Heroku Is Dead (And It Taught Us Everything We Know) Transcript, AI Summary & Key Points

00:00 Get around. Get comfortable, kids. Tonight, I'm going to read you another bedtime story. This is the story of Heroku, the best place to run software that anybody ever built and one of the least important things its owner has ever owned. Your company has a platform team probably, maybe you run it. They're building an internal developer platform. It has a name with Forge or Runway or something like that in it.

00:24 a portal, a golden pot and a YAML file you fill in to get a database. It has been in progress for about 2 years probably. Hold on to that because everything on that road map already shipped, finished, priced and documented in 2011 and the thing that shipped it is still running tonight. You can sign up in under a minute. Your platform team is going to build it anyway.

00:50 Some of you deployed to it every day. Some of you have never heard the name. Those two facts are the same story. Once upon a time in June 2007, three people decided that writing software should be as easy as opening a web page. They were called James Lindenbomb, Adam Wiggins, and Orion Henry. And to understand what happened to them, you have to remember what shipping code actually meant in 2007.

01:18 You rented a server. You installed an operating system on it. You installed the language runtime and a web server and a database. And then you spent an evening discovering that the version of the library your application needed was not the version your operating system had opinions about. You wrote a deployment script. The deployment script worked on your machine.

01:36 You sshed into the server at 11 at night and run it and something in the middle failed. And you fixed that. and then something after it failed again. The application was 10 files long. Getting it onto the internet took a weekend and the weekend was not the interesting part. So they named their company Heroku which is heroic and Hiku pushed together because they wanted the difficult thing to come out short.

02:06 They went through Y Combinator in the winter of 2008 and their first product was completely wrong. It was a code editor that run in the browser. You would write your Ruby application in a web page and Heroku would run it. It was genuinely impressive and people signed up and then they did not come back because almost nobody wants to write software in a text box.

02:28 But the founders noticed something in the logs. The people who did stay were not using it to write anything. They were pasting in code they had already written somewhere else because Heroku was the fastest way they had ever found to get it running on the internet. The editor was the product, the machinery that took your code and run it on the internet was the plumbing underneath built to make the editor possible and nobody was ever supposed to look at it.

02:58 Their customers had quietly informed them they had it the wrong way round. So they threw away the product. The plumbing became the company. Heroku launched properly in April 2009. And what it launched with was one command. You wrote your application. You typed git push Heroku master. And that was the entire deployment procedure. No server, no operating system, no runtime to install, no web server to configure, no script that worked on your machine and nowhere else.

03:33 You pushed your code the same way you already saved your code and a few seconds later there was a URL and the URL worked. I deployed to Heroku before I understood what I was doing. That was not modesty. I genuinely did not know what the web server was and it did not matter because nobody asked me before that the gap between I've written something and other people can use it was measured in evenings.

04:01 Heroku made it one line in a terminal I already had opened not faster a different category of thing. It deleted the whole class of work from my job and it deleted it so completely that people who started after me never found out the work had existed. And here's the part I keep having to explain in meetings. That one line is still the target. Not a nice bit of history, the target.

04:31 Almost every internal platform I get shown has a slide near the front promising developers something very close to it. 17 years on with far better technology underneath and most teams are still building towards it. That's the highest compliment you can pay a piece of infrastructure. It's also the murder weapon. Heroku named its platform generations after trees in alphabetical order.

04:59 The first was argant or argent aspen in 2009. It ran ruby and only ruby. The second was Bususe Bamboo I think in 2010. And in 2011, Heroku shipped Seladon Sedar, which is the one that really mattered. Setter was the moment Heroku stopped being a Ruby company. It run Ruby and Node and Python and Java and Scala and Closure. And it did that through a thing called builtpuck.

05:31 A small replaceable piece of logic whose only job was to look at your code, work out what it was, and turn it into something runnable. That word built should sound familiar to some of you and it should sound familiar for a reason. If you deploy to digital oceans app platform or you use gitlub or glow cloud or hashior corpse waypoint or kpak you're using billpucks today this week.

06:00 Hiu invented them here in 2011 for itself. Hold on to that because where they end up is the strangest thing in this entire story. Underneath your application run a dino. A dino was a container, a real one isolated with Linux namespaces and croups. The same kernel primitives Docker would use later. It held no memory of yesterday. If it died, another one started.

06:27 If you needed 10 more, you move the slider and there were 10 more. Heroku did not invent containers. The Linux primitives had been sitting there for years, but nobody had built the tooling to make them usable. So Heroku wrote their own. They were running containers in production in 2009. Docker went public in 2013. Four years and Docker came out of cloud which was a Heroku competitor.

06:56 They built container tooling to run their platform worked out that tooling was worth more than the platform threw the platform away and became Docker. Heroku had the same thing and bet the other way. Keep the containers underneath sell the experience on top. That bet is exactly why Hioku was so good. It also handed somebody else an entire industry. Hioku's other invention of 2011 was not a piece of software at all.

