Handling AWS Support Cases With an AI Agent

Handling AWS Support cases with an AI agent means letting an assistant gather the context a case needs, draft the case or the reply, and monitor status — while a person with the right permission remains the one who actually opens the case or sends the response.

---

The slow part of an AWS Support case is almost never the writing. It is the assembly: which account, which region, which resource, what changed, what the error actually said, and what you already tried. That information exists across several systems and one person’s memory, and gathering it is what turns a ten-minute case into an afternoon.

It is also exactly the sort of work an agent is good at — bounded research over systems it has been granted access to. The filing is the part that should stay human, and it does.

---

What the integration does

In Insulin, AWS Support opens as its own window on the workspace desktop. It shows the same cases as the console’s AWS Support tab, because it talks to the same backend and the same AWS account — so this is a different surface onto your existing cases, not a parallel system.

From there you can check an open case, open a new one, and reply to AWS without leaving the workspace. The underlying AWS Support API supports opening a case, listing and filtering cases by date or identifier including resolved ones, adding correspondence and attachments, adding email recipients, and resolving cases. Attachments are capped at three files of up to 5 MB each, which is an AWS limit rather than a workspace one — worth knowing before you plan to attach a debug bundle.

Creating and replying are permissioned separately

This is the detail that makes the integration safe to enable broadly. The button to create a case appears only if your role grants permission to create AWS Support cases, and creating and replying are permissioned separately.

That separation is more useful than it first sounds. A large group can be given read access — enough to check status, read the correspondence thread, and see what AWS has asked for — while a much smaller group holds the ability to say something to AWS on the company’s behalf. Case visibility stops being a bottleneck without making everyone a spokesperson. It is the same role-based access reasoning that applies everywhere else in the workspace.

The agent gathers; the human sends

Scope the agent to the systems the evidence lives in, and let approval hold the outbound step. An agent scoped to its integrations can pull the account and resource identifiers, read the relevant monitoring or ticket history, check your own runbooks in a knowledge base, and produce a draft case with the evidence attached and the reproduction steps written out.

What it should not do is file it unattended. Anything that reaches AWS is external communication about your infrastructure, which is precisely the category that belongs behind an approval — and in Insulin, consequential actions run through approval workflows where you see the plan before it runs. The act-or-ask boundary is the same question in a different context.

Monitoring is the other half, and it is the half teams under-use. A scheduled job that reads open cases and flags the ones where AWS is waiting on you catches the single most common way a case stalls.

Check the support plan before you plan around this

The AWS Support API is not available on every plan, and the plan lineup is currently changing.

AWS’s documentation states you must have a Business Support+, Enterprise Support, or Unified Operations plan to use the Support API. It also carries end-of-support notices: Developer Support, Business Support and Enterprise On-Ramp are all being discontinued on January 1, 2027, with Enterprise On-Ramp customers upgraded to Enterprise Support during 2026. Those three plans remain available in the AWS GovCloud (US) Region.

So confirm your current plan and what it becomes after that date before building a workflow that depends on API access. Read it from AWS’s own page rather than any summary, this one included — it is the kind of detail that changes.

One reassurance for the audit conversation: AWS Support case-management operations support CloudTrail logging, so actions taken through the API are recorded on the AWS side as well as in your workspace’s own run history.

---

Frequently asked questions

Can an AI agent open AWS Support cases on its own? It should not. The agent gathers evidence and drafts the case or reply; a person with case-creation permission approves and sends. Anything reaching AWS is external communication about your infrastructure.

What can the integration actually do? List and filter your organization’s cases, open new ones, read status and the full correspondence thread, reply to AWS, and attach files — up to three files of 5 MB each, which is an AWS limit.

Who can create versus reply to a case? Different people, if you want. The create button appears only for roles granted that permission, and create and reply are permissioned separately — so many people can read cases while few can respond.

Which AWS Support plan is required? AWS states the Support API requires a Business Support+, Enterprise Support, or Unified Operations plan. Developer, Business and Enterprise On-Ramp are being discontinued on January 1, 2027 — verify your plan on AWS’s page.

What is the highest-value thing to automate here? Monitoring rather than filing. A scheduled job that flags cases where AWS is waiting on your response catches the most common reason a case stalls.

Takeaways

• The slow part of a support case is assembling evidence, and that is the part an agent should do.
• Filing and replying stay human — anything reaching AWS is external communication about your infrastructure.
• Create and reply are permissioned separately, so case visibility can be wide while the ability to respond stays narrow.
• Attachments are capped at three files of 5 MB each; that is an AWS limit, not a workspace one.
• Confirm your support plan first: AWS requires Business Support+, Enterprise Support or Unified Operations, and is retiring three plans on January 1, 2027.

---

Bring your operational systems into one place: see agents, jobs and knowledge bases, or book a demo.