Aviso's 7-Question Test for Telling a Real AI Agent From a Demo

Aviso published a buyer’s framework for a problem it says is rampant in the current agentic-AI sales market: vendors conflating “generates a recommendation” with “actually does the work.” Its seven-question test — Trigger, Context, Reasoning, Action, Coordination, Completion, Control — is built to separate the two. The clearest line in the piece: a conventional system might flag that deal engagement has declined and leave a human to interpret it; a real agent runs the full loop itself — detect, understand, reason, decide, act, coordinate, verify — without waiting for someone to close it out.

It’s a self-interested framework — Aviso is grading its own architecture against its own test — but the seven categories map onto real, independently documented failure points. Salesforce’s own writeup on scaling past a single agent names the two failure modes directly: a monolithic agent that degrades as more domains get bolted on, eventually misrouting requests in ways that take engineers reading through thousand-line instruction blocks to debug, or a set of narrow specialist agents that’s more reliable individually but fragments the user experience and complicates cross-agent handoffs — precisely Aviso’s Coordination and Completion tests, from a different vendor with no reason to flatter Aviso’s framing. Salesforce’s own bar for this: “95% reliability isn’t good enough” for anything high-stakes.

The Control test lands on ground Aviso has been working for a month already. Its earlier piece on the “RevOps Agent Owner” argued that agent accountability needs to be hardcoded into a specific job function — split between a Business Owner accountable for commercial outcomes and a Technical Owner accountable for infrastructure and compliance — rather than left as an informal, nobody’s-really-in-charge arrangement.

For a buyer sitting across the table from a vendor demo, the test is less about whether a specific answer is right and more about whether the vendor can survive being asked at all — a demo built on canned scenarios rarely holds up against Trigger or Coordination questions asked live.