Introduction
This article is an English translation of the article at the following URL, which was originally written in Japanese.
https://qiita.com/amarelo_n24/items/9f1457a999510a5714a4
On June 9, 2026, AWS FinOps Agent was released as a public preview.
Following the Security Agent and DevOps Agent, another Agent service has been made available.
I didn't get a chance to try the previous two during their previews, so I told myself I absolutely have to try FinOps Agent while it's still in preview! — and dove right in.
Here are my findings and impressions!
https://aws.amazon.com/jp/about-aws/whats-new/2026/06/aws-finops-agent-preview/
Note: This article reflects the author's personal opinions. The results are based on personal use only, so please treat them as a reference.
Note: This article was written based on information as of June 11, 2026. At the time of writing, the service is only available in the US East (N. Virginia) region.
What is AWS FinOps Agent?
It is a cutting-edge agent that continuously monitors costs across your cloud environment, investigates anomalies, and makes it easy to identify optimization opportunities.
It supports:
- Cost anomaly investigation triggered by events
- Cost queries in natural language
- Periodic cost reporting
- Responses using context files and memory
It appears to work in conjunction with the following services:
- AWS Cost Explorer
- AWS Cost Anomaly Detection
- AWS Cost Optimization Hub
- AWS Compute Optimizer
- AWS CloudTrail
Cited from AWS official documentation
https://docs.aws.amazon.com/finops-agent/latest/userguide/what-is.html
Setup
I followed this guide during setup:
https://aws.amazon.com/jp/blogs/news/aws-finops-agent-is-now-public-preview/
Deciding the Agent Name and Creating an IAM Role
Open the FinOps Agent console and create an agent by clicking "Get Started."
Once you've decided on an agent name, create an IAM role with the following permissions:
- Permissions for the agent to access AWS resources
- Permissions for the web app to access the agent
Third-Party App Integration
You can connect either Jira or Slack to FinOps Agent. I chose to integrate with Slack.
Set up your Slack workspace and channel beforehand.
Note: Steps to create a Slack workspace and channel are omitted here.
Select Slack and configure the integration via the "Add Integration" button.
Click the "Authorize with Slack" button to grant the FinOps Agent app access.
Warning: I was unable to configure agent permissions without first disabling multi-session in the AWS Management Console. Be sure to disable multi-session beforehand.
I thought the agent was successfully created — but then got an error...
It turned out that the FinOps Agent App needed to be added as a member of the Slack channel, but there was no prior explanation of this. While you can set up the Slack integration later, it would have been helpful to have clearer instructions upfront.
After adding the App as a channel member and then configuring the integration, everything worked fine.
Running the Agent
Next, run FinOps Agent from the web app by clicking "Open" in the Open Agent section.
The FinOps Agent chat screen opens. Enter a prompt and it returns an answer. Japanese prompts are accepted, but responses were returned in English.
Costs with credits applied were not included in the totals. When I rewrote the prompt to calculate costs before credit application, it was able to aggregate them correctly.
Tasks
On the left side of the FinOps Agent screen, there is a "Task" menu.
Tasks run in the background without interrupting your conversation. When you ask the agent to create a task, it processes it in the background.
https://docs.aws.amazon.com/finops-agent/latest/userguide/task-management.html
Tasks come in three types:
- Run once
- Scheduled execution
- Event-triggered execution
Interestingly, clicking "Run Once" displays "Create Task," but clicking "Run on a Schedule" switches to the "Automations" selection, and the "Create Task" button at the top of the screen changes to "Create Automation."
I was curious about the difference between Tasks and Automations and noticed this while looking into it. This kind of quirk is perhaps expected in a preview version — it's interesting!
Running a Task
I created and ran a one-time task ("Run Once"). It calculated usage charges for May (before credit application), compiled the results into a PDF report, and sent it to Slack.
The PDF report was delivered to Slack. It was neatly organized and easy to read.
The analysis results were also stored in Artifacts.
Scheduled Execution (Automations)
I set up a task to send a weekly report to Slack every Thursday at 8:00 AM (Wednesday 11:00 PM UTC).
On Thursday, June 11 at 8:02 AM, the aggregated results were sent to Slack. It took 2 minutes for the aggregation and Slack delivery, but I was able to confirm that the scheduled execution worked correctly.
Impressions
Being able to manage costs easily using natural language — without needing to build a cost management agent or dashboard from scratch — and being able to schedule those tasks is really great.
I didn't configure it this time, but with event-triggered execution ("Run when an event occurs"), it seems like you could set up notifications for:
- Which service is causing a sudden spike in costs
- Recommended actions for cost optimization
One thing I'm curious about is what the FinOps Agent usage fee will be once it reaches general availability (GA). Since I use AWS for personal purposes and can't spend too much, whether using FinOps Agent to optimize costs will yield a good return on investment is something I'll be keeping an eye on.
It may be challenging to get impressive results with personal usage, but I still want to find interesting use cases.
Closing Thoughts
Since I use AWS for personal purposes and pay out of pocket (or use credits), I try to be careful about cost management to avoid unexpected charges. Even so, I often end up leaving resources running by mistake and incurring unnecessary costs. With that in mind, I'd like to start leveraging FinOps Agent to better manage and optimize my costs going forward.
I hope this article is helpful to someone out there.
Thank you so much for reading all the way to the end!
Top comments (0)