This week, Klaviyo expanded Customer Agent —our AI-powered platform for customer service—with three major additions: Custom Skills, Custom Tools, and Agent Guidance. Any brand can now build entirely new AI workflows for their specific use cases, connect Customer Agent to any system in their stack, and govern how the agent behaves across every interaction.
These purposefully are not feature updates. They’re the capabilities that turn Customer Agent from an out-of-the-box product you deploy into a platform anyone can build on. And the difference between those two things is what I want to talk about.
Most AI service tools are products. Here’s where products hit a ceiling.
Most AI service tools give you a fixed set of capabilities. For the basics, like shipping questions, order status, and return policy lookups, that works. Resolution numbers look good. The ROI story is clean.
But brands quickly run into what you can’t do when all you have is configuration:
- Enforce a strict sequence of operations on a return.
- Route a gifting experience's answers to a separate data scope so they don't pollute the buyer's profile.
- Run a product preference quiz, save the answers to the customer profile, and pick up the thread on the next visit.
You can’t configure your way to those outcomes with a prompt editor. The more consequential the interaction, the more you need a system you can extend, not just one you can adjust.
How Customer Agent is built, and why that makes it a platform
Customer Agent is a collection of agent skills. On the back-end, each skill is a specialized agent, purpose-built to own a specific type of customer request end-to-end. One handles order tracking. Another processes returns. Another makes product recommendations. A routing layer evaluates each incoming message, picks the right skill, and re-routes mid-conversation when a customer's needs shift.
Your customer experiences one continuous conversation. Underneath, each specialist runs its own instructions, calls its own data sources, and follows its own skill for the job it owns. This allows you to get AI precision where it needs to be strict, with built-in flexibility where the conversation calls for it.
This is a platform and not just a well-architected product because each agent is independent and composable, meaning brands can add their own.
- Custom Skills let you define a new AI workflow for any use case your business requires.
- Custom Tools let you connect to any external system.
- Agent Guidance lets you set the behavioral rules.
In this model, you don’t have to wait for us to build the capability you need. You can build it yourself, on infrastructure designed for exactly that.
The data foundation is what makes the platform possible
Other AI service tools use similar techniques like specialized agents, routing, and tool integrations. What makes Customer Agent different isn’t the architecture alone. It’s the foundation the architecture is built on.
Most AI service tools sit on top of helpdesks. Helpdesks manage tickets. They don't know the customer. No purchase history, no loyalty tier, no browsing behavior. When the conversation ends, nothing writes back.
That means that the next marketing campaign is no smarter than the last one.
Customer Agent is built directly on Klaviyo's data platform, where the customer's full history already lives. Before a customer sends a single message, the agent knows who they are, what they've purchased, what they've browsed, and what they're worth to the business.
Every signal from every conversation writes back to the same profile that drives your marketing. Service makes marketing smarter. Marketing context makes service better. That feedback loop is the advantage that compounds.
And it’s what makes the platform work. Every skill a brand builds, every custom tool it connects, every interaction its agent handles—all of it reads from and writes back to the same customer profile. That shared context is what separates a platform you can build real experiences on from a product you can only configure.
What brands are building
Naked Wardrobe deployed Customer Agent across web and text, resolved 84% of support conversations autonomously, drove a 28% increase in average order value, and generated meaningful incremental revenue, all without adding headcount. Service became a revenue channel.
With Custom Skills, the range of what brands can build opens up considerably:
- A reservation concierge that learns preferences across sessions and books the right table at the right restaurant on the right night.
- A personalized gifting advisor that runs a guided quiz scoped to the recipient, makes a confident recommendation, and keeps the buyer's profile clean.
- A device failure agent that reads telemetry, diagnoses the issue, collects a shipping address, and ships a replacement autonomously.
These aren't hypotheticals. Brands on Klaviyo are building them right now.
The platform advantage
The first wave of AI service gave brands a product to deploy. A model, a prompt, a set of pre-built skills. That wave delivered real value, and it set the bar for what resolution rates and response times should look like.
The next wave gives brands a platform to build on. Custom experiences. Connected systems. Intelligence that compounds across every interaction because every interaction reads from and writes back to the same customer data.
Brands that start building on that platform now are going to have a structural advantage that’s hard to replicate later. The architecture makes the composition possible. The data foundation makes it intelligent. And doing both on a B2C CRM is what makes it matter.
Custom Skills for Customer Agent is available now in managed beta. Join the waitlist .
Related content
