nader dabit@dabit32h ago

📝 Automating Exceptional Product Tutorials at Scale with Devin You can automate pixel-perfect product tutorials with a cloud agent that has its own computer and records itself using your app the way a human would (like Devin). This has technically been possible for a while, but with Opus 5.5 the results have progressed to something better than what I make by hand. I've been creating these types of tutorials for 10+ years, so I don't say that lightly. It's especially useful for the basic yet important tutorials that don't need a talking head. And it's 100x easier and faster. TLDR: Your cloud agent already has it's own computer. Give it an account in your system with everything set up, access to your docs, a style guide, and any other resources. Then ask it to create tutorials for you, as a human would. Here's a screenshot of my environment when doing this (signed in cloud computer on the right, prompt on the left): This works especially well in Devin, which has native macOS, Linux, and Windows machines in the cloud, so it can run pretty much anything: web apps, native iOS, iPad, and macOS apps, Windows apps, and Android apps. It can also download any app, authenticate into any app or service, securely access credentials, and use any tool a human would. So you can create these exact tutorials for your iPhone app, Android app, etc... Once you have a good workflow you can create these demos for literally every feature you'd like to in parallel, as cloud agents can spin up hundreds or even thousands of their own computers at once. You can automate the process so that when a new changelog goes up, you trigger the videos to be built simultaneously. In fact, Devin also made a tutorial to show you how to do it with Devin! Testing and verification All of this builds on something Devin already does every day: it tests apps end-to-end, running everything and recording a video of the entire process, with on-screen annotations, highlighting where it clicks and types. Since it's just a computer, nothing is mocked. It installs the app, signs in, and uses it for real. A tutorial is basically that same run with a style guide on top, so it's already verified to work before you publish it. Here's a Devin-created tutorial showing how this works! Better in most ways Software moves extremely fast, it's nearly impossible to keep video tutorials up to date without hiring an army of video creators. Every software company I've worked for that has external facing docs has been unable to scale this. It's always been a challenge (impossible) to maintain these types of videos. As a human creating these videos, the process is very tedious. There are so many tiny details to get right that it's very hard to be consistent. Computers are extremely good at consistency. They can programmatically keep every frame consistent, blur out information you don't want seen, set borders and font sizes and margins, really anything. It seems we've hit the tipping point where I no longer feel that human-made tutorials of this type are better, or maybe I am saying they are objectively worse. Another example: The prompt Here's the prompt: https://gist.github.com/dabit3/5d3969b61db56dc150c161b46350584a It's worth noting that this is the worse that it will ever be. I've barely iterated on this prompt or workflow and think that, with a tiny bit of additional elbow grease, you could get results that are indistinguishable from a polished, professionally produced tutorial. Launch videos and other things You can obviously apply this same technique to anything, the most obvious being launch videos. Here are a couple of things I've been able to achieve with similar techniques. Most of these have a general look and feel of our branding while being creative: And these just kind of went off the rails but were still pretty cool: https://x.com/i/article/2103573841176141824