Ampersand Raises $15M for Enterprise Agent Integrations

Years before enterprise agents became their own software category, an engineer told Ayan Barua that he was not doing his job. At Siftery, the product roadmap was being pulled apart by CRM and ERP integrations that looked secondary until they started costing revenue and consuming the engineering team.

That operating lesson now sits behind Ampersand's $15M Series A, announced October 6, 2026 and led by Bessemer Venture Partners. Existing investors Matrix and Flex Capital participated, while Yelp, Tenacity Capital, CTO Fund, Mana Ventures, and unnamed angel investors joined the round. Bessemer partner Lauri Moore is also joining Ampersand's board.

The financing matters because enterprise AI has moved from answering questions to taking actions inside systems that were never standardized across customers. Ampersand is building the integration runtime between those agents and the CRMs, ERPs, and other systems of record that still carry the data, permissions, and workflows of the business.

What Happened

Ampersand says the Series A brings its total funding to $20M. The company previously announced a $4.7M Seed round in April 2023 led by Matrix, with Base Case Capital, Flex Capital, 2.12 Angels, and individual angels participating. No valuation was disclosed for the new round.

Barua, Ampersand's co-founder and CEO, previously co-founded Siftery and later served as VP of Engineering at G2. Co-founder and CTO Lauren Long built Firebase Extensions at Google. The founders say they interviewed more than 100 companies before writing Ampersand's first line of code, testing whether the integration burden they had experienced was a local problem or a structural one.

Ampersand is based in San Francisco and says it has spent three years building alongside customers including 11x, Orb, and Square. Bessemer's founder interview reports that 90% of teams completing an Ampersand proof of concept become paying customers. That conversion figure is company-reported, but it identifies the commercial point of the platform: integration depth can determine whether an enterprise pilot reaches production.

Why Agents Make the Old Problem Harder

A software vendor can build a Salesforce connector and still discover that each customer's Salesforce behaves like a different product. Objects have been customized, fields renamed, permissions tightened, approval rules layered, and workflows shaped by years of operating decisions. The software logo stays the same while the implementation carries the identity of the business.

Traditional integration work already consumed engineering time during onboarding and maintenance. Agents raise the stakes because they read and write continuously. An expired credential, changed schema, or incorrect field mapping can interrupt a production workflow, and a bad write can damage a pilot or renewal rather than merely create another support ticket.

Ampersand's position is that this variation should not be flattened away. It treats customer-specific structure as permanent and meaningful, then gives developers a way to define the intended behavior once while adapting it to each customer at runtime. That is a more demanding promise than connecting an API in a demonstration, because the system has to keep working after credentials, schemas, and customer configurations change.

How Ampersand's Runtime Works

Ampersand organizes integrations around five primitives: Read, Subscribe, Search, Write, and Proxy. Developers define the integration in version-controlled configuration through amp.yaml, including the objects, fields, and mappings the product needs. Ampersand's runtime handles authentication, token refreshes, retries, backfills, monitoring, and other machinery required to keep the connection running.

The company also lets a customer's administrator map fields inside the software vendor's onboarding experience. That shifts part of the implementation knowledge toward the person who understands the customer's configuration while keeping the integration logic visible to the vendor's engineering team. Ampersand argues that the result can ship with the product instead of arriving as a separate services project after the deal closes.

Ampersand has released Andi in beta as an AI integration agent. The company says Andi can help build connectors, advise on architecture, and investigate logs, but the important boundary is that it operates against Ampersand's deterministic primitives, configuration, observability, and validation layer. The agent assists with implementation while the runtime retains the controls needed for production infrastructure.

What Bessemer Is Underwriting

Bessemer's investment thesis is that legacy systems of record remain one of the strongest barriers between agentic software and enterprise adoption. The firm describes per-customer implementation cost as part of the moat protecting entrenched systems, because a vendor has to understand both the platform and the customer's particular configuration before an agent can act safely.

That makes Ampersand's market broader than a library of connectors. The company is trying to own the operational layer that keeps those connections alive across many customers. Hatch CTO John Pena said Ampersand enabled the company to serve customers previously blocked by complex integration requirements, a company-distributed testimonial that illustrates how infrastructure can expand the reachable market for an AI vendor.

The investor group also connects the round to both software infrastructure and the customers using it. Matrix and Flex Capital returned, Bessemer led, and Yelp joined after acquiring Hatch in 2026. Ampersand still has to prove its approach across the full variety of enterprise systems, while the financing gives the company more room to build where software architecture and forward-deployed expertise have to meet.

What the $15M Changes

Ampersand has named three priorities for the new capital. First, it plans to expand from deep CRM coverage into ERPs and industry-specific systems used in healthcare, insurance, logistics, financial services, and other verticals. Those environments bring their own APIs, permissions, scripting languages, data models, and implementation histories.

Second, the company will continue building Andi so engineering teams can create what Ampersand calls an integration factory rather than generating one-off code for each customer. Third, it will grow a forward-deployed team with experience in systems such as Salesforce, NetSuite, SAP, Workday, ServiceNow, Oracle, and Epic. Ampersand's current careers page lists multiple full-time hybrid roles in San Francisco.

The combination matters because difficult integrations are not purely a code-generation problem. Someone still needs to understand why a particular field exists, how an approval process works, which write carries risk, and what has to remain observable when the customer's environment changes. Ampersand is funding software and human domain expertise together because enterprise infrastructure eventually has to survive the enterprise.

The Operating Question Ahead

Ampersand now has to show that its runtime can handle the depth and variety of major systems across a growing customer base without recreating the services burden it is meant to reduce. The company has reported strong proof-of-concept conversion and named credible customers, but it has not disclosed valuation, revenue, audited growth, or an independent measure of market share.

Agent developers want to sell outcomes, yet those outcomes depend on reaching the systems where enterprises record customers, invoices, approvals, cases, employees, and inventory. Ampersand is building the handoff between an agent's reasoning and the customized operational record underneath it, and the new capital puts more of that handoff inside the product roadmap rather than inside the next customer's implementation calendar.