Gumloop.

A Company That Sells the Idea That You Should Not Have to Be a Programmer

Gumloop wants every employee to build their own AI agents. Benchmark wrote a $50 million check to find out if they will.

Here is a fact about modern work that is both obvious and slightly insulting once you say it out loud: an enormous amount of what people do at their jobs is the same thing, over and over, forever. Someone pastes a lead from an email into a spreadsheet. Someone reads a support ticket and decides which bucket it goes in. Someone checks a dashboard every morning and copies three numbers into a Slack message. These are not hard tasks. They are, in the specific and technical sense, tasks a computer could do - and the reason a computer wasn't doing them is that telling a computer how to do them required being a programmer, and most people are not programmers.

Gumloop's entire business is the wager that this is about to stop being true. The company builds a no-code platform where you drag boxes onto a canvas, connect them with lines, and out the other end comes an automation that reads your email, or scrapes a website, or routes a support ticket, or enriches a lead - the boring middle of your job, running on its own. What makes it a 2020s company rather than a 2010s one is that the boxes contain AI. You can drop in a node that says, roughly, "read this and figure out what the customer wants," and it will, because there is now a model that can.

That comparison is doing a lot of work, and it is worth pausing on, because it is also the whole competitive argument. Zapier - along with Make, n8n, and a dozen others - won the no-code automation market years ago. They connect App A to App B when Event C happens. In that world, AI is a feature: one more app you can wire into the chain, one more node among hundreds. Gumloop's pitch is that this gets the architecture backwards. If AI is the thing that can actually make decisions - the thing that replaces the human judgment in the middle of the workflow, not just the plumbing - then AI orchestration should be the core primitive, and everything else should be built around it. You don't add intelligence to an automation tool. You build an automation tool around intelligence.

The Bedroom, the Discord, and the Denied Visa

Origin stories are usually cleaned up in retrospect. This one is stranger than the cleaned-up version.

Gumloop was founded in 2023 - originally under the name AgentHub - by Max Brodeur-Urbas and Rahul Behal, classmates from McGill University. Brodeur-Urbas had been an engineer on Microsoft's Azure Linux team; Behal had done time at Amazon's AWS and at a machine-learning operations role. The founding moment, as it gets told, involves Brodeur-Urbas being denied re-entry to the United States and, stuck abroad, watching the Discord server for AutoGPT - one of the first viral autonomous-agent projects - fill up with people asking things like "what is GitHub?" and "how do I use a terminal?"

That is a more useful origin story than most, because it contains an actual insight rather than just a vibe. The lesson Brodeur-Urbas took from the confused Discord was not "AI agents are the future," which everyone already believed. It was narrower and more actionable: the technology had arrived, and the interface had not. There were thousands of people who understood exactly what they wanted to automate and had no way to tell a computer to do it. The gap between wanting and doing was the product. Everything Gumloop has built since - the drag-and-drop canvas, the obsession with non-technical users, the thousands of customer calls and hundreds of workshops the team ran - is downstream of that one observation.

The money followed the way money follows when a thesis starts working. A $3.1 million seed round led by First Round Capital in July 2024, on the back of a Y Combinator Winter 2024 batch. A $17 million Series A led by Nexus Venture Partners in January 2025. And then, in March 2026, the round that made people pay attention: $50 million, led by Benchmark, the firm whose brand is built on writing a small number of very large checks into companies it thinks will define a category. The partner leading it, Everett Randle, had joined Benchmark only months earlier. Nexus, First Round, Y Combinator, Box Group, The Cannon Project, and - tellingly - Shopify Ventures all came along. When one of your customers' venture arms invests, it is a slightly louder vote of confidence than a term sheet from a stranger.

What You Can Actually Do With It

Strip away the funding narrative and the product is concrete. A revenue team builds a pipeline that takes a raw list of companies, enriches each one with data pulled from the web, scores them, and drops the good ones into the CRM - no analyst copy-pasting for an afternoon. A support manager deploys an agent into Slack or Microsoft Teams that reads incoming tickets, categorizes them, drafts responses, and escalates the genuinely weird ones to a human. An operations person automates the morning report they have written by hand two hundred times. A marketer wires together web scraping, content generation, and publishing. None of these people write code. All of them, in the Gumloop telling, ship something that used to require asking engineering for a favor and waiting a quarter.

The platform now spans three pieces. There is the core Gumloop canvas, with its 130-plus native integrations - Slack, Gmail, Salesforce, HubSpot, Notion, Airtable, GitHub, Linear, Stripe, Google Workspace, and on - plus custom nodes, code sandboxes for the people who do want to write a little logic, and parallel execution for jobs that need to run at scale. There are Gumloop Agents, the proactive kind that live in your chat tools and act on their own when something happens rather than waiting to be run. And there is Gumstack, the less glamorous but arguably more important enterprise layer: audit logs, role-based access controls, monitoring of exactly what data the agents are touching, SOC 2 Type II certification, and virtual-private-cloud deployment for companies that will not let their data leave the building.

That last part is worth dwelling on, because it is where a lot of AI-tooling stories quietly end. It is easy to build a demo where an agent does something impressive. It is hard, boring, and expensive to build the governance that lets a Fortune 500 security team say yes - to answer "who can see what," "what happens when an agent misbehaves," and "prove to our auditors this is safe." Gumloop appears to have done the unglamorous work, and the enterprise adoption suggests it is paying off. When you give a thousand employees the power to build agents that touch company data, someone has to be able to watch the agents. Gumstack is Gumloop admitting that, which is a more mature thing to sell than pure magic.

There is one more thing about Gumloop that is either a marketing flourish or a genuine statement of belief, and it is that the company openly talks about wanting to build a billion-dollar business with a tiny team - a headcount you could count on a few hands. Roughly 42 people work there now. In most eras that ambition would read as bravado. In this one it reads as the thesis eating itself in a good way: a company whose entire product is "let a small number of people do the work of many" would be slightly embarrassing if it needed a thousand employees to do it. Gumloop is, in a sense, its own first customer. Whether it becomes a billion-dollar company is unknown, and the competition is real, and the category is loud with well-funded companies making similar promises. But the shape of the bet is unusually clean: if understanding a task really is the only thing standing between people and automating it, Gumloop is selling the shortest path across that gap - and a lot of serious people have now put serious money on the crossing.