nav[aria-label="Primary Navigation"] { padding: 0; & ul { list-style: none; width: 100%; display: flex; flex-direction: row; justify-content: start; align-items: start; gap: 30px; padding: 0; & li { margin: 0; } & ul li { list-style: none; } } }

🚀 Coming soon: Classify in Automations with Jev

Ben Costello

Hey Automation Community 👏

We've been working on something new over the past few weeks, and we want to give you an early look.

A few weeks ago, TypeSafe AI released Jev, its first public System One model.

Since then, we've been building a new Automation step that brings Jev into your automation flows.

We showed it on stage at Team '26 Europe, and we're now opening an Early Access Program (EAP) for customers who want to try it first. Your feedback will help shape the experience as we get it ready for more customers.

Loom walkthrough

Watch this to get a closer look at the new step and see some examples in action.

https://www.loom.com/embed/c09aea0a46624671aaaab664d2493816

What is Jev?

Jev is a new type of AI model that specialises in making decisions and classifying text. It answers questions like:

What priority should this work item have?
What request/issue type is this work item?
What team should I route this work item to?
What label should I add to this Confluence page?
Should I delete/archive this page?

How can Jev help you?

Jev is designed for classification, it analyses text and allows you to ask questions about it, then outputs the answers as structured data - this is perfect for setting and updating fields and labels.

This makes it ideal for many of the use cases you’re already using agents & automation for - triaging, prioritisation and more.

What does Jev look like in Automations?

We’re still working on how to make this as usable as possible, but here’s what we are testing internally.

A new Classify with Rovo automation step.

With 3 types of classification to choose from

Check - ask a True/False question of text (in Jev this is called a noul)
Categorize - categorize text into one of 2 or more categories (Jev calls this a Choice)
Score - choose from 2 or more ranks or ratings

Currently we’re supporting multiple simultaneous classifications on the same text in the same step.

What about other decision models?

Jev is the first but not the only decision model released recently. We are in the process of evaluating others, stay tuned.

When will Classify with Rovo be available?

We’re working to get this into customer hands ASAP.

In the meantime, f you’re keen to use Jev - and ideally share your experiences - please sign up for the Classify with Rovo Early Access Program (EAP)

I’ll post an announcement here as soon as we have an update.

-Ben (Studio Automations Team)

Comments

Viswanathan Ramachandran

Interesting and this looks promising, Ben!

For JSM, classifying a request’s type, priority and routing team in one step could make triage much simpler.

One thing I’d love to understand: Will the step expose confidence or support an uncertain outcome, so ambiguous requests can fall back to human review rather than forcing a category?

A way to test classifications against existing work items before enabling downstream actions would also be useful especially for higher-impact decisions like archiving or deleting content.

Looking forward to seeing how the EAP develops!

Mister Alex

This looks like a useful step toward smarter Jira automation. Using AI to classify issues, prioritize tasks, and route requests to the right teams could reduce manual work and improve workflow efficiency. I’m interested to see how well Jev handles real-world use cases and how much flexibility users will have when configuring classification rules.

Martin Runge

Hi, this is a great answer to usage-based pricing, as some customers can save on rovo credits here, right? Looking forward to getting hands-on with this!

Florian Bonniec

Looking forward to trying this out and seeing how it works for us.

Dang Quang

Hello,

This seems to be very excited, we have been using Rovo to do the classification for our works, but the reasoning makes it very slow to response. Jev seems to fit with it.
But if the categories are not something hard coded, but are dynamic (from the Asset objects/field options), it would be a very good fit, and even open us to utilise the native automation rules further

Mia Tamm _Simpleasyty_

Very interesting to see classification becoming a structured output rather than just another AI-generated text response.

One Confluence use case I'd be particularly interested in is classifying documentation against predefined metadata schemas.

For example, assigning a document type, business classification or review category using a controlled set of values rather than generating free-form labels.

Will the classification results be available as structured smart values that downstream automation actions can use, including integrations with metadata fields defined by Marketplace apps?

That could open up some really useful content governance workflows.

Arkadiusz Wroblewski

I will be very interested in testing how good it is actually in classifying. I see potential here to save some people money.

Ben Costello

Hey @Viswanathan Ramachandran , yes we are exposing confidence. For any classification we'll return
a) all classifications with associated probabilities and overall confidence
b) a single "answer" that is the highest probability response.

This allows most users to extract a single response and advanced users acecss to all probabilities and confidence.

Ben Costello

Hey @Mister Alex - we've biased towards maximum flexibility which means all classification types and multiple simultaneous classifications. We'll see how users work with this in the early access program and beta to see how users respond and work with this.

Am also looking forward to real world use and results, we're using it quite a lot inhouse but woulld love to see customer usage across more use cases.

Ben Costello

That's the plan @Martin Runge 😁, will announce pricing as soon as the that is all confirmed.

Ben Costello

Hey @Dang Quang - exactly! Jev is lightening fast compared to an LLM

Ben Costello

hey @Mia Tamm _Simpleasyty_ - that use case sounds great

The response are returned in fully structured smart values. That's one of the key benefits. I have been building agents to update fields for a couple of years now across Jira, Confluence and JM use cases and one of the hardest problems to solve is getting an agent to reliably return a structured result suitable for setting a field.

There's at least couple of approaches here:

1. Have the Classify step do the full classification
2.Add a Classify step after the agent, so no matter what format the agent returns it can be classified into a specific valid field value.

Good luck and please reach out and ask questions/share results.

Ben Costello

Totally agree! Good luck @Arkadiusz Wroblewskiand please reach out and ask questions/share results.

Dexter de Vera

Will check it out.

Dirk Ronsmans

@Ben Costello ,

Really interresting stuff! With all the AI advancements I had to look up what it all means but it sure looks promising!

Looking forward how this will work towards JSM Operations too and somehow linking this in to alert classification (or even broader event management) this seems like something that could really speed up getting the items that offer real value (or require human attention) to pop up much sooner.

Mia Tamm _Simpleasyty_

Thanks @Ben Costello! That's really interesting, especially the idea of using Classify to turn AI-generated output into controlled values.

I'm part of the team behind Simple Metadata for Confluence, where we're working on structured metadata, reusable schemas and reporting across pages.

I can see some interesting possibilities between AI-powered classification and structured metadata. For example, automatically suggesting metadata values based on page content.

Have you explored anything similar with Confluence metadata? I'd be curious to hear your thoughts!

Ben Costello

Thanks @Dirk Ronsmans - totally agree, I previously worked in a JSM team and I can see lots of use cases that this seems perfect for. I'm really looking forward to seeing these use cases in action when we release this more broadly than the early access program.