07:24 Adam Wiggins brought down the rules. 12 of them. Keep your configuration in the environment, not in the code. Treat your backing services as a touched resources. Build, release, and run uh are separate stages. Keep your processes stateless. Treat logs as a stream of events, not as files. Make your application disposable so it can start fast and stop cleanly.

07:45 He called it 12 factor up. And Hiu published it on the open internet for free with no license and no catch where anybody at all could read it. I remember that coming out. I could not tell you the date, but I remember reading it and just going, "Fuck yeah. Obviously, that's how we should be doing this. There was nothing to buy, nothing to install. It was somebody writing down what good looks like and it was so obviously right that arguing with it felt stupid.

08:19 Go and read it tonight. It's 15 years old and most of it has not moved. Now, there was one more thing which is the reason so many of you have a feeling about this company rather than an opinion about it. Hiu had a free tier and it lasted for over a decade. Anyone with no money and no credit card could put the real application on the real internet and send somebody the link.

08:44 The free dino went to sleep after 30 minutes of nobody visiting and woke up slowly when somebody finally did and everybody knew it and nobody cared because the alternative was nothing at all. Every pet project I had run on Hioku. every single one of them three four years and the company I worked for did not run a thing there not one application we paid other people for that same engineer same week sometimes two completely different answers and the only variable was whose money it was now that's anecdotal and plenty of

09:22 companies did pay them but I don't think it was unusual heroku taught an enormous number of us what deployment even is your first application on the internet, your first environment variable, your first database you didn't have to install and it did all that teaching on the house. It won every heart in the industry and almost none of the purchase orders.

09:45 On the 8th of December 2010, Salesforce announced it was buying Heroku for about $212 million in cash. Mark Benov explained why the next area of computing he said was social, mobile and real time. He called it cloud 2. And then he said the sentence that this entire episode hangs on. Ruby is the language of cloud two. Ruby. Seriously. To be honest, Ruby was right at the top of its arc that year until we started asking why the [ __ ] is this written in Ruby?

10:23 Why? More importantly, lesson learned is that any platform that chains itself to one language is going to fail. And that was as obvious in 2010 as it is today. Containers are the proof. The entire point of them is that how you run an application has nothing to do with what it's written in. The real reason for the purchase was better than the stated one.

10:48 And it was not stupid at all. Salesforce was extraordinary at selling software to people who sign purchase orders. It was terrible at reaching the people who write code. It had built its own developer platform force.com with its own proprietary language called AEX. And developers had looked at AEX and quietly declined. AEX. Seriously, what the heck?

11:14 A company that couldn't get anyone outside its own customer base to touch its programming language. announcing to the world which language the next era belongs to and that is exactly why they were shopping. They did not buy Heroku because they understood developers. They bought Heroku because they had proof they couldn't reach them. Heroku came with roughly a million Ruby developers already attached to it.

11:38 Salesforce was not buying a product. It was buying a constituency it had never once managed to earn on its own. Now, this is the bit where people start saying the founders sold out. That's [ __ ] 212 million in cash in 2010 for a 4-year-old company. That's not selling out. That's winning. Anyone who tells you that they would have turned it down is lying to you or has never had to make payroll on a Friday.

12:08 Salesforce did not abandon Heroku. Quite a lot happened over the next few years and it is worth listing what because the list is the diagnosis. In 2014, Heroku shipped Heroku connect which synchronize your applications data with Salesforce. In February 2015, Heroku Enterprise in January 2016 private spaces isolated networks for companies that needed their application inside a boundary.

12:35 In June 2016, Teams in June 2017, Heroku Shield for organizations with HIPPA and PCI obligations. Now look at who every single one of those is for. Network isolation, compliance boundaries, data residency, enterprise contracts. That is not a list of things a developer wants. That's a procurement checklist. They paid $212 million for a million developers and then spent six years building compliance features for insurance companies.

13:08 And here's the thing about that list. Compliance gets you through the paperwork. It does not get you adopted. What gets a platform adopted inside a big company is being able to say yes. Yes to the strange networking. Yes to the stateful thing. Yes to one process nobody's allowed to change. Heroku built everything you need to pass an audit and nothing you need to say yes.

13:34 So they stopped serving the developers. They bought and built an enterprise that still couldn't standardize on it. Six years and between two customers landing on neither. Meanwhile outside the ground moved. Docker arrived in 2013 and made containers something anybody could hold in their hand. Kubernetes arrived in 2014 and in 2015 Google gave it away to a foundation so it belonged to nobody and Kubernetes was and still is substantially worse than Heroku as a thing Heroku did.

14:09 It's enormous. It's complicated. Nobody has ever typed one command into Kubernetes and had their application running 30 seconds later. Nobody. Ask anyone who has spent an afternoon reading a YAML file out loud to a colleague trying to work out which of the four indentation levels is the one that's wrong. But Kubernetes could be extended by anyone in any direction.

