Last December, we announced the Community Public Preview of Healing Agent.
Today, I’m excited to share that Healing Agent is now available in Public Preview for our Enterprise accounts.
On behalf of the entire Healing Agent team, I wanted to thank you for your continuous support and feedback throughout this journey. Our team is already hard at work developing the new exciting features for Healing Agent, and we’ll keep you updated on our progress. In the meantime, we encourage you to continue testing Healing Agent and sharing your feedback via the Insider Portal.
For those new to this capability, here is a short summary:
Healing Agent is a set of AI-powered experiences across UiPath Robots, Orchestrator, and Studio, designed to help you reduce automation downtime and enhance efficiency. It provides:
Intelligent self-healing for UI-based automations: Whenever there is a change in the application interface, Healing Agent repairs the issue to ensure the automation continues uninterrupted.
Actionable recommendations: Healing Agent intelligently analyzes application interfaces and provides actionable suggestions on how to fix issues—without the need for further troubleshooting or debugging.
Demo:
How can you get started? Discover a detailed user guide on our dedicated Insider Portal page.
Once again, thank you so much for your continuous support. We look forward to hearing your feedback.
This is great and I’m really excited to try it out. But what kind of pricing are we looking at? Or is it just included in all enterprise subscriptions? Thanks.
Watched it twice now, and while it is pretty interesting, a couple of notes:
Would it be possible for the Healing Agent to generate fix recommendations without performing them live? This would allow for minimizing repro/debug time without making the automation non-deterministic (see later in the post on why that is a risk factor).
Would it be possible to enable Healing Agent features selectively? For example in a process I might be ok with pop-up dismissals, but not with semantic element selection (or vice versa).
And that’s my biggest question mark here - one of the (few) “features” that RPA has over “newer” automation approaches, is that it is fully deterministic. Yes, this means that it can fail if something changes, but also it will not do something unexpected.
Foreword:
I know that this was just a lab demonstration, and I would love for this to work flawlessly, but risk analysis is part of the job, so:
Looking at the demo video, and extrapolating it to a real process, the fixes applied are not as clear-cut:
Pop-up dismissal → aside of the pop-up on login, usually when there is an unhandled pop-up in the process, it most often signifies either an unhandled BREx or unhandled processing path. Dismissing it and just moving forward is not something that I’d call a safe default, but it is part of the “all or nothing” toggle.
Semantic element finding → Submit and Post might be semantically close, but have different meanings in a lot of cases - just being a close synonym from a pure language perspective could backfire with industry specific wordings (try telling an accountant that submitting and posting the invoice is the same thing, and see what happens ). Also, how would this work in selecting f.e. dropdown options? In a list of users, or sites/addresses etc. a one char difference can make a huge difference. If it’s similar to “closest selector” error messages, these can be very hit or miss in these scenarios.
Also as Oscar asked in the previous post, pricing is a big factor here. In a different thread, it was hinted that it could utilize a AI unit consumption model (which totally makes sense for what it does). That said, it also opens up for variable processing costs.
Imagine this scenario:
In a process like in the demo, you need to enter 10.000 invoices. That’s 20k healing actions with live healing (unless it does caching of activity failures). If the rearrangement of the form would be a bit more drastic, it could be up to 80k (if the 6 field selectors also broke).
Since there’s no pricing set (as far as I know), we could only speculate as to how much that would cost, but I’d guess that 80k healing actions would not be pocket change, and that could be just a weekend.
So in summary while I’d like to get excited about the Healing Agent, it would be great to get a bit more info on the practicalities.
). Also, how would this work in selecting f.e. dropdown options? In a list of users, or sites/addresses etc. a one char difference can make a huge difference. If it’s similar to “closest selector” error messages, these can be very hit or miss in these scenarios.
prashant1603765
ashokkarale
imad-pro
Matias_Clemente.Arg
Shah_Rukh