It is a Tuesday afternoon in San Francisco, and somewhere inside a building on Mission Street, an engineer is not writing the OAuth flow for the seventeenth payroll system this quarter. That is the entire point of Merge.
Who they are, right now
Merge is the company that decided integrations should be a feature, not a quarterly tax. Founded in 2020 by two college friends from Columbia, it now sits in a strange and useful position: too infrastructural to be flashy, too central to ignore. Its software powers the connective tissue of more than 7,000 B2B products - the boring, indispensable wiring between a SaaS app and the HR system, applicant tracker, ticketing tool, or accounting ledger its customers happen to use.
If you have onboarded an employee through Ramp, run a compliance audit on Drata, or scheduled a candidate on RippleMatch, there is a strong chance a Merge connector was somewhere in the loop. Most users will never know. That, in this category, counts as a compliment.
The problem they saw
Every B2B SaaS company hits the same wall around the same revenue mark. It usually arrives in the form of a procurement email: "We will purchase your product. We just need it to talk to Workday." Or BambooHR. Or Greenhouse. Or NetSuite. Or all four, depending on which division is signing the contract.
Each integration is its own snowflake. Different auth schemes, different rate limits, different ways of pretending the field "employee_id" is the same field as "employeeID". Building one connector takes weeks. Maintaining a hundred takes a department. Companies usually choose one of two losing strategies: hire an integrations team and watch margins erode, or politely lose the deal to whoever did.
Shensi Ding and Gil Feig spent their first jobs out of Columbia watching that wall claim other people's roadmaps. Ding was working at Expanse, a cybersecurity company that needed to push tickets into half a dozen ITSM platforms. Feig was at LinkedIn, then at other software companies, with a front-row seat to the same recurring tax. They compared notes over a meal, agreed it was absurd, and quit to build the alternative.
The founders' bet
The bet was deceptively simple. Hide every HRIS behind one HRIS-shaped door. Behind that door, normalize the data into a single schema - so a developer writes one query and gets a clean answer whether the underlying system is Workday, Gusto, or a Bavarian payroll provider you have never heard of. Repeat for ATS, CRM, accounting, ticketing, and so on. Build the boring middle, and let the customer believe they integrated with everything in an afternoon.
The hard part, of course, is the middle. The hard part is always the middle. A single category like HRIS has thirty serious vendors, each with its own concept of what a "compensation history" entry should contain. Merge's actual engineering work, much of it invisible, is the schema design and the unglamorous rate-limiting, retry logic, and webhook-deduplication that keeps the abstraction from leaking.
Founder-led, customer-obsessed
One detail worth lingering on: the early team ran sales themselves, for years. Not because they could not afford to hire reps - the seed round from NEA closed in 2021 - but because the founders believed they would learn faster doing the calls themselves. First Round Review later wrote up the strategy approvingly. The company's culture still tilts toward that posture: build the thing, then carry the pager.
The product, in three doors
Merge Unified API
The original product. One API surface, hundreds of integrations across HRIS, ATS, CRM, accounting, ticketing, marketing automation, and file storage. Data is normalized, authentication is handled, webhooks arrive in a consistent shape. The developer writes the integration once and ships to many.
Merge Agent Handler
Launched in October 2025. A platform for connecting AI agents to thousands of third-party tools - securely, observably, and with auth flows that do not embarrass the legal team. Built around the Model Context Protocol, with centralized governance, real-time logs, and proactive alerts when an agent does something it shouldn't.
Integration Observability
The dashboards, audit logs, and webhook management that customer-success and engineering teams use to actually see what is happening across a hundred customer-facing integrations. The kind of feature you do not realize you needed until your largest customer pings on a Saturday.
Two products live in production. A third (Gateway, for LLM routing) has been quietly mentioned in interviews. Translation: a busy roadmap.
A timeline, lightly annotated
The proof
You can argue with a thesis. Numbers are harder.
The customer roster reads like a tour of B2B SaaS in its current decade: Ramp, Drata, AngelList, Gong, Calendly, TripActions, Apollo, RippleMatch. Public reporting suggests several large AI labs and a major streaming service now use Agent Handler to give internal agents reach into Slack, Gmail, and Notion. The pattern is consistent. When a company's product becomes "the integration" inside someone else's workflow, Merge tends to be the thing under the floorboards.
Funding trajectory (USD millions)
Three rounds, two years, one obvious slope. The Series B alone is bigger than most companies' lifetime raises in this category.
The mission
Merge describes its mission, with the understatement common to infrastructure companies, as "making secure data access easy." That phrase does a lot of quiet work. Secure, because the company sits inside enterprise data pipes. Data access, because integrations are mostly a data problem dressed up as an engineering one. Easy, because the entire premise collapses if the abstraction is harder than the thing it abstracts.
The Agent Handler launch is a tell. It says the founders believe that the next decade of integrations will not be human-initiated at all. The customer of the future is not a procurement officer demanding a Workday connector - it is an AI agent that needs to read the calendar, post the message, and update the CRM, in that order, without supervision. Whoever owns the safe middle layer in that exchange owns a very interesting business.
Why it matters tomorrow
There are two ways for a company like Merge to fail. The first is that the platforms it integrates with build acceptable native solutions, and the abstraction stops earning its keep. The second is that AI agents and protocols like MCP make integration so cheap that no one needs a unified API at all - the agent can just figure it out.
Merge's bet is that neither will happen quickly, and that in the meantime, somebody has to be responsible. Somebody has to know whether the credential rotated last night. Somebody has to redact the PII before it leaves the tenant. Somebody has to keep the audit log. That role - the trusted middle, with paperwork - is unglamorous, defensible, and increasingly the entire game.
The opening scene of this dossier had an engineer not writing an OAuth flow. The closing scene is the same engineer, but their company has now shipped integrations into ninety customer accounts this quarter, and their roadmap actually has features on it. The wall is still there. Merge just rented it out.
Filed from a Tuesday afternoon, somewhere near Mission Street.