Every capable AI employee is built from the same four capabilities: eyes that see the user's screen, a voice that holds real conversation, hands that act in the product, and a brain that knows the business and remembers the account. The anatomy matters because each organ answers a different failure mode, and missing any one produces a recognizably lesser creature.
Eyes: perception of live state
The eyes see what the user sees, which page, what state, what's configured, what just errored, where the hesitation is, updated in real time, in the product and on the shared screen of a call. Eyes are what retire the description tax ("can you describe what you're seeing?") and what keep the agent's knowledge honest as interfaces change weekly. Without eyes: a chatbot, answering descriptions instead of situations, however sharp its answers.
Voice: conversation at working bandwidth
The voice holds real-time spoken dialogue, interruptible, silence-tolerant, multilingual with mid-conversation switching, and gives the agent presence: it can join a kickoff call, hold a demo's attention, talk a stuck user through a hard moment while their hands stay on the work. Voice matters because the highest-value exchanges are high-back-and-forth, and typing throttles them. Without voice: a coach behind glass, helpful, but absent from the rooms where outcomes get decided.
Hands: action in the product
The hands execute: annotate the screen, propose the next step against the agenda, then click, navigate, fill forms, run configurations, and report back, with permission, narrated, logged, and where the work demands it, integrate with APIs to run deeper tasks. Hands close the gap between advice and outcome, the difference between "here's how" and "it's done, watch." Without hands: an advisor, leaving every translation from instruction to click, and every translation error, to the user.
Brain: knowledge, memory, and the plan
The brain grounds the agent in the business, trained on the knowledge base, docs, and playbooks, rescanning the product as it changes, and carries what time requires: the agenda for multi-step work, follow-through, and memory of the account, history, context, next steps, shared with the team. The brain is what makes the other three organs an employee rather than a gadget: eyes, voice, and hands in the service of a plan that survives the session's end. Without the brain: a quick hit, impressive in the moment and amnesiac by the next one.
Why the anatomy only works whole
Each pair amplifies: eyes plus hands make action safe; voice plus eyes make diagnosis fast; brain plus everything makes sessions resumable and teams possible, the agent that demoed the account handing memory to the one that onboards it. The shorthand is talk, see, and take action, with the brain underneath. (Disclosure: this is literally the build sheet of Skippr's agents, Skippr AI demo, Skippr AI onboarding, Skippr AI technical support, and Skippr AI training are the same four organs wearing four jobs; use the anatomy to audit anyone's "AI employee" claim, ours included.)
Questions buyers actually ask
Which capability matters most?
The one you're missing. Diagnostically: blind agents interview, mute agents can't join calls, handless agents advise, brainless agents forget. The category requires all four.
Is this anatomy specific to one vendor?
No, it's a useful audit for any live-agent claim: ask what the agent perceives, speaks, performs, and remembers, and demand demonstrations of each.
How does the brain stay current?
By deriving knowledge from live sources, the product, the docs, rescanned as they change, rather than from renders that age.
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.