Acquisition and runtime infrastructure for production agents
Klaviyo's agreement to acquire Agency and AWS's general release of AgentCore show two production routes for agent products: bringing a team inside a company and providing a managed runtime layer.
Artificial Intelligence··Morning
A team and products brought in by acquisition
Klaviyo has agreed to acquire Agency, the AI-powered customer success startup Elias Torres founded in 2023; the deal terms were not disclosed. Before the acquisition, Agency had raised 32 million dollars from investors including Sequoia, Menlo Ventures and Felicis. The transaction brings a team and product responsibility into Klaviyo together: Torres is set to join as chief product officer and lead Agency's 25-person team. That team's work is developing two Klaviyo AI agents. Composer builds marketing campaigns, while Customer Agent handles post-sale tasks such as returns and order tracking. Klaviyo co-founder and chief executive Andrew Bialecki told TechCrunch that the company aims to bring those products to 200,000 businesses. The acquisition route therefore puts expertise, product management and two defined customer jobs inside one organization. The reported development establishes how the team and development work are being organized; the 200,000-business figure remains Klaviyo's target rather than an observed distribution result.[1]
A runtime layer delivered through configuration
Amazon Web Services announced a different entry point into production. The company said the agent-running harness in Amazon Bedrock AgentCore is generally available and published an open-source community node that brings it into n8n's visual editor. By AWS's description, the harness runs the orchestration loop, manages tool calls and the context window, keeps state across turns, recovers from failures and isolates sessions from one another. Each session runs in its own environment with a filesystem and shell, while memory across sessions and web browsing are also provided. A developer configures the agent through the model it uses, the tools it calls, the skills it can reach and the instructions it follows. The MIT-licensed open-source node is named @aws/n8n-nodes-agentcore. When its single deciding field, Harness ARN, is blank, the node creates the agent on the first run. This route gathers runtime work that developers would otherwise assemble into a managed, configurable layer. General availability describes the product's access status here, while the account of how its functions operate remains AWS's own product description.[2]
Two forms of production organization
The two developments emphasize different organizational pieces in the movement of agent systems into production. Klaviyo is arranging an acquisition that brings Agency's 25-person team, product leader and development work on two agents into its organization. AWS is offering a harness in which models, tools, skills and instructions can be configured, while runtime work such as memory, session isolation, failure handling and tool calls is managed as a layer. The first route concentrates product capability around defined marketing and post-sale tasks inside one company. The second creates a common infrastructure surface for running different agents and connects that surface to a visual editor through the n8n node. The announcements do not present these routes as competing answers to the same problem: Klaviyo is changing the team and product scope of a product company, while AWS is changing the availability of infrastructure developers can use. The reports do not measure which approach is more successful. Read together, they show production agents being assembled through organizational and technical layers including product ownership, specialist teams, tool access, memory and session management alongside model choice. Klaviyo's scale target and AWS's functional account remain company statements at this stage.[1], [2]
