The original title is "AI Agents and Apps With No API: The Real Engineering Tradeoffs"

Short summary

When APIs are missing or incomplete, AI agents can operate through UI layers — DOM, accessibility trees, screen pixels, or standard input — but this is a different engineering problem, not an easier one. The article compares six interaction methods (browser automation, accessibility interaction, OCR, keyboard/pointer, RPA, hybrid) and argues for picking the most structured permitted method per task. It also covers hard safety lines: never bypass CAPTCHAs or MFA, and require human checkpoints for consequential actions.

• •Six UI interaction methods compared: browser automation, accessibility tree, OCR, input, RPA, hybrid
• •Pick the most structured permitted method, not the broadest reach — determinism matters more than coverage
• •Hard safety lines: never defeat CAPTCHAs/MFA, always require human confirmation for irreversible actions

Generated with AI, which can make mistakes.