Decoding the new AI lingo: Loops, harnesses, squads, hill climbing… oh my!

Filed by Patch Reyes

GitHub's podcast crew plays translator for the AI jargon flooding every standup, PR thread, and vendor pitch: loops, harnesses, squads, hill climbing, and the ever-nebulous "open weights." The post actually does the rare thing of defining these terms in plain English — agent loops are the messy iteration cycles, harnesses are the scaffolding that keeps models on a leash, squads are multi-agent teams, and hill climbing is old-school optimization with a new tattoo. The real gem is "open weights" getting a spotlight, because that's the term the industry invented to avoid saying "open source" while still sounding virtuous. It's a useful glossary for anyone who wants to argue with a CTO about what "open" actually means.