Guide, 4 Nov 2026, read 8 Nov 2026
How to score an AI claim: the four-sentence test
Sentence one: what it does, as a process or as an answer
Read the page for a verb. 'Answers employee questions' is a chatbot; 'runs onboarding' is an agent. Shapes's home page names 'onboarding, comp cycle, employee questions' as what its agents handle; HiBob's agents page lists twelve named agents whose sample prompts are single actions ('Clock me in', 'Nudge everyone on my team who hasn't checked in their goal'); Humaans's MCP page walks through a comp cycle staged for approval. Score the verb, not the noun: a page with the word 'agentic' and no process named scores as a chatbot.
Sentence two: what happens before it goes live
The sentence to look for says who approves. Shapes: 'Ask for a change, they draft it, you apply it. Nothing goes live until you say so.' Rippling: 'Before it takes action, it gets your approval.' Factorial One: 'It asks before it acts.' Gusto MCP: 'Every write comes back to you for confirmation before anything moves.' BambooHR: prepared work 'still go[es] through the appropriate person before they're applied or sent'. A page that describes autonomy without that sentence, Humaans's 'autonomously runs workflows' for example, scores a point lower on agents here even when the audit log is good.
Sentences three and four: the log and the switch
An audit log that separates what the agent did from what people did is the control an auditor will ask for; BambooHR's Audit Trail records whether a person, Bamboo AI or both made a change, Shapes's audit log of 24 Oct 2026 lists sign-ins, role changes and account activity, Workday's Agent System of Record registers every agent. The switch is the sentence most pages lack: Shapes added separate switches for AI chat, AI-built Vibes and MCP access on 2 Nov 2026, Rippling says admins can enable or disable AI features, BambooHR has a setting under Account Settings. Of the fourteen platforms, three state a switch on the pages read.
The cap and the second source
A claim on a marketing page is a vendor claim, and the Index caps a cell that rests on one at 7 until a second source confirms it: a documentation page, a dated release note, a help-center article or a customer statement. Shapes's agents score 9 because the agents page is backed by dated release notes (the directory agent of 24 Oct 2026, the AI switches of 2 Nov 2026); its time to value scores 8 rather than 9 because 'days, not weeks' is a home-page claim. The vendor-claims register on this site lists every cell that is capped and why.
Sources
- Shapes agents, read 8 Nov 2026
- Shapes: turn AI features on or off, read 8 Nov 2026
- HiBob agents, read 8 Nov 2026
- Factorial One, read 8 Nov 2026
- Gusto MCP, read 8 Nov 2026
Last reviewed 8 Nov 2026.