K:BOS 2026 Demonstrates Klaviyo’s Marketing-Service Flywheel for Autonomous B2C CRM
AT K:BOS 2026, Klaviyo demonstrated how quickly it is expanding beyond its origins in email and SMS marketing automation. The company’s platform combines the Klaviyo Data Platform (KDP), personalization models, marketing and service channels, and two AI agents: Composer for marketers and Customer Agent for consumers. Klaviyo is also making its data and actions available outside its own interface through APIs, Model Context Protocol (MCP), and a command-line interface.
The announcements support Klaviyo’s ambition to become an autonomous B2C CRM. The company is moving from software that helps marketers complete individual tasks toward a platform in which AI can interpret a goal, analyze customer and performance data, recommend or create an intervention, and eventually monitor and execute work more proactively.
The result is a marketing-service flywheel. Marketing and service interactions produce customer signals, KDP unifies those signals, personalization models determine the next appropriate experience, Composer and Customer Agent act on those decisions, and the resulting outcomes return to the data platform.
New SQL Access Within KDP
Klaviyo’s principal announcement introduced SQL access within KDP. A marketer can ask a question in natural language through Composer, MCP, or another AI interface; Klaviyo translates the request into SQL, runs it against the organization’s Klaviyo data, and returns both the answer and the query used to produce it.
SQL access was in preview at K:BOS. Klaviyo also described a future semantic layer through which customers could add business-specific definitions and metrics. This is an important step toward making KDP intelligible to business users and external agents, because answers can be inspected at the query level rather than as opaque model output.
However, buyers should not confuse accessible SQL with a universal enterprise data layer. This capability’s immediate value is strongest where the relevant customer, commerce, marketing, and interaction data already resides in Klaviyo. Organizations should test data freshness, identity resolution, query performance, row- and field-level access controls, semantic consistency, and the treatment of data sourced from other systems. If the most authoritative customer, product, consent, service, or financial data remains elsewhere, Klaviyo will still depend on integration quality and external data governance.
Klaviyo also renamed Marketing Analytics as Personalization. The purpose is to emphasize that this capability can understand which customer to reach, what to present, when to engage, and through which channel. Klaviyo is using specialized models for different decisions rather than presenting one general-purpose model as the answer to every personalization problem. The architecture is sensible, but buyers should ask how each model is trained and evaluated, which inputs drive a recommendation, how conflicting models are reconciled, and whether uplift is measured through incrementality rather than Klaviyo-attributed revenue alone.
Composer Is Moving From Assistant to Marketing Agent
Composer has developed rapidly since its introduction in March 2026. Klaviyo states that more than 138,000 users now rely on it to analyze performance and create campaigns. Composer can analyze account data, audit flows and campaigns, draft campaigns and flows, create segments, edit segments, and apply reusable skills containing brand guidance or team practices. SQL access gives it a deeper analytical interface, while scheduled and recurring tasks allow it to surface opportunities or performance deterioration without waiting for a marketer to begin a new analysis.
Composer is now useful beyond simply conversational search. For example, it could identify a declining campaign, question whether a blanket 20% discount is appropriate for a particular segment, and recommend a better intervention. Klaviyo executives described a longer-term model in which Composer operates continuously in the background, drawing on integrated data and eventually external signals such as Meta activity – ultimately, providing proactive recommendations. More advanced simulations and idea ranking remain in the feature pipeline, so buyers should distinguish today’s scheduled analysis from a fully autonomous marketing agent that continuously observes the business and initiates action.
Composer’s usage model also deserves attention as proactive tasks expand. Klaviyo prices Composer through dynamically consumed capacity, with cost varying by task complexity and the quantity of data analyzed. Scheduled monitoring may be inexpensive at current scale, but a continuously running agent changes the cost model. Enterprises should require per-task estimates, budgets, usage alerts, and controls over task frequency before enabling background monitoring broadly.
Headless Klaviyo Extends the Platform Beyond Its Own Interface
Klaviyo is now “headless,” exposing more than 260 MCP tools and capabilities and more than 490 APIs. Users and agents can access Klaviyo capabilities through ChatGPT, Claude, Cursor, other AI systems, or custom applications. The company also introduced its first command-line interface. Klaviyo says new functionality will be accessible through both its user interface and programmatic surfaces.
