The archive · Developer & Business Tools · Product decision · 2023–2026
Gumloop bets employees, not engineers, build AI agents; Benchmark leads $50M
Two ex-Big Tech developers built an agent builder for nontechnical staff and planned a 10-person billion-dollar company — then demand forced a sales force.
Gumloop
What the business is
A no-code platform that lets employees build AI agents and workflow automations by dragging modular components onto a canvas, with prebuilt pipelines and integrations to tools like Gmail, GitHub and X.
Starting capital：$20M raised by January 2025 — a $17M Series A led by Nexus Venture Partners with First Round Capital, Y Combinator and angels including Instacart co-founder Max Mullen — then a $50M Series B led by Benchmark in March 2026: roughly $70M total.
How it started
Max Brodeur-Urbas (ex-Microsoft) and Rahul Behal (ex-AWS) started Gumloop in mid-2023 as a side project in a bedroom in Vancouver, solving a simple automation problem for nontechnical people in a Discord server. What began as a wrapper around the open-source Auto-GPT became a drag-and-drop workflow builder. The founders were explicit about the limits: AI left to run whole workflows would 'spin its wheels,' so the product composes narrow, trustworthy modules instead.
What happened
By January 2025 Gumloop had raised $20M, claimed thousands of users and customers including Instacart and Rippling, and declared it would stay a 10-person, billion-dollar company — 'Using AI to code let us have the throughput of a 20-person team.' A year later, enterprise demand broke that plan: in March 2026, Benchmark general partner Everett Randle (in his first deal at the firm) led a $50M Series B with Shopify participating; customers by then included Shopify, Ramp, Gusto, Samsara, Instacart and Opendoor, and Gumloop said it was building a dedicated sales force and scaling engineering.
How it ended up
The Series B happened even though Gumloop 'wasn't actively seeking capital' — the founders decided 2026 was the year to 'step on the gas.' Randle's due diligence found organic pull: one customer gave employees Gumloop alongside two competitors and six months later found staff using Gumloop daily while the rivals sat untouched. The company then abandoned its tiny-team vow, betting that being model-agnostic — 'plenty of enterprises have OpenAI, Gemini, and Anthropic credits. They want to use all of them' — is the moat against Zapier, n8n, Dust and Anthropic's Claude Cowork.
Background
Gumloop's founding bet was contrarian twice over: that the people who will build AI agents in companies are nontechnical employees, not engineers, and that AI-powered automation works best when humans assemble small, reliable modules instead of letting a model run a whole workflow unsupervised. Co-founders Max Brodeur-Urbas (ex-Microsoft) and Rahul Behal (ex-AWS) started in a Vancouver bedroom in mid-2023, solving a problem for nontechnical users in a Discord server; the product grew out of a wrapper around Auto-GPT.
The discipline showed in the economics. By January 2025 the company had raised $20M, claimed thousands of users and customers like Instacart and Rippling, and announced it intended to stay a 10-person, billion-dollar company, saying AI-assisted coding gave it 'the throughput of a 20-person team.' Venture money was a means, not a goal: 'We didn't need the money at all,' Brodeur-Urbas said of the Series A.
A year later the bet had outgrown its plan. In March 2026 Benchmark's Everett Randle — his first deal since joining from Kleiner Perkins — led a $50M Series B with Nexus Venture Partners, First Round, Y Combinator, BoxGroup, The Cannon Project and Shopify participating. The customer list had broadened to Shopify, Ramp, Gusto, Samsara, Instacart and Opendoor, and the founders were building a sales force and scaling engineering — the '10-person' plan died of success. Randle's evidence: a customer gave employees Gumloop plus two competitors; six months later only Gumloop had stuck.
The competitive argument is model-agnosticism. Gumloop lets customers route each workflow to the best model at the moment, which matters when 'plenty of enterprises have OpenAI, Gemini, and Anthropic credits' — but it also puts Gumloop between credit-rich enterprises and every major lab, against both established automation platforms (Zapier, n8n) and agentic rivals like Anthropic's Claude Cowork.
What has to be true
- Gumloop sold narrow, composable modules instead of 'AI does the whole job,' which is what makes risk-averse enterprise users trust it enough to pay.
- Betting that nontechnical staff would become the builders turned every user into a potential creator of new agents, so internal automation compounded without engineers in the loop.
- Staying model-agnostic let enterprises spend their existing OpenAI, Gemini and Anthropic credits through one surface — a buying reason competitors tied to a single lab can't match.
- It raised on demand, not schedule: a small Series A with 'we didn't need the money,' then a large Series B only after organic enterprise adoption proved the category.
- AI-assisted coding let the team run at 2x headcount with 10 people, forcing product focus — a constraint the founders only gave up when customers made sales capacity the bottleneck.
What can be applied
When a product's marginal maker is a nontechnical employee rather than an engineer, adoption compounds on its own — and the demand it creates is the argument for breaking your small-team vow.
Aftermath
As of September 2026, Gumloop is scaling: ~$70M raised — a $17M Series A and Benchmark's $50M Series B — customers including Shopify, Ramp, Gusto, Samsara, Instacart and Opendoor, and a sales force built after years of running tiny. Its bet: a model-agnostic builder outlasts rivals tied to single labs, and 'enterprise automation' — 'the biggest category in enterprise AI,' per Randle — stays big enough for an independent layer between companies and labs. The shared risk: foundational labs may eventually bake agent-building into their products, leaving model-agnosticism the only defense.
Sources
- Gumloop, founded in a bedroom in Vancouver, lets users automate tasks with drag-and-drop modules
- Gumloop lands $50M from Benchmark to turn every employee into an AI agent builder
spotted an error? The archive wants to know.
Your turn
You just read one. Describe what you are building, and see who is betting on the same thing.
Free account · 3 free questions · no card