How the lab works
Every briefing here is researched, written and independently fact-checked by AI agents — then published with its workings on show: the pipeline that built it, the sources, the scores, the fact-check. We do this so you can judge the quality yourself, and so you can see what well-skilled agents can genuinely do for a business like yours.
Five agents, one briefing
Research + draft
An agent works to a house brief — what changed in AI, who feels it, where the workflow drag is — and drafts the briefing from primary sources.
Humanise + sharpen
A second pass strips the AI tells, finds the real hook, takes a position, and makes it read like a sharp operator wrote it — not a machine.
Score, independently
A separate scoring agent grades it on soul (is it worth reading?) and rigour (is it sourced and accurate?). Anything weak gets rewritten and re-scored.
Fact-check, independently
A different, web-enabled agent (Codex) checks every statistic, named case and regulator claim against its actual source — and flags anything unsupported.
Correct, then publish
Flagged items are fixed before anything goes live. The briefing ships with its scores and fact-check shown, so you can judge it yourself.
The separation matters: the agent that scores the work is not the agent that wrote it, and the agent that fact-checks it is different again, with live web access. No single model gets to grade its own output.
What this shows an SME about AI
Most SMEs are told AI is either magic or a threat. Neither is useful. What you actually need to know is narrower and more practical: what can agents reliably do, where must a human stay in control, and how do you tell good output from confident nonsense?
These briefings are a worked answer. The agents did the high-volume, checkable work — research, drafting, sourcing, cross-checking — fast and to a standard you can inspect. What they did not do is make the judgement call. That division — machine does the drag, human owns the decision — is exactly the pattern we think most traditional businesses should copy. It is also what the regulators expect.
Judge it for yourself
At the foot of every briefing is a "How this was made" panel. Here is how to read it — the same four questions you should ask of any AI output, yours included.
Sources
Every factual claim is numbered and links to a real source — government data, regulators, named benchmarks. Count them. Follow them.
Independent fact-check
A separate agent, not the one that wrote it, verifies the claims against the web. That separation is the point — it is not marking its own homework.
Soul + rigour scores
Soul = is it actually worth your time. Rigour = is it sourced and accurate. Both are shown, out of 10, so quality is visible, not asserted.
Human control
AI does the research and the drafting. It does not make the call that needs judgement. The same rule we tell SMEs to apply to their own work.
AI is the method, not the magic
We are not selling you AI. AI is how we research and produce this work to a high standard, quickly. The value to you is the insight and the evidence — where AI genuinely fits your workflows, and where it does not.
And we are deliberately public about the limits. An agent that drafts a briefing is doing checkable, sourceable work. An agent deciding whether to accept a client, approve a loan or sign off a compliance return is not — those need an accountable human, and the briefings say so. Showing the workings is how we keep that line honest.
See it in practice.
Read a briefing and scroll to the panel at the foot — or run a free scan to see where AI fits your own workflows.