Klaviyo’s acquisition of AI startup Agency is less about adding another chatbot than about securing the product leadership, customer context, and execution experience needed to make autonomous software useful in commercial workflows.
Klaviyo has agreed to acquire Agency, a three-year-old artificial-intelligence startup founded by serial entrepreneur Elias Torres, in a deal that reflects a broader shift in enterprise software. Established platforms are no longer treating AI agents as experimental add-ons. They are trying to build them into products that can make decisions, take actions, and handle customer interactions at a scale that justifies their cost.
The terms of the transaction were not disclosed. Agency had previously raised $32 million from investors including Sequoia, Menlo Ventures, and Felicis. Torres will join Klaviyo as chief product officer, while Agency’s 25-person team will move to the publicly traded marketing-automation company.
Klaviyo says the acquisition will accelerate its AI-agent strategy, including Composer, a product that helps businesses create marketing campaigns, and Customer Agent, which is designed to manage post-sale requests such as returns and order tracking. The company’s larger ambition is to combine Agency’s technology and product experience with Klaviyo’s existing customer data, integrations, and base of roughly 200,000 businesses.
That combination matters because the central challenge in enterprise AI is no longer proving that a language model can produce a convincing answer. The harder question is whether an agent can reliably complete a business task, within defined limits, using current information and without creating expensive errors. Klaviyo believes its position inside merchants’ marketing, purchasing, and support workflows can provide the operating context that standalone AI companies must build one customer at a time.
The acquisition therefore offers a useful test of where the AI-agent market may be heading. Will customer-facing agents become core features of incumbent marketing, commerce, and CRM platforms? Or will independent companies such as Decagon and Sierra establish themselves as a new software category before larger vendors can catch up?
Klaviyo’s move suggests that established software companies are preparing to compete on both fronts: by developing agents internally and by buying the teams that understand how to turn AI capabilities into products customers will trust.
The strategic value is broader than Agency’s technology
The most important asset in this transaction may not be Agency’s standalone product. It may be the company’s people and product judgment.
AI startups can build impressive demonstrations quickly. Turning those demonstrations into dependable software requires a different set of skills. Products must connect to customer records, order-management systems, payment tools, inventory platforms, shipping data, and internal approval processes. They also need controls for situations in which the agent lacks enough information or the requested action has financial or reputational consequences.
Those requirements are particularly important in customer service. An agent that answers a routine question about shipping status can create value by reducing workload and shortening response times. An agent that incorrectly approves a refund, changes an order, or provides inaccurate policy information can create a direct financial loss and damage a merchant’s relationship with its customer.
Agency’s team has been working on these kinds of customer-success applications, giving Klaviyo access to people who have already confronted the practical limitations of AI in commercial settings. The group is small by enterprise-software standards, but that can make it strategically valuable. A focused team with a shared product vision can often move faster than a large organization divided among established priorities.
Torres’s appointment as chief product officer also gives the acquisition significance beyond a conventional acqui-hire. He will not simply be advising Klaviyo’s AI team or overseeing a narrow product line. He will take on a companywide product role at a moment when Klaviyo is deciding how deeply AI should reshape its platform.
Torres has experience building and selling software businesses. He co-founded Performable, which HubSpot acquired in 2011, and later co-founded Drift, which Vista Equity acquired in 2021 for $1.2 billion. Klaviyo Chief Executive Andrew Bialecki was one of Torres’s early hires at Performable before founding Klaviyo himself.
That relationship makes the transaction unusually personal, but the business rationale is straightforward. Bialecki is acquiring a senior product leader who understands both startup execution and the evolution of customer-facing software. Torres, in turn, is joining a company with a large installed base and the resources to distribute AI products more broadly than a young startup could manage independently.
In an emerging category, those distribution advantages can determine whether an agent becomes a meaningful business or remains a promising but limited application.
Klaviyo’s data advantage is real, but not automatic
Klaviyo’s argument rests partly on the value of its data foundation. The company serves e-commerce businesses and helps them manage marketing communications, customer segmentation, and engagement. Over time, its platform can accumulate information about customer behavior, purchasing activity, campaign responses, and interactions across a merchant’s operations.
That information could make AI agents more useful. A customer-service agent with access to an order record, shipping status, customer history, loyalty information, and applicable policies can do more than a generic chatbot. It can potentially identify the customer, understand the context, recommend an action, and execute that action through connected systems.
The advantage is not simply having more data. It is having data that is structured around a business workflow and available at the moment a decision must be made. An agent that knows a customer’s order status but cannot update the order is limited. An agent that can update the order but lacks the merchant’s return policy is risky. An agent that has both information and permission still needs safeguards to determine when a request should be escalated to a human.
Klaviyo’s integration with business workflows could help address those gaps. The company says it plans to combine Agency’s product with its existing agent efforts and offer the resulting capabilities to its customer base. If it succeeds, the company could provide a more comprehensive system than an independent customer-service startup that must integrate with dozens of marketing and commerce platforms.
