and deciding what to say. So we do that once and let the system turn it into each format, with each one designed for where it'll be seen.
Illustration: a made-up example, not a real post.
Most agencies still rebuild the same client report by hand every month. Automate it, or keep doing it?
03 · Guardrails
Built to be trusted, not just to be fast
Speed is easy. What makes a system like this usable in a real business is everything around the AI: the checks, the limits and the people.
Quality
Every step reviews its own work against a written standard. A separate check challenges each idea and each finished piece, and sends back what isn't good enough. "Correct but dull" counts as a fail.
Accuracy
Every claim is traced to its original source and checked by a separate step. Anything that can't be verified is cut, and nothing is ever invented to fill a gap.
Security
Agents never hold passwords or log in as anyone. They can't publish on their own: only what a person approved goes out. Access is limited to what each step needs, and everything is logged.
Cost
Each step uses the smallest model that does the job well. Work only moves up to a more capable model when a check fails or a task is unusually hard, and that switch is recorded.
04 · For your team
The same principles, applied to your work
We build systems like this for agencies and marketing teams, shaped around their work and their brand: small steps, people deciding where it matters, and checks you can see. Three places it tends to pay off first:
Client reporting
Data pulled together and turned into a first draft of the analysis, so your team spends its time on what it means.
Content production
A brief turned into formats for each channel, in your brand's voice, with your team approving every piece.
Research and briefing
Sources read, summarised and traced back, ready for a strategist to sharpen into a point of view.
Let's work out what you need.
Book a free 20-minute call. We'll tell you honestly whether we can help.
Book an intro call