Agents In UseSpecific projects, practical patterns and mistakes worth learning from. Each collection explains why the cases belong together, with source links and limits in every case.
A goat that eats your notes, a browser jam room and a printer that prints its own photograph. Six specific builds with something to try.
Start with one interaction you can demonstrate. Keep the first version small enough to inspect yourself.
Robots, microscopes, projector firmware and home lighting: six accounts where a successful command needed a check against actual equipment.
Give the agent observable feedback, retain control of physical constraints and check the result on the actual equipment.
Six concrete workflows for invoices, meeting actions, applications, job matching and receipt entry, including the review work their authors retained.
Define folders, fields and recipients before connecting tools. Reconcile outputs with the original records.
Missed fares, wrong timezones, conflicting cart totals and repeated coding mistakes: six cases with specific checks a reader can borrow.
Check the original evidence and the intended outcome separately from the agent’s completion message.
A delayed connection, toddler timing, childcare and offline travel notes. Six accounts of agents working around the details generic itineraries miss.
Put the inconvenient constraints in the brief. Ask for separate confirmation of reservations, companions, local times and unresolved steps.
These are editorial selections of existing sourced accounts. Reported outcomes remain the authors’ claims unless a case explicitly says otherwise. Browse the full directory →