Ampersand raises $15M to connect enterprise agents to customized systems

Why it matters

Enterprise AI vendors need dependable access to customer-specific business systems, and the integrations often require ongoing engineering work. Ampersand is raising capital to own that layer, where reliability matters as much as access.

Ampersand raised a $15 million Series A led by Bessemer Venture Partners, betting that AI agents will need a durable connection to the customized CRMs and ERPs businesses already rely on. The October 6th announcement says existing investors Matrix and Flex Capital also participated, alongside new investors Yelp, Tenacity Capital, CTO Fund and Mana Ventures. Bessemer partner Lauri Moore will join the board.

For co-founders Ayan Barua (@ayanb) and Lauren Long (@laurenzlong), the round extends a problem they have worked on from two different sides of the software stack. Barua saw integrations consume engineering capacity at Siftery, the software-discovery company he co-founded and later sold to G2, where he became vice president of engineering. Long led Cloud Functions and the launch of Firebase Extensions at Google, building tools that connected apps with outside services. Ampersand puts that experience to work on a harder version of the same task: connecting a software vendor's product to each customer's individually configured business systems.

The financing follows a $4.7 million seed round announced in April 2023 and led by Matrix, according to TechCrunch's coverage. The two announced rounds total $19.7 million.

The integration problem moves into the agent runtime

Ampersand's case is that a working demo is easy compared with keeping an agent's access reliable once it reaches production. A customer may have custom fields, objects, permissions and workflows in Salesforce or an ERP. An agent needs to interpret those differences and make changes without losing access when credentials expire or a system's schema changes.

Moore's statement in the announcement makes the distinction directly: protocols such as MCP can describe a tool, while the recurring work of keeping customer-specific connections running remains. That framing places Ampersand in the operational layer around agent software. Its platform is designed to let agents read and write data inside those systems, with the customer-specific configuration intact.

The problem is familiar from Ampersand's earlier pitch to software developers. In a LinkedIn post, Barua described custom integrations as a persistent burden at both Siftery and G2. Ampersand says Barua and Long spoke with more than 100 companies building integrations before writing its first line of code. That figure comes from Ampersand; it describes customer discovery, not adoption or revenue.

The distinction from a broad connector catalog is depth. Ampersand says each end customer's data model can account for custom fields, permissions and workflows, allowing software vendors to build two-way actions into their products. That can help a vendor serve a customer whose configuration would otherwise require bespoke engineering. It also leaves Ampersand with the maintenance work: keeping those connections functioning as customer setups and third-party systems change.

A founder's integration thesis gets an agent test

Barua has described Ampersand as a continuation of his work on how businesses select and manage software. In an interview with Sacra, he recalled that a substantial share of Siftery's roadmap went into integrations, including getting data from business systems and writing it back. Ampersand's premise is that enterprise AI will multiply that demand: every new agent product still has to reach the systems where a customer's operational data lives.

Long brings a complementary product-building background. At Firebase, she led Cloud Functions and Firebase Extensions, a role that involved connecting apps to third-party services, according to her speaker biography. The founders' combination of enterprise integration experience and developer-platform work helps explain Ampersand's focus on making custom connections part of a software product rather than a separate project for each customer.

Ampersand also announced the beta release of Andi, an AI integration agent intended to help developers with implementation work in each customer's environment. The release gives Ampersand a second layer to its pitch: its own software will help developers do the work of configuring and deploying the integrations that let other agents operate. The announcement does not establish how much of that work Andi can complete without human engineering support.

Ampersand also cites a customer example: John Pena, CTO of AI communications platform Hatch, said the product helped Hatch serve customers it could not previously support because of their integration requirements. The announcement identifies Hatch as acquired by Yelp in 2026, and Yelp is also a new investor in Ampersand's Series A. That account is a testimonial supplied by Ampersand, not a disclosed measure of how widely the product is deployed.

The Series A funds a focused bet: as AI vendors pursue enterprise customers, customer-specific integration work may become a bottleneck that a third party can own. Barua and Long have built around the claim that agents are only useful when they can reliably reach and act on business data. The demanding proof will be operational: whether Ampersand can keep those connections accurate across customers while the agents using them are allowed to write into systems businesses cannot afford to have changed incorrectly.