*Note: This was from Fable for a linked In post. I gave it the puzzle solver the 2 examples and it build the rest Including the tag line. I thought it was interesting from both Fable direction, me, and AI. So sharing. At my core, I'm a puzzle solver. Give me a problem with a clear "this shouldn't be happening" and I'll hunt it down. Two favorites from my analyst work: → A device in the field was dropping 0.3% of its API calls. Every dashboard said "fine." I isolated the device traffic onto a clean network, ruled out layers one at a time, and traced it to the IP's firewall silently scanning credentials. Days of work for a fraction of a percent — a fraction of a percent of thousands of calls is real failures for real customers. → Historical revenue reports kept shifting after the books were closed. The data model checked out. The queries checked out. The cause: manually backdated payments, invisible unless you knew exactly where in the data to look. Both cases were solved the same way: refuse to accept "probably fine," follow the evidence one layer down, repeat. That attitude now shapes how I build AI agents. Every automation I ship is auditable end to end. My GitHub triage agent logs every decision and escalates to a human the moment it can't parse a situation. My negotiation engine validates every LLM output against a deterministic rules engine — 638 tests pin the behavior. When something misbehaves, I can tell you where, why, and exactly what the model saw. Agents will go wrong. I build them so you can find out why. If you can't trace it, you can't trust it.

I turned my remarkable paper pro (an electronic paper tablet. One of my favorite) into Tom Riddle's journal, but more importantly it's a new interactive surface. I can simply write in normal text and it gets sent directly to the model and the model response directly on the tablet itself in normal handwriting! Pretty cool! Going to see about connecting it to a couple icms! I used Maximes main repo to get it running !