What I’ve learned from building AI Agents in OutSystems (so far)

A developer shares practical lessons from building three AI agents in OutSystems: a spellchecker, a translation agent, and a compliance policy checker. Key takeaways include token usage being far higher than expected with real-world data (not just demo inputs), response times exceeding two minutes causing user frustration, the value of keeping agents small and single-purpose for easier debugging and lower costs, and the importance of having budget conversations with stakeholders before go-live rather than after complaints arise.