Open any software category's release notes this year and you'll find the same word: agent. The bolting-on is rational, incumbents have distribution, customers, and boards asking about AI, and it mostly underdelivers, for a reason that has nothing to do with effort or talent: the form factor underneath was built for a different job, and agents inherit their container's constraints.
Why the bolting is universal
The incentives are airtight. The capability wave, perception, action, real-time voice, arrived as infrastructure anyone can build on, so not shipping agent features reads as falling behind. Incumbents own the deployments and the data, so an agent inside the existing product is the shortest credible path to an AI story. And the first demos are genuinely easy: an assistant that drafts, summarizes, or suggests inside your current UI ships in a quarter and demos beautifully. None of this is cynical; it's what any rational roadmap does with a general-purpose capability. The trouble starts after the demo.
The container problem
A bolted-on agent lives inside its host's form factor, and form factors have load-bearing assumptions. A tool built around authored artifacts, recorded tours, walkthrough flows, rendered courses, gains an agent that authors artifacts faster: the bottleneck moves, the artifact remains, still identical for every viewer, still aging as the product ships, still silent when a question arrives. A tool built around the conversation layer gains an agent that converses better, and still can't see the screen where the problem lives. A dashboard product gains an agent that explains the dashboard, and still can't act on what it measures. In each case the agent inherits the container's ceiling, because the container defines what the product structurally is: the artifact, the thread, the chart. Intelligence was never the constraint; the form factor was.
Why the gap is architectural, not temporal
The optimistic incumbent read is "we'll get there iteratively," and for features, iteration works. Form factors are harder: session-native work, an agenda held across days, memory that belongs to the account rather than the thread, agents that hand work to each other as a team, isn't a feature you add to an artifact pipeline; it's a different spine. Retrofitting it means carrying two architectures while your pricing, your onboarding, and your customers' habits all assume the old one. Some incumbents will manage the crossing; most will ship increasingly good agents inside increasingly irrelevant containers, and the release notes will look great the whole way down.
What buyers should take from this
Ignore the word "agent" entirely and interrogate the spine: does the session survive the window closing? Does memory belong to the account? Can it perceive and act on the live product, or does it produce artifacts about the product? The answers sort the market faster than any feature grid. (Disclosure: we build at Skippr, where the spine came first, live, session-native, team-based, and the reader deserves the same skepticism pointed at us: run the session-two test on everyone.)
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.