14:35 Red Hat built Open Shift on top of it and sold it to banks. VMware built Tanzu bought Rancher. Amazon, Google and Microsoft each sold the managed one. They don't build a business watching it. An entire economy of a thousand companies grew on top of Kubernetes and every one of them had a commercial reason to go out and evangelize the thing they were standing on.

14:56 Now Heroku had a marketplace as well. It launched with 85 add-ons in 2012 but 2015 there were about 150. Today there are a bit over 200. 85 to 200 in 14 years. That's the number I keep coming back to and I think is the actual cause of death. Kubernetes did not win because it was good. Kubernetes is not good. Kubernetes won because it made a thousand other companies rich and a thousand rich companies will carry you anywhere anywhere you want to go.

15:31 Heroku substraction was so complete, so beautifully sealed that there was nowhere for anybody else to attach anything. You could sell Heroku a database to plug in. You could not build a product on top of Heroku and sell it the way Red Hat built an entire enterprise business on top of somebody else's scheduleuler. There was no surface and so nobody in the industry had a single financial reason to argue for Heroku in a meeting ever except Heroku itself.

16:04 On the 25th of August 2022, Hokku announced that the free tier was ending. It was gone by the 28th of November. The stated reason was fraud and abuse. By 2022, the free tier wasn't teaching anybody anything anymore. Nobody was learning to deploy on Heroku. They were learning on Docker and on Kubernetes and whatever their company had already standardized on years earlier.

16:31 So killing the free tier wasn't the moment Heroku lost the developers. It was the moment they admitted it had already happened. In April 2025, after 14 years, Heroku planted a new tree. They called it fear. And Heroku's own announcement was titled planting new platform roots in cloud native. Fear was genuine rebuild. Underneath it ran on Kubernetes, it ran on unprocessors.

16:58 It used OCI container images and cloudnative buildp packs which meant that for the first time you could build your application on Heroku and then run that exact artifact somewhere else on your own machine in Docker anywhere. And it had native open telemetry traces, metrics and logs from your application and from the platform underneath it in an open standard exported anywhere you liked.

17:24 Heroku had gone and fixed it. The sealed box that nobody could see into and nobody could build on could now be seen into and built on. And that is the right answer. Properly done as well. Not a patch on the old thing. A rebuild everything that had been wrong with Hiro for a decade. They went and fixed 11 years after it would have mattered. 10 months later on the 6th of February 2026, Heroku's chief product officer Nitin But published a post called an update on Heroku.

18:01 It said that Heroku is moving to a sustaining engineering model focused on stability, security, reliability, and support. It said that enterprise account contracts will no longer be offered to new customers, though existing ones would be honored and could renew. It said there would be no change for customers using Heroku today. It did not say the platform was closing because it isn't.

18:24 There is no shutdown date. You can still sign up tonight with a credit card and it will still work and it will still be lovely. It said an investment is going to enterprise AI instead sustaining engineering. Nobody was confused by that, right? The industry decoded it inside a day. The migration guides went up almost immediately. It means start planning your exit.

18:47 Stay as long as you need to. Support will be there. Your build will keep arriving. And do not plan on still running on Heroku in 5 years. Is the politest possible way of telling several thousand companies to leave. Just go. So why would anyone shove a product that works and has customers and makes money? Arithmetic. Salesforce does something like $46 billion a year.

19:14 Nobody publishes Heroku's number, but nothing I can find puts it above a fraction of 1% of that. A rounding error that still needs security patches, compliance audits, incident response, and a product team. While the board asks, "What are you doing about AI?" Killing it would be stupid. It makes money. Investing in it would also be stupid because nothing it could possibly become would move that needle.

19:41 So you keep it alive and you stop feeding it. I would probably sign the same memo and the best developer experience anyone has ever built ends up as a maintenance line item. Heroku is not dead. It is running tonight quietly, competently in a way that would embarrass most of what we have replaced it with. and every application on it can leave whenever it likes because Heroku is the company that taught this industry how to build application that can leave.

20:12 Its ideas did not meily survive. They won completely. The 12 factors are simply how software is built now. Configuration in the environment, stateless processes, logs and streams, disposable replaceable identical instances. It is in your Docker file. It is in your Helm chart. It is in the pull request you approved this afternoon. Most of the people following those rules can never even heard the phrase fail factor up, let alone the name of the company that wrote them down.

20:39 And the build pucks Heroku invented became one of the cloud native computing foundations top tier projects in August 2026, 6 months after Hioku was told to stop building things. And your platform team is still going. Two years in, six engineers, a portal, a golden pot, a YAML file you fill in to get a database, building something that looks a great deal like Heroku for a company that could never have used Heroku. Good night, kids. Sleep well. Your golden pot is in review.