Why Klaviyo is opening its stack to builders
Generated by Klaviyo AI
Klaviyo is opening more of its stack to the developers and AI agents now building software. Klaviyo, the autonomous B2C CRM, began as a database company and is treating that foundation as infrastructure others compose with: an MCP layer used daily by marketers and agency partners, machine learning models like personalized send time that predate LLMs and still outperform them, and a new Stripe Projects integration that provisions an account, scoped API credentials, and an LLM context file from a single terminal command.
- The durable value in software is moving from the interface to the customer understanding underneath it.: As agents, applications, and workflows all draw on the same shared view of the customer, the UI matters less than the data layer every one of them reads from.
- Klaviyo is the first and only B2C CRM in the Stripe Projects developer preview.: One command stands up a Klaviyo account, writes scoped API credentials into the environment file, and places an LLM context file alongside them, so any coding agent immediately knows how to call the APIs.
- Proprietary data, not model access, is the moat.: Anyone can reach an AI model, but the evaluations, the guardrails, and 14 years of first-party customer data are what make output brand-true. The models are commodity; the intelligence drawn from the data is not.
In May, I wrote that the biggest architectural shift in enterprise software is the decoupling of capability from interface. As AI reshapes how businesses operate; the durable value moves from the UI toward the shared understanding of the customer that agents, applications, and workflows draw from.
That was the bet. What has grown clearer since is that AI is doing much more than just changing how we do business at a high level. It’s changing who (and what) builds software.
The set of people who build software, and the tools they reach for, is more dynamic than ever. A founder can describe an idea to a coding agent and see a working product within the hour, and a marketer can stand up the infrastructure behind a campaign without waiting on IT to wire it together. The slowest part is no longer writing the code, but choosing the infrastructure, getting credentials, and stitching services together so data flows where needed.
Where Klaviyo sits in the stack
This is where the shape of our infrastructure matters. Klaviyo began as a database company. While databases were once thought of as applications, the industry long ago came to understand them as infrastructure, much as the market moved from buying server software to relying on the hyperscalers' managed services. Klaviyo’s investments have built on that foundation but, for most of our history, we exposed only the topmost layer that customers touch directly. We are now opening more of our stack to the builders and agents who want to compose with it.
Klaviyo also runs machine learning models that predate the current wave of LLMs and still outperform them on specific, high-value problems, such as predicting the right send time for an individual. Even where we build on top of language models; the value lives in the evaluations, the guardrails, and the brand-true output that fourteen years of proprietary customer data makes possible. Anyone can access AI models, but the data, and intelligence we draw from it, is not.
Meeting builders wherever they build
The latest proofpoint is that Klaviyo is joining the Stripe Projects developer preview as a co-design and launch partner, as the first and only B2C CRM in the program.
Stripe Projects is a command-line workflow that lets developers, and the AI agents working on their behalf, provision services, store credentials, and manage billing without leaving the terminal.
With a single command, a builder can stand up a Klaviyo account, receive scoped API credentials, and start integrating customer data and messaging into their project.
The credentials are written into the environment file, and an LLM context file is placed alongside them, so any coding agent immediately knows how to use our APIs. It builds on our MCP layer, used every day by marketers and agency partners, and extends that approach to a more technical audience of developers.
A shared, default stack
At most companies of enterprise size, the marketing stack is still a Frankenstack: it’s a set of separate tools a technical team must plumb together and maintain. With AI, marketers will direct agents to do the wiring, spend far less effort on it, and focus on the outcome rather than the plumbing.
Our aim is for Klaviyo to be the part of the stack builders reach for by default. It is agent-friendly, increasingly designed to be consumed by agents as naturally as by people. It is robust and proven, tested at the scale of billions of signals and hundreds of millions of messages a day across 196,000 brands. And it is evolving quickly, because meeting builders in the terminal, in the IDE, with the agents they are orchestrating, and in whatever tool turns an idea into a working product is a direction we intend to keep pushing.
Related content
