Eight words carry the whole story of customer-facing AI's last decade: chat answered questions; live finishes jobs. The line sounds like a slogan and works like a diagnostic, because the gap between an answered question and a finished job is where most customer experience still leaks, and most tooling still lives on the wrong side of it.
An answered question is not a finished job
Watch what actually happens after a perfect chat answer. The user reads the steps, switches windows, misremembers step three, meets an interface that shipped since the article was written, improvises, and returns to ask a different question that's really the same question wearing the consequences. The bot answered; nobody finished. The job, the integration connected, the account configured, the issue resolved and verified, the skill actually built, was never inside the conversation at all. It was on the screen the conversation couldn't see, made of actions the conversation couldn't take, spanning a timeline the conversation couldn't remember. Answering was the part language could reach; finishing needed organs language doesn't have.
What finishing requires
Name the job honestly and the requirements fall out. Finishing requires perception: you can't complete what you can't see, and state, not knowledge, is where jobs live. It requires action: someone has to click the thing, and every handoff back to the user's hands is where completion leaks. It requires continuity: real jobs span sittings, so memory that dies with the window kills the job with it. And it requires a definition of done: an agenda with an exit, the milestone confirmed, the fix verified working, the workflow performed unassisted, because jobs, unlike conversations, end somewhere in particular. These are exactly the capabilities the live stage added, which is why the eight words split the eras where they do.
The metric version of the line
Every team can translate the slogan into instrumentation. Count answered questions and you'll optimize answering: deflection up, articles sharper, and the silent cohort of unfinished jobs untouched, sometimes growing, because each friction-filled loop teaches users to stop asking. Count finished jobs, activations confirmed, resolutions verified on-screen, competence demonstrated, and the whole system reorients toward the thing customers were actually paying for. The line is a question to ask of any dashboard: which side of the gap does this number live on?
The standard worth adopting
The practical takeaway isn't anti-chat, answering questions well remains real value, and the chat layer earned its place. It's a raised bar: for the work that decides revenue, demos, onboarding, support, training, "we answered" stopped being an acceptable terminal state. Ask of every customer-facing tool: where does the job finish, and who watched it finish? (Disclosure: finishing is the business we're in at Skippr, agents measured on outcomes, meetings booked, accounts activated, issues resolved, users trained, which is the eight words, operationalized.)
See what a live agent actually does
The category is easier to watch than to define. Fifteen minutes is enough to see where the mechanism differs from everything it gets confused with.