Pipedream

Glue code is a tax nobody agreed to pay.

Every modern app is really a switchboard. It talks to a payments API, a messaging API, a data warehouse, a half-dozen SaaS tools that marketing signed up for without telling anyone. Connecting them is unglamorous, repetitive, and weirdly time-consuming. Developers call it glue code, and they mean the word "glue" with some contempt.

The trouble is that glue code never stays small. First you write the integration. Then you write the error handling. Then you find somewhere to run it, monitor it, secure the credentials, and rotate the tokens. A two-hour job becomes a small piece of permanent infrastructure that one engineer now owns forever, usually against their will.

No-code tools tried to solve this by removing the code, which works right up until you need the code. Traditional integration platforms solved it for the enterprise buyer, which works right up until a developer wants to actually customize something. There was a gap in the middle: a tool that respected developers enough to let them write code, and respected their time enough not to make them run it.

Bet on the developer, not the buyer.

In 2019, Tod Sacerdoti and a group of former BrightRoll colleagues started Pipedream. Sacerdoti had done this rodeo before - BrightRoll was a video-advertising company Yahoo bought for around $640 million - so the founding team knew something about plumbing data between systems at scale, and about how much of an engineering org's time disappears into it.

The bet was contrarian for its moment. Most integration companies were chasing the enterprise buyer with procurement decks and per-seat pricing. Pipedream aimed one level down, at the engineer who would adopt a tool on a Tuesday afternoon because it solved their immediate problem. Make that person 10x faster, the thinking went, and the company they work for follows.

It was also a bet on openness. The integration registry - the catalog of triggers and actions - was open source. The community could read it, fork it, and extend it, which meant the catalog grew faster than any single team could build it. Pipedream was, in part, selling a library its own users helped write.

Drag and drop, or write the code. Your call.

The reason developers tolerate - then like - Pipedream is that it never forces a choice between simple and powerful. A workflow starts with a trigger: a webhook, a schedule, an event from one of thousands of apps. Then you chain steps. Each step can be a pre-built action or a block of custom code with full package access. Need to remember something between steps? Data Stores handle that without you standing up a database.

The pricing tells the same story as the product. There is a genuinely usable free tier for developers, paid plans (roughly $45 and $74 a month) metered on credits, active workflows, and AI tokens, and a Connect tier from about $150 a month for companies embedding integrations in production. Adopt it alone, expand it as a team. The classic developer-first funnel, run correctly.

When AI agents needed hands, Pipedream had them.

For years Pipedream was infrastructure most people never saw - the layer underneath someone else's product. Then 2025 happened, and large language models started wanting to take actions, not just answer questions. An agent that can talk but cannot click is a very expensive parrot. The missing piece was a safe, standard way for a model to call thousands of real apps.

That piece was the Model Context Protocol, and Pipedream already had the hard part built. Its registry of 3,000+ apps and 10,000+ tools became MCP servers almost overnight. Suddenly an AI agent could read your Slack, file a Jira ticket, update HubSpot, and process a Recurly charge - through one connection. Pipedream MCP shipped inside OpenAI's ChatGPT and Agent Builder, with OAuth and static URLs for production use.

Credentials matter when agents start acting on their own, which is why the SOC 2 Type II and HIPAA certifications stopped being a checkbox and became a selling point. Then Workday - mid-transformation from HR software into an enterprise AI platform - looked at its agents and realized they needed to act across Slack, Asana, Jira, and thousands of other apps. In November 2025 it signed a definitive agreement to acquire Pipedream. The integration layer had become the point.

Make connecting things boring - in the best way.

The mission never really changed. Make developers dramatically more productive by removing the work that should not exist. What changed is what counts as "connecting things." In 2019 it meant wiring a webhook to a database. By 2025 it meant giving an autonomous agent safe hands across a company's entire software stack. Same plumbing, far higher stakes.

There is a quiet ambition in choosing to be the layer nobody notices. Pipedream is not trying to be the app you open every morning. It is trying to be the thing that makes the apps you do open actually work together. The best infrastructure is invisible, right up until it disappears and everything breaks.

The agents are coming. Someone has to wire them in.

Software is shifting from things people operate to things that operate on their own. Agents will book the travel, reconcile the invoice, escalate the ticket, and answer the customer. Every one of those actions crosses an API boundary, and every boundary needs a secure, reliable, well-documented connection. The integration layer is no longer back-office plumbing. It is the floor the whole AI economy stands on.

That is the wager Workday is making by absorbing Pipedream, and it is why a company of roughly two dozen people, with $22M raised, ended up mattering far beyond its size. The connectors are the moat. Whoever owns the most of them, maintained best, owns the most of what AI can actually do.

So picture that developer again - the one staring at two apps that won't talk. The weekend of glue code is gone. The 3 a.m. pager is gone. There is a trigger, a few lines of code, and now an AI agent that can run the whole thing without being asked twice. The annoyance Pipedream was built to kill is still there in the world. It just isn't their problem anymore.