The first wave of business AI came in one shape: a single assistant, bolted onto a website, expected to handle everything. Mikhail Antipkin, an operator with a background in payments and mid-market technology, believes that model is already giving way to something more familiar from consumer software. Specialized agents, installed on demand, each doing one job well.
“We have seen this movie before,” Antipkin says. “Phones did not get useful because one app tried to do everything. They got useful because a marketplace let a thousand developers each solve one problem. Business AI is following the same path, just faster.”
Why The All-in-one Assistant Hits a Ceiling
The logic of the monolithic assistant made sense at first. Companies wanted one interface, one vendor, one integration. The problem, Antipkin argues, is that business functions are not interchangeable. The knowledge, tone, and guardrails required to handle a billing dispute have little in common with those required to onboard a new user or recover an abandoned subscription.
“When one system tries to cover billing, sales, retention, and technical support, it is mediocre at all four,” he says. “Operators know this instinctively. You would never hire one person for those four jobs and expect excellence. The software version of that hire fails the same way.”
The result, in his view, is a predictable pattern: companies deploy a general-purpose assistant, watch it perform well on frequently asked questions and poorly on everything else, and conclude that AI support “doesn’t work for our business.” What actually failed was the architecture, not the technology.
The Composable Alternative
The emerging model separates the foundation from the specialization. A core AI assistant handles the common layer, trained on the company’s documentation, policies, and past conversations. On top of that foundation, businesses install purpose-built agents for specific functions, published by partners who understand those domains deeply.
Antipkin compares it to how modern commerce stacks already work. “Nobody builds their own payment processing, their own shipping logic, and their own tax engine anymore. You compose your stack from specialists. Support and customer engagement are simply catching up to how the rest of the operation is already built.”
The practical advantage shows up in speed and risk. Instead of a long implementation project with a single point of failure, a company can start with the core assistant, measure results, and add capabilities one agent at a time. Each addition is small, reversible, and judged on its own performance.
What This Means for Mid-market Operators
For the companies Antipkin knows best, those large enough to feel operational complexity but too lean to build in-house AI teams, the marketplace model changes the calculus entirely. The build-versus-buy question dissolves when specialized capability can be added in a click.
He suggests operators evaluate platforms on three questions. Does the core assistant genuinely learn from your existing knowledge, or does it require months of manual scripting? Is there a real ecosystem of specialized agents behind it, with partners incentivized to keep improving them? And can you start small, prove value, and expand, rather than committing to a big-bang rollout?
“The mid-market cannot afford science projects,” he says. “The right question is not whether the technology is impressive. It is whether you can be live this week, measure it this month, and expand it this quarter.”
An Ecosystem Taking Shape
Antipkin points to the platform layer as the piece worth watching. VivoChat, a free AI customer support platform, pairs a core assistant trained on a company’s own knowledge base with a growing marketplace where partners publish specialized agents for billing, onboarding, sales, and retention that businesses can install in one click. The structure mirrors exactly the shift he describes, a foundation plus an ecosystem, rather than a single tool trying to be everything.
“The winners in this category will not be the ones with the smartest single model,” Antipkin says. “They will be the ones that make it easiest for a business to assemble the exact capability it needs, piece by piece, and swap pieces as the business changes. That is a marketplace problem, not a model problem.”
NewsBlaze editorial and news staff were not involved in the creation of this article.
