How does JAMEs safely make changes to Jira?

You have seen your Jira enough to know how complex the configurations can be. Worse, even small changes can easily creep beyond their intended scope and bring your teams’ work to a full stop. So why should you consider letting an AI agent handle it for you?

JAMEs is built for exactly this purpose. When you make a request, it analyzes the potential impact on your Jira and proposes a risk-mitigated implementation plan you can easily understand. Nothing runs without your explicit confirmation. JAMEs then carries out the steps, each one explicitly shown for you to track, and verifies your goals were delivered.

Execution runs with your API token, so JAMEs can never do more in Jira than you can. Additionally, it has no delete capabilities. It only creates and updates objects.

JAMEs lets you stay in control while it mitigates the risk and carries the implementation.

Below we cover each step, what you approve, how to undo a change, and why human approval matters.

Key takeaways

  • JAMEs is supervised AI administration: an AI that plans and executes Jira changes, but only after a person approves each plan.
  • Planning starts by reading the existing configuration of your Jira site, so every plan is built against what is actually there and the risks it actually carries.
  • Execution runs through the Jira REST API with your own API token; JAMEs can never do more in Jira than the connecting user can.
  • JAMEs cannot delete anything: it creates and updates Jira objects only.
  • Every event is auditable for 30 days; automatic rollback is a Premium-tier feature.

What happens when you ask JAMEs to change something?

“Add an approval step to the purchasing workflow” means something different on every Jira site, and the risk it carries depends on what that site has configured.

A request to JAMEs produces a plan, and the plan starts from your real site. Before proposing anything, JAMEs reads the existing configurations and homes in on the objects the change would involve.

The plan gives you a detailed description of what was found, risk exposure, and the open questions you need to consider. Use JAMEs to shape the plan until the requirement reads exactly the way you want it. Once the questions are answered, the risks identified, and the steps finalized, the plan is ready to accept into execution. Until you do, nothing runs and your configuration stays as it is.

What do you approve before anything runs?

You approve the plan itself. This is the concrete list of changes JAMEs intends to make on your Jira site, laid out for you to read before any of them run. JAMEs writes nothing to Jira before that approval. There is no small-change exception, because small is exactly what a review exists to judge. Renaming a field is trivial on one site and breaks a report on another.

Rejection is just as simple. Turn a plan down and it never executes; no approved plan, no write. You can ask again in different terms, or leave the site as it is.

How does JAMEs execute approved changes?

JAMEs executes approved changes through the Jira REST API, signed in with your own API token. That is the same interface admin scripts and integrations already use: no private back door into your site, just the access path Atlassian already supports.

It is also where the second safety constraint is applied: an API token acts as the user it belongs to, so JAMEs inherits that user’s permission limits. If your account cannot edit a permission scheme, JAMEs cannot edit it either. Connect a tightly scoped account and you have scoped the tool to match. If a step fails mid-run, JAMEs declares it loudly, so you have clear visibility in case your intervention is required.

That makes the connecting account a real decision, and it is yours: the token you hand over is the reach the tool has. What JAMEs stores while it works, where that data lives, and the rest of the security model are on JAMEs and security.

Wouldn’t a fully autonomous agent be more useful?

Not for Jira administration. Plenty of AI agents are sold on removing the person from the loop, and for some work that is the right trade. Your Jira configuration is not that work. A change may look small or subtle yet go against business reasoning that is reflected in the existing Jira implementation. JAMEs can spot the Jira collision; you can spot the business-logic collision. That is why the approval gate is a core feature of JAMEs, not a training wheel you outgrow.

While AI-based solutions carry great promise, Gartner predicted in June 2025 that over 40 percent of agentic AI projects will be canceled by the end of 2027, citing among its reasons unclear business value and inadequate risk controls (Gartner’s prediction). The approval gate answers the risk-control concern on both sides. This is the pattern of supervised AI administration: the AI plans and executes changes, but only after a person approves each plan. European regulation points the same way: the EU AI Act makes human oversight a design requirement for high-risk AI systems (Article 14).

