About

Field notes on building AI systems that fail honestly, written from the middle of building one.

I am Cam Sweatman. I work in Pensacola, Florida, and I build AMCO — a campaign orchestrator that takes a business brief and produces the material a marketing campaign is made of: strategy, channel copy, content calendars, visual direction, email sequences, and the import files a client loads into whatever tools they already pay for.

It is not a demo. It generates output that goes to real businesses, which is the only reason any of this writing is worth reading. A system that never ships produces opinions. A system that ships produces bugs, and bugs are where the actual knowledge is.

What I write about

Mostly one thing, from different angles: a language model produces confident, well-formed, plausible output whether or not it has any basis for it. Every serious defect I have shipped comes back to trusting the shape of an answer instead of its substance — a citation that was never retrieved, a test that asserted my assumption back to me, a format designed against a platform that works differently than I imagined.

So the recurring subjects are evidence discipline, verification that actually catches things, and the gap between "the tests pass" and "it works."

How I try to work

• Run it against the real thing before believing it. Mocked tests verify plumbing. They cannot tell you whether the response you handled correctly will ever occur.
• Prefer the honest gap to the plausible filler. An empty field, a stated "unverified," a missing merge tag — all better than a confident value that fails in front of somebody else's customer.
• Read the vendor's own documentation. Not a summary of it. Several summaries I checked were confident and wrong.
• Write the failure down. Most of these posts exist because I wanted the next person hitting the same wall — often me, months later — to hit it once instead of twice.

Contact

You can reach me at nosweatautomotive@gmail.com. I also run No Sweat Automotive, a family collision repair shop in Pensacola that has been open since 1988 — which is a different job entirely, and the reason I care about tools that work for small businesses rather than tools that demo well.