This reflects a broader shift in enterprise software. As conversational AI becomes a common work surface, vendors can no longer assume that users will enter their application to obtain value from its data or functions. Headless access reduces change-management friction, allows technical and nontechnical users to work through familiar interfaces, and enables Klaviyo to participate in workflows assembled outside its product.
Opening the interface does not eliminate Klaviyo’s moat. Durable value will come from the quality and recency of its customer data, identity graph, consent and channel context, commerce integrations, decision models, marketing knowledge, and ability to execute and measure experiences. However, headless access also makes individual Klaviyo functions easier to combine with or substitute for capabilities from other vendors. Klaviyo must therefore prove that its integrated data and execution loop produces better decisions than a collection of external models and point solutions operating on the same data.
The security implications are also worth highlighting. Klaviyo’s MCP documentation shows that its remote server supports both read and write tools and that read-only mode is optional. It also provides a setting to disable tools that expose user-generated content because such content could contain malicious instructions for an LLM. This is a welcome acknowledgement of prompt-injection risk, but safer configuration should not depend on every customer discovering and activating optional parameters. Enterprise deployments should apply least-privilege roles, default to read-only access where possible, separate development and production accounts, require approval for consequential actions, restrict access to user-generated content, and retain auditable records of prompts, tool calls, changes, and approvals.
Customer Agent Connects Service and Commerce Data
Customer Agent is Klaviyo’s customer-facing AI agent. Klaviyo reports that it has exchanged more than 1.3 million messages and now supports over 100 languages, with automatic locale detection through Shopify Markets and access to the appropriate regional catalog, pricing, and currency. It currently operates across web chat, email, SMS, and WhatsApp.
Customer Agent’s available skills include product and policy questions, recommendations, order tracking, returns and exchanges, order changes, subscription management, and loyalty lookup. Customer Agent uses the shopper’s Klaviyo profile and interaction history, brand content, guidance, and connected tools to respond or take action. Klaviyo provides eight skills out of the box, while fully custom skills and tools that connect to HTTP endpoints remain in open beta.
This makes Customer Agent attractive to B2C brands that already use Klaviyo and want marketing and service to operate from the same customer context. Its differentiator is the ability to combine service interactions with commerce, behavioral, channel, and campaign data already held in Klaviyo before feeding the results back into future personalization.
Klaviyo also teased that voice will become a new Customer Agent channel, capable of handling a call from beginning to end with access to the customer’s history. The strategic logic is understandable. Voice interactions generate valuable intent, sentiment, and service data that can enrich the profile and improve the marketing-service flywheel. A Klaviyo-native voice agent could also apply the same commerce knowledge, guidance, and customer context used across digital channels. News on release, architecture, integration, governance, and pricing is forthcoming.
Our Take
Klaviyo’s showcased a compelling marketing-service flywheel that powers its autonomous B2C CRM. KDP provides the context, personalization supplies decisioning, marketing and service products provide channels, and Composer and Customer Agent turn insight into work. Headless access allows this value to appear inside other agents and applications rather than remaining confined to Klaviyo’s interface.
The company’s strongest advantage is the closed loop between customer data, engagement, service, personalization, and measurable commercial outcomes. Customer Agent can turn a service conversation into new profile and intent data, Composer can use the enriched context to improve the next campaign, and Klaviyo can activate the resulting decision across its channels. For prospects considering Klaviyo, this broader operating model is the vision to keep in mind.
The buyer risk is scope mismatch. Klaviyo’s capabilities are mature, while its autonomous service, proactive orchestration, cross-platform execution, and voice capabilities are at different stages of development. Existing Klaviyo customers should prioritize use cases that benefit directly from Klaviyo-held data and channels, validate native functionality before adding specialist agents, and retain established service platforms where operational depth is required. Prospective enterprise customers should also evaluate data residency, identity resolution, consent, model transparency, integration ownership, availability commitments, and usage-based costs.