However, data access is not the same as data quality. Enterprise systems often contain duplicate records, outdated fields, inconsistent policies, and incomplete histories. A company may have information about a customer distributed across its commerce platform, help-desk software, payment processor, warehouse system, and marketing database. Connecting those sources reliably is a technical and operational problem, not a marketing claim.
Klaviyo will also need to persuade customers that giving an agent access to sensitive business information creates more value than risk. Merchants may be willing to automate low-stakes questions while withholding permission for refunds, cancellations, discounts, or changes to shipping addresses. The commercial success of Customer Agent will depend on whether Klaviyo can support that graduated approach rather than forcing businesses to choose between full automation and no automation.
Customer service is the first major proving ground
Customer service is an attractive market for AI agents because the economics are visible. Businesses spend heavily on support teams, particularly during seasonal peaks and periods of rapid growth. Many incoming requests are repetitive: customers want to know where an order is, whether an item can be returned, how long delivery will take, or whether a payment was processed.
Automating those interactions could reduce response times and allow human agents to focus on complex cases. It could also make support available around the clock, which is valuable for merchants with customers in multiple time zones.
But customer service is also one of the most unforgiving environments for AI. A marketing assistant can generate a draft that a person reviews before it is sent. A customer-service agent often operates in real time, with a customer expecting a definitive answer. The cost of a mistake is not limited to an incorrect sentence. It can include a lost order, an unauthorized refund, a privacy incident, or a frustrated customer who takes the complaint to social media.
This creates a difficult balance for vendors. The more authority an agent receives, the more value it can potentially deliver. Yet each additional permission increases the consequences of failure. A system that can only explain a return policy is relatively safe. A system that can approve a return, issue a credit, and update inventory is much more powerful—and much more difficult to govern.
Klaviyo’s opportunity is to make that governance part of the product. Merchants will need controls that define which actions agents may take, which conditions must be met, and when a human must intervene. They may also need audit trails showing what the agent saw, what it decided, which systems it contacted, and why it took a particular action.
The companies that establish trust in these workflows could build durable relationships with customers. Those that focus only on reducing support headcount may discover that the savings disappear when automated mistakes generate refunds, rework, or customer churn.
Composer points to a wider platform strategy
Customer Agent is the most obvious example of a customer-facing application, but Klaviyo’s AI strategy also includes Composer, which helps businesses build marketing campaigns. Together, the products indicate that Klaviyo is positioning AI as a layer across the customer lifecycle rather than as a single chatbot.
Composer can help with campaign creation, while Customer Agent can address requests after a purchase. In theory, those systems could benefit from sharing context. Marketing messages could reflect a customer’s purchase history and preferences. Support interactions could inform future segmentation or campaign decisions. A merchant could eventually manage acquisition, engagement, and post-sale service through connected AI tools.
That vision would give Klaviyo a stronger competitive position than a vendor selling an isolated support agent. It would also raise the stakes of execution. The company would need to demonstrate that its products work together without creating irrelevant personalization, conflicting recommendations, or inconsistent customer experiences.
The platform approach is attractive because it can increase the value of existing customer relationships. Klaviyo already has distribution, billing relationships, and integrations with businesses that may prefer to buy additional functionality from a vendor they know. If AI features are embedded in the platform, adoption could be easier than it would be for a new company asking merchants to deploy another system.
This could also improve Klaviyo’s economics over time. Software companies generally benefit when they can sell more products to existing customers, because acquisition costs are lower than the cost of winning entirely new accounts. AI agents may create new consumption or usage-based revenue opportunities, although the company will need to manage the cost of model inference, data processing, and integration support.
The risk is that AI features could increase operating costs faster than they generate revenue. Every automated interaction consumes computational resources. Complex tasks may require multiple model calls, searches across business systems, and checks against policy rules. If pricing does not reflect that usage, higher adoption could pressure margins rather than improve them.
Klaviyo will have to decide whether to charge for agents as premium modules, price them based on usage, bundle them into existing plans, or combine those approaches. The choice will affect customer adoption and the predictability of the company’s revenue.
Incumbents versus AI-native challengers
Klaviyo is entering a field that includes AI-native customer-service companies such as Decagon and Sierra. These companies have an advantage that incumbents often lack: they can design their products around agents from the beginning. They do not need to preserve older workflows, reconcile competing product priorities, or convince internal teams to change established systems.
An AI-native company can also present itself as neutral across platforms. A merchant may prefer a specialist that integrates with its existing marketing, commerce, and help-desk tools rather than a vendor whose core business is focused on one part of the stack.
The challengers’ weakness is distribution and context. They must build integrations, learn each customer’s operating model, and prove that their systems can function across a wide range of businesses. They may have sophisticated agent technology, but that technology still needs access to reliable data and permission to act.
