Every era of customer-facing software has a unit of work it optimizes. The ticket era counted resolutions; the content era counted views; the chat era counted deflected conversations. The live agent thesis is that the unit that matters now is the session: a continuous engagement with an agenda, and that businesses which reorganize around sessions will quietly outperform those still counting the old units.
Units of work discipline everything around them
The unit you count becomes the shape of what you build. Teams that count views produce content, more of it, better targeted, endlessly refreshed, and the customer's unfinished job survives every view. Teams that count deflections produce absence: fewer tickets, which is not the same as fewer problems, and sometimes its opposite. Units aren't neutral measurements; they're commitments about what the work is. The deep claim of the live agent thesis is that customer-facing work was always session-shaped, a demo is a meeting with a storyline, onboarding is a plan across days, support is diagnosis-to-verified-fix, training is practice toward competence, and we counted fragments of it because fragments were all our tools could hold.
What a session commits you to
Three things the old units never demanded. An outcome definition: a session ends somewhere, the milestone confirmed, the issue verified fixed, the workflow performed unassisted, so "done" stops being a feeling and starts being a state. Continuity: sessions resume, which requires memory that belongs to the account, and makes multi-day jobs one engagement instead of serial introductions. And accountability: a session has an agenda it can be judged against, in a way no view, click, or deflection ever could. This is why session-counting changes behavior upstream of any technology: the moment you ask "how many jobs did we finish with customers this week?" instead of "how many interactions did we log?", different work gets funded.
The reorganization that follows
Take the thesis seriously and the org chart's seams start looking like session-breakers: the demo team's context dying at the sales handoff, support meeting the customer as a stranger, training scheduled by cohort instead of by need, each a place where a session-shaped job was cut to fit fragment-shaped tooling. The remedies share a spine: agendas that persist, memory that travels, agents (human and AI) that hand work to each other without dropping it. Live AI agents matter here not as gadgets but as the first tooling that makes session-native operation affordable at every tier, the session that once required a scheduled human now scales like software.
The falsifiable version
Theses should bet something. This one bets that session metrics, jobs finished, agendas completed, time-to-outcome, will predict retention and revenue better than view, click, and deflection metrics do, at any company that instruments both. Run the comparison on your own funnel; the thesis is happy to be tested. (Disclosure: we build Skippr on exactly this bet, sessions with agendas are the product's spine, and the metric argument above is the one we'd want used against us.)
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.