Live Helper Chat vs Plain

Plain's changelog is dated entries with no titles, and nearly all of it concerns two AI surfaces: Ari, the customer-facing responder, and Sidekick, the internal agent. Sidekick spent July gaining reach — its own full-page home, reusable workspace skills, and connection to arbitrary MCP servers. The newest release turns to Ari and adds what was missing around it: a way to test it before customers see it, provenance behind each answer, configurable tone, and explicit escalation rules.

The sequence reads as capability first, then governance. Plain rebuilt Ari into an agentic first responder, opened Sidekick to any MCP server, and let teams write skills the whole workspace shares — all of which widen what the agents can do and reach. The current release closes the loop the other way, making that autonomy inspectable and bounded before it is switched on. That is the order support-AI vendors have to complete in: teams will not put an agent in front of customers on capability claims alone.

With testing and provenance in place, the next step is likely measurement — resolution and escalation rates per skill, or per-answer quality review feeding back into the skills teams write. The entries are short summaries linking off-site, so per-release depth is limited.