Meet Joe: the assistant inside your workspace

2nd Shift ships with an assistant named Joe, and this post is the honest manual for him — what he does, where he does it, and what happens when the AI isn't switched on at all.

Where Joe shows up

On your website widget, Joe is the first voice: he reads the customer's question and answers from your knowledge base — the published articles you wrote — with a link to the source. When the answer isn't in the KB, he says so and hands the conversation to a human with the transcript attached. He is specifically instructed that admitting uncertainty is better than inventing an answer, and the widget bot's grounding is enforced the same way for every surface.

Inside the workspace, every team member gets an assistant bubble in the corner of the app — same grounding rules, same sources, pointed at your workspace's own articles. It's the "how do I do X in here" question without leaving the page.

On tickets, Joe reads the subject and body and suggests the flows most likely to resolve the ticket, so the agent's first move is a click instead of a search.

In flows, with your AI provider connected, Joe can draft a new guided flow from a description — and every draft must pass the same structural validation the flow canvas uses before it can be saved. He doesn't get a looser rulebook than the humans.

In purchasing, when a vendor's quote arrives as a paste of email text, Joe parses it into structured lines and matches them against your request items — for your review, before anything is saved.

The settings, briefly

AI in 2nd Shift is bring-your-own-key: platform-level enablement is a superadmin decision, keys are per-organization, and they're stored encrypted and never returned to any screen after saving. Usage meters, budget limits apply, and every AI surface degrades gracefully: if the key is missing or the budget is exhausted, you get a readable error and a UI that keeps working — the flow suggestions fall back to keyword matching, the widget hands off to humans, nothing crashes, nothing pretends.

What Joe is not

He's not autonomous. He doesn't send emails you didn't approve, change records on his own, or touch money. He answers, suggests, drafts and parses — and the meaningful actions stay behind buttons operated by people. That boundary is deliberate, and it's why we're comfortable putting him in front of your customers: the worst case of a Joe mistake is a wrong answer with a visible source, not a wrong action with invisible consequences.

Try him

If your workspace already has a knowledge base, open the widget on your site and ask it your most common customer question. If the answer comes back with a source link to your own article, that's Joe working as designed — your writing, at scale.

One rule governs all of it: Joe's answers come from your workspace's own knowledge base, or he says he doesn't know. He doesn't improvise product claims, invent policies, or guess at your numbers. When the knowledge base is thin, that honesty is the feature — it tells you exactly what to write next.

And when he's wrong, it's checkable: every grounded answer names its source, so a wrong answer is one click away from being seen, not a mystery.

Related: start with why we built 2nd Shift, or see Joe's grounding source up close in the knowledge base post.