82547870-a225-4398-9034-75c41500ab8b

[giscus[bot]]

82547870-a225-4398-9034-75c41500ab8b

An AI agent that verifies its UI change in a real browser should keep the screenshot or video it took. Put that evidence on the PR and the Done comment, so bugs get fixed before a human reviews and reviewers see the change without checking out the branch.

https://www.ssw.com.au/rules/done-video-by-ai

[Aibono1225]

I've seen AI Done Video once. It was pretty cool, but I'd like to see more and give it a few tries

[jeoffreyfischer]

I see the benefit of getting agents to repeatedly check and fix UI issues, but I wonder if there’s a risk of this leading to over-complexity in the code itself. Could the visual feedback sometimes cause the agent to over-engineer the fix?

[danielmackay]

Done videos can take a lot of time. Automating this is a great idea. Historically done videos include footage of the developer so it's clear who did the work (and how awesome SSW is). With this new approach we need to ensure done videos have a title screen that includes the developers name, so that they get attribution towards the work. This is especially important when working for a client with a mixture of client devs and consultants.

[TobyChurches]

I really like this AI done video! One less thing to do at the end of a prompting session 😂 🤖 This really does put a massive emphasis on a well defined backlog in my opinion. If the PBI is lacking details, AC clarity or context about common workflows in the application (Which tbh - Is most of the time), I could see the implementation and resulting demo really deviating from the initial intent.

[vladislav-kir]

I have not used AI to generate Done Videos before, but saw that done and was pretty impressed!

[isaaclombardssw]

Sometimes... an agent will try and recreate a version of the page to take a screenshot of rather than running the whole solution. I've had to discourage that, but otherwise it's a great practice.

[suiyangqiu]

I like the concept behind agents "seeing" their changes, and being able to iterate before it comes back to the human driver - but I'm on the fence about fully AI done videos/screenshots.

Allowing AI to automate these done videos and screenshots are great for time, but at the same time, most people don't "teach" their agents how to communicate like them, or in a way that a human would really want to read, so it ends up padding content unnecessarily and being a painful process to review.

[RobJT-wq]

I can see this saving me a lot of time when making a video for non-hardware related projects.

[calumjs]

When AI can reliably get data and use it to create pages to display in a format a user wants, ~as fast as a browser can load one, will front ends still be relevant? 🤔

[0xharkirat]

I was expecting SSW.AgentSkills for this so that I could ask my AI agent with that skill to follow the same SSW standard to make done video.

Now I have to find out the voice tool, and setup myself.
Calum 👀

[calumjs]

https://github.com/calumjs/ARMADA/blob/master/skills/logbook/SKILL.md
That's the skill from the user group - to use it you need a text to speech service for the AI to narrate its video so I haven't incorporated it into the AI Skills repo yet (because no one would be able to use it)... feel free to contribute a PR if you can come up with a solution!

[harley-j]

Definitely a great use of AI in the pipeline, keen to see how AI done videos contribute to our work!

[bahjatalaadelssw]

Great resource for updating PBIs

[ricksu978]

AI video and voice would cost a bit extra, and it may take a few attempts to get the output right.

Happy to use this feature on a project once it is fully set up.

[Ben0189]

If it’s just a quick and simple screenshot, I would capture it myself rather than wait for the agent to do it, but AI Done Video looks cool and I’ll try it next time.

[18-th]

It's a nice way to capture a bunch of screenshots quickly. I still re-test the UI myself to make sure that the agent hasn't followed a path biased by the context of our conversation, e.g. making sure that the new popup is there despite exploding the rest of the page.

[kulesy]

Yes, I do this with pretty much all of my UI changes. An example is a CTA button was invisible on the teal panels. The PR carried before/after grids of both the resting and hovered states on teal, black and white, plus a capture of the four live production pages the fix would change.

The next level is getting done videos for them, not sure about the AI voice though 😅

[ianhoffmanx]

If it's made part of Playwright tests, it's probably a good 2-birds-1-stone solution.

[ZenoWang1999]

Yeah, it's often hard for other devs to reproduce my changes locally, especially UI that only shows up in rare cases, and most of them just care about the logic. So using AI to mock it up and share screenshots really helps.

[thom-wang]

I'm a little bit worried about this because most of our PBI is written by Yakshaver. It's already generated. So if it has a wrong description or AC. It could go wrong for the whole AI toolchain.

[stephanfakossw]

Great way to save time on PR approvals

[larrycarland]

Where visual updates on changes matter I can see this helping alot.