Heroku made servers somebody else’s problem.

Consider the pen. In Heroku’s account of PensionBee, an online pension provider, a robot puts an actual signature on actual paper. Behind it, applications on Heroku connect Salesforce to the signing machinery. Customers authorize their signature once; subsequent correspondence can move without another round of paperwork. A cloud platform ends up doing something gloriously physical. The useful question is how much engineering the pension company had to do before it could get to the pen.

That question gets to the heart of Heroku. Developers want to make a service work. Between their code and a customer sits a procession of jobs: assembling dependencies, running processes, connecting databases, handling releases. Heroku built its business by taking responsibility for much of that procession.

01 First, they removed something

James Lindenbaum, Adam Wiggins and Orion Henry founded Heroku in 2007. The early platform included an editor inside the browser. You could write Ruby applications there and see them run. It was an appealing package, particularly for people getting started.

By January 2009, the team reported more than 20,000 applications in its private beta. Yet demand had exposed an awkward truth. Commercial users needed a different set of trade-offs. Lindenbaum wrote that the lessons included features “some we need to remove as well.” The team built a second platform for production, separating the original experience into Heroku Garden.

The distinction matters. A browser editor tells developers where to work. A deployment platform helps them publish the work they already do. Heroku’s commercial launch came in April 2009. Its sharper product boundary made room for a larger ambition.

“some we need to remove as well”

James Lindenbaum, 2009, on the features customers needed

02 A small machine for shipping

Heroku’s unit of computation is the dyno, a managed Linux container. Web dynos handle requests. Workers do background jobs. Temporary, one-off dynos handle chores such as database migrations. Developers describe the processes; Heroku manages the machinery that keeps them running.

Buildpacks turn source code into runnable applications. Pipelines organize delivery stages, while Review Apps let teams inspect proposed changes in running environments. The practical achievement is a shorter distance between a change in a repository and something a colleague can test.

Around that runtime sit Heroku Postgres, a managed key-value store, Kafka and an add-on marketplace. Heroku Connect synchronizes Postgres with Salesforce. The expertise here is joining operations into a coherent experience: application processes, data services and delivery tools that a small team can use together.

03 The bill has two columns

Heroku raised a $3 million venture round in 2008, followed by a $10 million Series B led by Ignition Partners in 2010. In December 2010, Salesforce announced an agreement to acquire it for approximately $212 million in cash, net of cash acquired. Salesforce was buying a route into the developer’s working day.

Today, customers buy compute and data services. Eco costs $5 a month for 1,000 dyno hours shared across an account. A Basic dyno costs up to $7 a month. Those figures describe compute plans, not an entire production system. Databases, additional processes and paid add-ons can add to the bill.

The second column is time. How many hours would the team otherwise spend operating infrastructure? That makes Heroku attractive to teams short on operational capacity. It also means the cheapest virtual machine is an incomplete comparison. A purchasing decision needs both columns.

04 Paperwork, poetry and traffic spikes

PensionBee illustrates the Salesforce connection. Its customer-facing application and internal systems needed consistent data; Heroku Connect supplied the bridge. The company’s CTO, Jonathan Lister Parsons, said: “With Heroku, we don’t need a dedicated DevOps team.”

LitCharts arrived through a different door. Its original shared hosting struggled with traffic spikes, instability and outages as readership grew. Moving to Heroku gave the literature-guide publisher managed infrastructure and room to scale. Its team also used add-ons instead of building every supporting service itself. These are businesses selling pensions and literary understanding. Running servers is necessary work, but it is far from the thing their customers came to buy.

05 Convenience has an operating cost

The promise has had difficult chapters. In April 2022, a security incident exposed GitHub OAuth tokens and other data. Heroku’s incident review described stolen customer-account data and pipeline configuration variables. The company rotated credentials and tightened access. Outsourcing operations still leaves customers dependent on someone else’s security practices.

Later that year, Heroku retired its general free plans. Its explanation pointed to the effort spent fighting fraud and abuse. Free entry had helped make the platform accessible; keeping it open had an operating cost. Learners and hobbyists had to adjust.

Then, in February 2026, Heroku announced “sustaining engineering.” New Enterprise Account contracts stopped. Existing subscriptions could renew, and credit-card customers could continue. March brought clarification: aligned improvements would still ship, including larger compressed application packages and CLI v11. The platform remained supported, with a narrower investment commitment.

06 What travels beyond Heroku

There is still competition for this job. Render and Railway offer managed deployment; Fly.io offers application VMs. Teams must compare their actual workloads, networking needs and operating costs. Heroku’s own conventions favor disposable processes and durable data stored outside the application’s local filesystem. Software built around persistent local files needs adaptation.

The portable lesson comes from Twelve-Factor thinking: separate configuration from code, treat backing services as attached resources and make processes replaceable. Those habits help even when the hosting company changes.

Heroku’s 2025 Fir platform brought Cloud Native Buildpacks and OpenTelemetry into the experience; its AI offering adds managed inference and agent tools. The original question survives each new layer: which work can the platform take away? Before choosing it, ask what your team would gain from that relief, and whether its present commitments fit your application’s future.