Signal & noiseThe detector
Signal & NoiseThe Bullshit Detector
Positions on AI hype and transformation theatre, and a tool for translating corporate language.Short opinions on AI hype and transformation theatre. And a machine that turns jargon back into English.
Corporate Jargon Translator
Type a sentence from a meeting, a strategy deck or a press release.Paste the sentence that made you close your laptop a little too hard.
Or try one of these
Awaiting input
Enter a sentence. The translation appears here.Feed me a sentence. I’ll tell you what it actually means.
Jargon density—
Standing positions
What I believe, briefly.Things I will say in your steering committee.
An AI strategy needs a P&L line.An AI strategy without a P&L line is a mood board.
If a use case cannot name the cost it removes or the revenue it adds, it is research. Fund it as research, with a research budget and a stop date, or do not fund it.
Pilots need an exit date.Pilots are where good ideas go to retire.
A pilot without a scale-or-stop date becomes a permanent pilot. Set the date before kickoff, agree what “good enough to scale” means, and make the call on the day.
Count systems changed, not slides produced.Your transformation office has more slides than systems.
A progress report should list the processes that now run differently. If that list is short, the programme is mostly communications.
Governance works when teams experience it as a service.Governance is a product, not a committee.
Teams follow governance that makes their work faster: templates, pre-approved patterns, a register that answers questions. They route around governance that only adds meetings.
Data sovereignty is an architecture decision.Data sovereignty is not a legal footnote.
Where data lives, who can move it and how is decided in the design. Under PIPL, GDPR or the EU AI Act, discovering it during an audit costs far more than designing it in.
A lighthouse matters only if others copy it.A lighthouse nobody copies is an expensive lamp.
Suzhou mattered because thirteen more sites used it as their benchmark. Build the template at the same time as the showcase, or the showcase stays a one-off.
Simplify a process before automating it.Automating a mess gives you a faster mess.
Most automation business cases assume the current process is worth keeping. Remove the steps nobody needs first; the remaining automation is smaller, cheaper and easier to trust.
An agent needs a job description.An agent with admin rights and no job description is an incident waiting for a date.
Before an agent touches a live system, write down what it may decide, which tools it may use, when it must hand over to a person, and the one number it exists to move.
Measure agents per task, not per demo.A demo is not a deployment.
Track cost per completed task, error rate and how often a person had to step in. If those numbers do not exist yet, the agent has been shown, not deployed.
A vendor demonstration is not your environment.Their demo data always works. That’s why it’s theirs.
Ask for a proof of value on your data, inside your security constraints, before the contract. It takes weeks and saves years.