Tray.ai

A demo room in San Francisco. A salesperson types a sentence. Across the room, a screen lights up: Salesforce, Slack, Snowflake, NetSuite, a half-dozen other tools - all moving in quiet, choreographed agreement. The thing doing the choreographing is not, technically, the AI. It is the layer underneath it. That layer is Tray.ai.

01 / WHO THEY ARE NOWThe unglamorous middle of the AI stack

Tray.ai is an enterprise integration and automation platform - an iPaaS, in industry shorthand - that has, over the last two years, repositioned itself as the connective tissue for AI agents. The product connects more than 700 enterprise systems, runs the workflows between them, and now ships an agent builder called Merlin that lets non-engineers stand up AI agents that actually do things across that stack.

It is the kind of company whose name a chief information officer can pronounce without flinching. It does not promise sentient assistants or general intelligence. It promises that the agent your team built last Tuesday will still talk to Workday on Wednesday. Which, in 2026, turns out to be the harder problem.

The CEO's favorite sentence, repeated until it stuck.

02 / THE PROBLEMAgents without plumbing

For most of 2023 and 2024, the AI conversation focused on models. Bigger ones, smaller ones, cheaper ones, faster ones. The conversation Tray.ai's customers were having behind the scenes was less photogenic: how do we get this thing to read from our CRM, write to our ticketing system, follow our approval rules, and not embarrass us on a Friday afternoon?

That is the gap Tray.ai bet on. An LLM can draft a beautiful response. It cannot, by itself, look up the customer's order history, check warranty status, refund the right line item, and post a note in the right Slack channel - not without something deeply unglamorous in between. Pipes. Permissions. Auth tokens. Audit logs. The very things software engineers usually try to avoid talking about at dinner.

The story Tray.ai's leadership tells is that the agent era's biggest blocker is not the agent. It is the integration debt accumulated over a decade of SaaS sprawl. Most enterprises now buy 200 to 400 SaaS tools. Stitching them together used to be a quarterly engineering project. Now it is the prerequisite for any AI strategy at all.

03 / THE FOUNDERS' BETThree Londoners, one pivot

The company was founded in London in 2012 by Rich Waldron, Alistair Russell, and Dominic Lewis. The original Tray was a consumer product - a tool that pulled a person's data together from various online services. Charming. Hard to monetize. After a stint at the AngelPad accelerator in New York, the founders did the thing most ambitious founders eventually do: they looked at the unglamorous business problem next door and realized it was worth more.

By 2015 the pivot was complete. The consumer dashboard was gone. In its place: a low-code workflow builder aimed squarely at enterprise IT and revenue operations. The bet was that companies would buy automation if it did not require an engineer to maintain. They were right, mostly. They were also early - which is the polite way of saying they had to spend a few years explaining what iPaaS even was.

A decade later, all three co-founders are still associated with the company. That is rare in venture-backed software. It is also, arguably, why the product still has a coherent shape rather than the patchwork look of a platform built by committee.

04 / THE PRODUCTWhat you can actually do with it

Stripped of marketing, the platform has four jobs.

Connect 700+ apps without writing connectors

Pre-built integrations to the systems most enterprises already run - Salesforce, Workday, NetSuite, ServiceNow, Snowflake, HubSpot, Slack, Microsoft 365, the lot. A connector SDK for the long tail nobody else has built yet.

Automate workflows visually

Drag-and-drop logic, not YAML. The kind of thing a revenue operations lead can actually maintain. The kind of thing that, in less polite circles, used to be called Zapier-for-grown-ups - though Tray sits closer to Workato in the enterprise market.

Build and run AI agents (Merlin)

Launched in December 2024, the Merlin Agent Builder lets teams compose agents from the same building blocks the platform already uses for workflows. Agent Accelerators ship as templates: an SDR agent, an IT support agent, a finance ops agent. You add the data and the guardrails. The thing runs.

Govern everything in one place

An MCP-aware gateway sits between agents and tools. Logs, permissions, rate limits, audit trails. The part that makes a security team breathe out for the first time all quarter.

05 / THE PROOFReceipts, in numbers

The pitch is only persuasive if the numbers cooperate. Mostly, they do.

Tray.ai customers tend to talk less about throughput and more about avoidance: the IT projects they did not have to scope, the integrations engineering did not have to staff, the agents that did not have to be ripped out six weeks after launch because they could not see the rest of the stack.

The recognition has caught up. The company has been named a Visionary in the Gartner Magic Quadrant for iPaaS for three years running, and a Leader in the 2025 iPaaS Technology Value Matrix by Nucleus Research - the kind of analyst stamps that make procurement departments stop asking awkward questions.

06 / THE MISSIONComposable, governed, and unglamorous

Ask the company what it's trying to do and the answer is consistent in a way that suggests they have practiced it. The mission is to make every team faster by connecting their software and orchestrating the agents that act on top of it. The vision is a composable enterprise, where integration, automation, and agents share one governed platform instead of three competing ones.

It is not a mission designed for a keynote. It is the kind of mission that wins five-year contracts.

The cultural tell is in the hiring. The company has stayed remote-friendly without becoming remote-fragile, leans engineering-heavy without becoming engineer-only, and has kept the same three founders in the building for more than a decade. In a category that has watched founders rotate out at the first hint of a Series D, that continuity is its own kind of moat. The result is a product roadmap that reads less like a deck and more like a list of things customers actually asked for - in order, with dates.

There is something almost old-fashioned about the bet. While much of the AI industry has spent the past two years chasing the most dramatic possible demo, Tray.ai has been quietly insisting that demos are not the point. Production is the point. And production, in an enterprise, is mostly integration, governance, and the willingness to call a workflow boring without flinching.

07 / WHY IT MATTERS TOMORROWThe plumbing wins

Most enterprise AI projects in 2025 stalled at the same place: not the model, not the use case, but the integration layer underneath. The companies that figured it out early - the ones with mature iPaaS and clean connectors - shipped agents that worked. The ones that skipped the layer ended up with extremely articulate chatbots that could not check stock levels.

The Vanti acquisition, closed in March 2025, doubled down on the AI side of the platform. The Merlin Agent Builder kept shipping refinements through 2025 and into 2026, including a major release that focused, with refreshing specificity, on the gap between building an agent and getting employees to actually use it. That gap is the next problem. Tray.ai has clearly decided to own it.

Back to the demo room in San Francisco. The salesperson types another sentence. Behind the screen, the same hidden choreography runs - Salesforce, Slack, Snowflake, NetSuite. The agent looks fast and clever. The thing that makes it fast and clever is, still, the layer underneath. That layer has a name now, and a Gartner mention, and four thousand customers. It is no longer hidden. It is just, finally, taken seriously.