Klaviyo moves Composer into public beta and expands Customer Agent
Klaviyo’s Composer and Customer Agent push CRM toward agentic execution, linking service signals to campaign automation across channels.
Klaviyo has released its Composer AI marketing agent into public beta and broadened its Customer Agent into a more extensible “agent platform” spanning marketing and service workflows.
The update matters because it pushes AI in CRM from “assistive” features (suggestions, predictions) toward agentic execution, where the system can assemble campaigns, resolve common support tickets, and feed learnings back into a unified customer profile for reuse across channels.
Table of contents
Jump to each section:
- What Klaviyo shipped: Composer, Customer Agent platform, and Social Marketing
- How the agents work in practice (and where the limits are)
- Customer Agent’s “build on it” shift: connectors, APIs, and multilingual support
- Competitive context: Klaviyo vs Salesforce, HubSpot, Braze, and Bloomreach
- What this signals about AI-powered CRM and workflow automation
- Practical takeaways for retention and customer experience teams
ContentGripWhat Klaviyo shipped: Composer, Customer Agent platform, and Social Marketing
Composer is positioned as an AI marketing agent that can build a complete campaign from a plain-language brief, including audience selection, cross-channel copy drafts (for example email and SMS), and send timing. The control point Klaviyo emphasizes is approval: campaigns are staged for review rather than automatically pushed live.
On the service side, Klaviyo is expanding Customer Agent from a prebuilt service agent into a platform, adding a “Conversational Agent Builder” experience, plus connectors and open APIs that let the agent take actions in downstream systems rather than only answer questions.
Klaviyo also added Social Marketing features aimed at converting social engagement into owned subscribers, such as DM-based auto replies that capture consented email or phone opt-ins, plus richer engagement events and a social content library for organizing UGC and owned posts.
How the agents work in practice (and where the limits are)
The core design choice is that both agents operate on the same underlying customer profile and event stream, so service interactions can update marketing context and vice versa. In practice, that means a service conversation can capture preferences and intent signals (size, color, product interest), write those back to the profile, and make them available for future segmentation and personalization.
The value is not “AI copywriting.” It is the orchestration layer: auditing what is running, identifying opportunities (for example underperforming flows), and assembling the set of assets and steps needed to execute. That can reduce the operational burden for lean retention teams managing large automation libraries.
The constraint is governance. “Grounded in your customer data” can still fail if your data is incomplete, if policies are inconsistent across channels, or if the business has edge cases that require human judgement. Teams will still need QA, holdouts, brand and legal guardrails, and a clear escalation path when an automated resolution should not happen.
Customer Agent’s “build on it” shift: connectors, APIs, and multilingual support
The shift from “agent” to “platform” is mostly about extensibility. With prebuilt skills (order tracking, returns, recommendations, loyalty), brands can start quickly, but the newer promise is that a non-technical operator can describe a custom experience and generate the underlying conversation flow and logic.
Connectors and open APIs are the practical enablers: the agent can complete actions (for example initiating a return or applying loyalty benefits) rather than just explain what to do. That is also how Klaviyo can support brands beyond standard commerce stacks, including headless and custom storefronts, by embedding the agent in more surfaces.
Multilingual expansion (100+ languages) and support across chat, email, SMS, RCS, and WhatsApp is less about novelty and more about reach. If the same agent logic is meant to drive resolution rates, it must work across the channels where support demand actually appears, not only in on-site chat.
Klaviyo has cited performance signals such as 65% questions resolved automatically and individual brand examples reaching high autonomous resolution rates within 90 days. For marketers and CX leaders, the important question is measurement design: what counts as “resolved,” how often the agent drives re-contact, and whether automation changes refund rates, AOV, or customer satisfaction over time.
Competitive context: Klaviyo vs Salesforce, HubSpot, Braze, and Bloomreach
Klaviyo’s pitch is a B2C CRM that unifies customer data, marketing automation, analytics, and service workflows, with AI agents operating directly inside that system. That puts it in competitive territory with large CRM suites like Salesforce and HubSpot, and with marketing engagement platforms like Braze and Bloomreach that also emphasize orchestration and personalization.
Where Klaviyo can differentiate is ecommerce- and B2C-first workflow depth: lifecycle programs (welcome, cart recovery, winback), identity and consented messaging across channels, and a tighter loop between purchase behavior and campaign execution. For brands that live inside high-frequency retention cycles, “audit, rank opportunities, build campaigns” targets the day-to-day bottleneck.
Where competition remains intense is enterprise breadth and cross-department standardization. Salesforce and HubSpot can anchor broader go-to-market and service operations across more functions, while Braze and Bloomreach compete on sophisticated segmentation, experimentation, and personalization patterns. In that landscape, agents are becoming table stakes; durability will depend on data quality, governance features, and how reliably agent actions translate into measurable commercial outcomes.
What this signals about AI-powered CRM and workflow automation
This update aligns with a broader shift toward AI-powered CRM where the system does more than recommend. The “agent” framing is effectively a claim that the CRM can take on execution tasks that used to require specialists: assembling creative variants, coordinating timing, and handling high-volume service interactions.
It also reflects the direction of marketing workflow automation: fewer point tools, more integrated systems where service, marketing, and analytics share the same underlying profile. The strategic bet is that first-party data and unified identity make agents more useful, because the agent can operate with business-specific context rather than generic prompts.
For the market, the implication is competitive pressure on CRM and marketing automation vendors to deliver not just generative features, but controlled automation: auditability, human approval checkpoints, simulation/testing, and clear ROI reporting.
Practical takeaways for retention and customer experience teams
Start with a narrow set of “safe” automations. For marketing, that might be winback or abandoned cart improvements where success metrics are clear. For service, start with tracking, returns, and policy FAQs before moving into more complex exceptions.
Treat agent outputs like a new production pipeline. Define brand voice rules, compliance constraints, and channel-specific QA steps. If the system can draft and stage campaigns quickly, the limiting factor becomes review quality and experimentation design, not copywriting speed.
Instrument outcomes beyond resolution rate. For Customer Agent, track repeat contacts, refund and exchange rates, CSAT, and whether recommendations change conversion or AOV. For Composer-led work, track incremental lift, deliverability impact, and whether automation reduces message collisions across large flow libraries.
Finally, plan for organizational ownership. Agentic workflows cut across marketing ops, lifecycle marketing, and CX. Without clear ownership of data hygiene, approvals, and escalation, the benefits can be diluted even if the underlying tools are capable.
