Action Layer Determines AI Winners
Decagon
The strategic issue is that model providers can move up the stack and sell customer service products themselves, which compresses room for pure application vendors. AWS already sells Amazon Connect AI agents that can answer questions, update accounts, and work inside contact center flows. Google sells a full contact center stack through Contact Center AI Platform and Dialogflow. OpenAI is now packaging enterprise agents with forward deployed engineers for workflows including customer support, so the same companies supplying models can also chase the application budget.
-
For Decagon, the defense is product depth above the base model. The workflow layer is where value lives, custom integrations into ticketing and back office systems, guardrails, QA, and the operating logic that lets an agent issue refunds, update records, or escalate cleanly. That is why Decagon and Sierra both rely on high touch deployments rather than simple self serve setup.
-
The overlap is real, but the products are shaped differently. Cloud vendors sell broad toolkits and contact center platforms that large enterprises can configure themselves. Specialists sell faster time to value for a narrow job, replacing support labor at roughly $1 to $1.50 per resolution versus about $10 to $15 for human handling, with 60% to 80% containment in strong deployments.
-
This same pattern has already shown up inside customer support software. Intercom bundles Fin with ticketing, inbox, knowledge base, and agent assist, while Decagon can also sit on top of Intercom for escalations and human handoff. That means specialists are squeezed from both sides, by model platforms below and by systems of record above.
Going forward, the winners in AI customer service will be the companies that own the action layer, not just the chat layer. Foundation model vendors will keep entering enterprise workflows, and help desk incumbents will keep bundling AI. Specialized players that become the fastest path to safe automation across chat, email, and voice will keep a durable place in the stack.