What we build

Before, an idea sparks from somewhere — another story, a lateral connection, something subconscious. It gets developed by your own research, talking it through with a colleague, or in the morning meeting with the editor.

The meeting still happens. What changes is how far the idea has travelled before you walk in. Pressure-test it with AI first and you arrive with something substantial rather than a hunch.

A meeting between people, and it stays that way.

Two places where AI can help. When an editor pushes back and asks you to confirm figures or sources before green-lighting, that loop closes faster — unless you are waiting on others. And with the right setup and everyone's consent, the meeting itself can be captured and transcribed, with the agreed changes drafted for approval before the room has emptied.

This is where AI shows significant edge — from hours of searching and skimming through a large quantity of material online, to finding the right pieces of the puzzle.

What AI cannot do is taste the wine, talk to the artist at an exhibition, or walk into the LHC and feel its grand scale. Those hours are yours. And they matter the most because they are human — experiences that shape your taste and perspective.

What changes is what happens around them. You talk the experience through, and AI fills the gaps: historic context, what others have written, the detail you missed on the night.

This step is new. Before, gathering and judging happened together — you assessed each source as you came across it, and the verification was distributed through the slow process. But you might get tired and lose focus. The solution was more coffee.

Now the gathering is faster, with one degree of verification baked in. Which means you must reallocate the time you saved to the more important work. Full verification becomes its own deliberate act — this is where more of the time goes now. But instead of the old one-directional search and find, it's a faster, iterative learning process, because you don't know what the AI will come back with. With the right setup, you are usually pleasantly surprised.

Both sides bring bias. Yours comes from experience — the Cognitive Bias Codex maps 188 documented varieties. The model's comes from its training data and the context it has been given. A good partnership corrects for both. A bad one compounds them.

Governed by someone else's diary. But this is usually the meat of the pie.

This is uniquely human — knowing when to push, reading the facial expression before an answer arrives, and being present.

As more organisations adopt AI, these human capabilities will be the differentiator between quality content and AI slop. An intern writing with AI and a reporter with decades of experience will leave the same interview with entirely different stories.

In most cases you don't need to transcribe the whole interview — you mark the audio and extract the quotes you need. But if the editor asks for a full transcript, it could take hours depending on the interview.

Now the whole thing is transcribed before you've left the building, and you ask for a quote conversationally rather than scrubbing audio.

Like the research stage, AI saves you time in one place — so reallocate some of it to verification. Double-check names, titles and key facts. Just like an audio recorder in a noisy environment, AI can mishear too. This is no different from before: use your judgement, and when in doubt, double-check with the source.

Some writers structure the whole picture in their heads as each piece of the puzzle arrives; some do so once they've gathered enough material. Others move subheadings around in a Word document.

Sometimes the piece writes itself. Other times the hardest part is the blank page, or the empty space between subheadings.

Now, instead of starting from a blank canvas, AI structures all the pieces you have available. No procrastination, and no more rearranging subheadings as the clock ticks away.

The real change isn't only efficiency. It's momentum and a clear direction. With the right setup it's like talking it through with a friend or an editor during the coffee break — except there's almost no lag between a new idea and testing it out.

The first draft was usually the first big hurdle — a feedback loop between the piece and yourself.

With your AI partner, the structure and the first draft blend together. After one conversation you are already in the editing phase before you know it.

Writing is rewriting. At its best this was writer, editor, sub-editors and picture desk, each with their own filters — which is why a piece could pass through five or six people and come out transformed. For better or worse, that feedback loop could be smooth or arduous depending on the team and the culture.

With the right setup: you share the piece, everyone contributes in the same chat, and everything gets incorporated with your approval. No more keeping track of long email chains, multiple Word documents back and forth, or debating who should be in the CMS right now.

In SMEs, the editorial team is sometimes one junior staff member wearing every hat at once. AI restores some of what's missing — only if it's used with human judgement rather than replacing it.

Yes, we never put the word 'final' on anything without jinxing it. But you'll be surprised how quickly you can get to a place where you can use that word confidently. Sometimes it's a false sense of confidence — which is exactly why we keep stressing the importance of showing up and maintaining your agency.

If you have access to more powerful models, going through this process with a more advanced one might push the piece further still. Even if that opens another round of feedback, it will still be faster than before, given the right setup.

You or the editor read through everything again, raise anything outstanding, and double or triple-check depending on the piece and its sensitivity. Use alternative sources or methods for any fact-checking at this stage. Asking the same AI that drafted the piece to check its own work is not verification — it's asking for reassurance.

AI empowers you to 'write' less. It does not let you take less responsibility. Show up, read it properly, and check everything thoroughly, every single time. Don't be lazy.

You'd think we'd have this all automated by now. Unfortunately most content management systems are old, and every platform wants something different.

You could automate it, but to be reliable it would need a proper agentic architecture with multiple layers of guardrails — and human judgement may still be required, depending on your distribution platforms.

Between the time it takes to make it reliable, the token cost and the supervision, we don't recommend it. But it's doable.