What separates JAMEs from a manual change or an admin script?

A change in Jira can happen in one of two ways: an admin edits the configuration by hand, or a script does it through the API, whether that is a ScriptRunner job or a one-off REST-based script. Supervised AI administration is a third way, and the table shows where it differs.

AxisManual admin changeAdmin scriptJAMEs
Who does the workAdmin is responsible for analysis and implementationYou write and maintain the code; the script carries out the changeJAMEs analyzes and plans, you approve, and JAMEs executes according to the plan
Where errors come fromHuman error in analysis or implementationErrors in scoping or code, with post-execution visibility onlyDeep, guided analysis surfaces issues for review; a failed step makes no change to your site
Effort per changeManual, for each changeA code change for every new scope or analysis updateAutomated review of the current system setup
What record it leavesJira’s native audit log, under the admin’s own accountWhatever logging the script itself writesEvery event auditable for 30 days; automatic rollback on the Premium tier
How far it reachesEverything the admin’s account is allowed to touchEverything the credentials it runs with allowOnly what the connecting user’s API token allows
Can it deleteYes, anything the account can deleteYes, if the code makes a delete callNo; JAMEs has no delete tools

None of this is an argument against scripts; a reviewed script run by a careful admin has changed Jira safely for years. The difference is where the safety lives. With a manual change or a script, review is a habit someone has to keep on a busy day. With supervised AI administration, review is built into the mechanism itself: without an approved plan, nothing runs.

What if you need to undo a change?

A business requirement in Jira can be updated multiple times, and even rolled back occasionally. To undo a change, you provide the agent with the previous state, and it plans the restoration the same way it plans any other change: reviewed and approved by you before it runs.

Your history is kept for 30 days, where you can audit every change made while speaking to the agent. Premium goes further: a detailed search across conversations and users to trace a change, and support for an automated rollback where possible.

Whatever tier you are on, the worst case stays bounded: JAMEs cannot delete, so a wrong change is never data that is gone.

What can’t JAMEs touch?

JAMEs cannot delete anything, cannot change automation rules, and does not manage boards yet. The first limit is absolute: JAMEs has no delete tools, so no plan, however you phrase it, can end in something being removed from your site.

Automation rules are read-only. JAMEs can query your automation rules and answer questions about them, but updating them is not supported. Board management is coming soon rather than out of scope.

If some of the changes you need are still unsupported, reach out to us at hello@viter.io.

For the full feature table of what JAMEs supports, see what JAMEs can manage.

Frequently asked questions

Does JAMEs train on my data?
No. JAMEs does not train on customer data. What it reads from your site and what you type at it is used to do the work you asked for, and it does not become training material.
Where is my data stored?
JAMEs stores conversation history (kept for 30 days, longer on the Premium tier) and, should you choose, your automation rules. It may also save your API key, depending on your preference. All the data is kept in the EU.
Does JAMEs need admin rights?
It needs yours, whatever yours are. JAMEs authenticates with the API token of the user who connects it and inherits that user’s permission limits. The account you connect sets the ceiling: JAMEs can never do more in Jira than that user is allowed to do.
What happens if I reject a plan?
Nothing. A rejected plan is never executed. No approval means no write, and your Jira configuration is left untouched.
Is JAMEs available today?
In beta, yes. There is no general release date. In the beta there is no subscription fee; you pay only for the credits you use. Access is arranged directly with Viter, and the JAMEs page covers how to join.

If spending too much time inside Jira configuration is your pain, JAMEs carries that for you, while letting you stay in control. The beta is open: JAMEs explains how to join, and it starts with an email to hello@viter.io.

Sources

External claims on this page were checked against these documents in August 2026:

About the author

Ben Friedman is an Atlassian consultant and Forge developer with over 8 years of hands-on Jira work, certified ACP-610, ACP-620 and ACP-120. He has delivered for hundreds of customers across all business verticals, supporting their configuration, administration, and ongoing support and maintenance. Ben is building JAMEs, the AI Jira admin.