Klaviyo starts with the opposite profile. It has customer relationships and workflow data, but it must adapt its organization and products to a more autonomous model of software. Its historical strength has been helping businesses communicate with customers. The next phase requires it to make decisions and execute tasks on behalf of those businesses.
The competitive outcome may not be determined by which company has the best language model. Many vendors can access similar underlying models. Differentiation is more likely to come from integrations, deployment speed, reliability, controls, pricing, and the ability to quantify business results.
This is why the Agency acquisition could matter. Torres and his team may help Klaviyo close the gap between having AI capabilities and delivering an agent that works inside a customer’s actual business. But the acquisition will only create an advantage if Klaviyo allows the team to move quickly while providing access to the company’s broader resources.
Large software companies often acquire startups to accelerate innovation and then slow them down with layers of process. Klaviyo will need to avoid that pattern. Agency’s value is partly its startup operating model and product focus. Preserving those qualities inside a public company will be a management challenge.
The return of the founder network
The transaction also illustrates how relationships can shape competition in enterprise technology. Bialecki and Torres worked together at Performable, then pursued separate entrepreneurial paths before reuniting at Klaviyo. Their shared history may reduce the time required to establish trust and align on product decisions.
That matters in AI, where companies are making choices under considerable uncertainty. Leaders must decide which tasks to automate, how much autonomy to grant agents, how to price usage, and how aggressively to invest before customer demand is fully proven. A strong working relationship can make those decisions faster.
It can also improve recruiting. The AI market has made experienced product and engineering leaders scarce. A founder with multiple exits can attract employees, communicate a clear ambition, and navigate the transition from startup experimentation to enterprise distribution. Agency’s 25-person team provides Klaviyo with a compact group that already understands its own product and can now operate with larger resources.
Still, personal history cannot substitute for product-market fit. Agency’s prior funding and investor backing establish that the company attracted significant interest, but they do not by themselves demonstrate how broadly its technology has been deployed or how it performs under enterprise conditions. Klaviyo will need to show that the combined product can produce measurable improvements for merchants.
Those measurements could include faster resolution times, higher customer satisfaction, fewer escalations, reduced support costs, improved conversion, or greater retention. The most credible results will be tied to business outcomes rather than the number of conversations an agent handles.
AI-agent consolidation may arrive early
Klaviyo’s acquisition may also be an early signal of consolidation in the AI-agent market. Many startups have been able to raise capital and launch products because the underlying model infrastructure is increasingly accessible. But the market may not support dozens of independent vendors once customers begin demanding deep integrations, security controls, service guarantees, and global support.
The cost of enterprise sales is another pressure. Selling an agent to a large business can require lengthy security reviews, data-governance assessments, workflow customization, and executive approval. Startups with strong technology may struggle to fund that process independently. Being acquired by an established platform can provide access to customers and distribution that would otherwise take years to build.
For buyers, acquisitions are a way to acquire speed. Internal development can provide control, but it may take too long in a rapidly changing market. Buying a startup can bring specialized talent and a product that has already been shaped by external customers.
The danger is that acquirers may overpay for technology that becomes interchangeable. If the underlying models improve rapidly, a startup’s technical lead can narrow. The lasting value may instead lie in customer relationships, proprietary workflows, data permissions, and operational knowledge.
That makes Agency’s integration into Klaviyo especially important. The acquisition will be judged not only by whether Klaviyo can launch features, but by whether it can create a differentiated system that competitors cannot easily replicate.
The question is who controls the customer relationship
At the center of this competition is control of the customer relationship. A customer-service agent may become the primary interface between a shopper and a merchant. The platform that powers that interface could gain valuable information about customer intent, friction points, and purchasing behavior.
Marketing software companies want to use that information to improve campaigns. Commerce platforms want to use it to increase sales and retention. Customer-service specialists want to own the support interaction. Large technology companies may want to provide the model, infrastructure, and general-purpose agent layer underneath all of them.
The winner may not be the company that generates the most natural-sounding responses. It may be the company that becomes most deeply embedded in the chain of decisions surrounding a transaction.
Klaviyo has a credible path because it already sits close to merchants’ marketing and customer data. But it will need to demonstrate that its agents can operate across the messy boundaries between marketing, commerce, logistics, and support. Customers will not experience those boundaries as separate software categories. They will simply expect a correct answer and a completed task.
The acquisition of Agency gives Klaviyo additional talent and a stronger product-development base for that challenge. Torres’s move into the chief product officer role signals that the company sees AI as central to its product direction, not as a minor feature category.
The broader strategic bet is that a trusted business platform with customer context can build better agents than an isolated application. That bet is plausible, but it remains unproven. Data must be accurate, permissions must be carefully managed, and the economics must work at scale.
Klaviyo’s next phase will show whether its advantage lies in owning the data around customer interactions—or whether AI-native specialists can use superior focus and execution to capture the agent relationship first. The Agency acquisition is an effort to ensure that the incumbent does not surrender that opportunity while the market is still being defined.