Vol 2, Issue 10 | Mar 11 2026
Same. Prompt,Different Model, Worse Results
Dear Reader,
New LLM models drop every few weeks. Features change between updates. The prompting advice that was best practice six months ago has been revised — sometimes by the vendors themselves. And if you’re like most people, the way you actually use AI today is basically the same as it was three months ago.
That’s not a criticism by any means. AI moves fast enough that keeping your setup current is itself a recurring task — one that almost nobody schedules.
Last Saturday I made time for it. I went through every skill in my Claude Code configuration — documented procedures for repeatable work, things like running an editorial checklist, deploying a website, processing the morning email. I use this setup every day. I decided to see what still matched reality.
Aaaand…I found duplicates doing the same job. Skills that had grown bloated enough to need splitting into smaller, focused pieces. A whole cluster of related skills that collapsed into half their number once I looked at what they actually did. Legacy formats that hadn’t kept pace with how the rest of the system had evolved. Files in the wrong place. etc. etc.
The worst finding was that the descriptions I’d written for each skill — the short summaries that help the system know when to use them — had drifted so far from reality that they weren’t doing their job at all. The system was still finding the right tool most of the time, but through lucky guesswork rather than anything I’d designed. The descriptions were effectively decorative.
All in all it took about an hour and a half of passive maintenance while catching up on season 1 of ‘Shrinking’. But all of that drift had accumulated in weeks, not years — on a system I use for hours every day.
What you’re missing
This week’s full issue digs into why your prompts don’t travel between platforms — the structural differences between how Claude, ChatGPT, and Gemini process your instructions, the vendor guides that almost nobody reads, and a five-step checklist for auditing your own setup in twenty minutes.
This is what Signal Over Noise looks like from the paid side: tested patterns, honest assessments, and the implementation detail that makes the difference between reading about AI and actually using it well.
Read the full issue — subscribe to Signal Over Noise →
€7/month or €69/year. No sponsors, no affiliate deals — just honest, tested AI implementation work.
Until next time,
Jim
€7.00 / month
Signal Over Noise
Skip the AI Hype.
A newsletter about using AI to actually run a business and build things — past the hype and the doom,… Read more