Choose a regular script when the API is stable and inputs are structured; choose an AI agent like GPT-6.1 Terra when the work lives on changing screens, scanned PDFs or portals without an API. OpenAI has added fast computer use and visual reasoning to its API, so one model can read a screen, decide the next step, and click.
Terra is positioned as a general-purpose agent that drives browsers and desktop interfaces at usable speed. The pitch: no bespoke scraper per site, no RPA license per workstation - the model sees the UI and acts.
What it changes for small-business automation
For small shops, the rule stays the same: if a vendor exposes a clean API, a 50-line Python script wins on cost and reliability. Terra gets interesting the moment you hit a supplier portal with no API, a legacy ERP, or a stack of scanned invoices and tender files - the kind of input I usually feed into a tender analysis pipeline. You pay per token and per screenshot, so the business case is "this task used to need a human clicker", not "let's agent-ify everything".
Why it matters: computer-use agents are moving from showcase territory to API primitives, which means processes that previously needed a human in front of a screen now have a plausible automation path - at a cost you can actually estimate per task.
Source: OpenAI