Trying the Public Preview of AWS FinOps Agent

Introduction

On June 9, 2026, AWS FinOps Agent entered public preview.

According to Announcing the public preview of AWS FinOps Agent, AWS FinOps Agent is "an agentic AI solution that investigates the root cause of cost anomalies and answers cost questions in the tools engineers across your organization already use." I tried it right away, so this article walks through setup, real queries, and actual responses.

References

• AWS FinOps Agent Documentation
• Announcing the Public Preview of AWS FinOps Agent
• FinOps Foundation: What Is FinOps?

What Is FinOps?

The FinOps Foundation defines FinOps in What is FinOps? as follows:

FinOps is an operational framework and cultural practice which maximizes the business value of technology, enables timely data-driven decision making, and creates financial accountability through collaboration between engineering, finance, and business teams.

In short, FinOps combines process and culture to maximize technology value while aligning engineering, finance, and business teams around data-driven cost accountability.

What You Can Do with AWS FinOps Agent

With AWS FinOps Agent, you can:

• Ask cost-related questions in natural language and get investigation reports based on actual cost and usage data
• Send investigation reports to Jira or Slack
• Investigate cost anomalies by using events as triggers
• Configure daily, weekly, or monthly schedules and export recurring cost reports in HTML, PDF, or PPT formats

Because the agent has guardrails configured, it only answers within the FinOps domain, such as AWS cost management and cost optimization.

Setup

During preview, AWS FinOps Agent is available only in US East (N. Virginia) (us-east-1). Also, during preview, usage is free with monthly limits. However, standard charges still apply to AWS APIs the agent calls internally (for example, Cost Explorer APIs).

1. Create an agent

Open AWS FinOps Agent in the AWS Console.

2. Enter the agent name

Enter any agent name and description.

3. Create an IAM role

Choose whether to create a new IAM role or use an existing role. In this example, I chose to create a new role and kept the default role name.

4. Configure IAM permissions for the web app

Grant IAM permissions so the web app can access the agent. This controls what the web app can do with the agent (conversations, task creation, automations, document management, and more). Here, I selected creating a new role and kept the default role name.

5. Configure third-party integrations

Configure integrations with third-party tools. At this time, Jira and Slack are supported. Select the tool(s) to integrate, then choose or enter the required information. Integration requires account-level pre-installation. If not set up yet, you can skip this step and add integrations later from the agent details page after creation.

6. Review the configuration

Confirm your inputs are correct, then click Create agent.

Launch the web app

After agent creation completes, a web app is generated. You can open it in a new tab from Open agent in the console.
Enter a natural-language question in the chat box and press Enter to get a response. You can type queries in Japanese, but according to the documentation, English responses are the default. As shown later in the "Ask in Japanese" section, I actually got a response saying only English responses are supported. So in practice, this currently appears to be English-only rather than "English by default."

Running questions

I asked an actual cost question. This time, I requested a summary of cost trends for LLMs, especially the Anthropic Claude series.

Request a summary of Claude costs

Summarize cost trends about LLM especially Anthropic Claude series.

As shown below, the report summarized costs by month and by model. Because I specifically asked for Claude cost trends, cost data for Cohere embedding models appeared only in comments.

Send to Slack

When I asked it to send the report to Slack, it posted to the Slack channel I configured during third-party integration setup. A practical workflow is to generate a requested FinOps report with the agent and quickly share it in Slack.

In this way, AWS FinOps Agent can send investigation reports directly to Slack.

Export reports in HTML, PDF, and PPT formats

AWS FinOps Agent can export reports in HTML, PDF, and PPT formats. I tried all three.

Export in HTML format

I used the following prompt to request an HTML report:

Summarize cost trends about LLM especially Anthropic Claude series in an executive-ready report in HTML.

Then a link to the HTML report appeared in the Artifacts pane.

Clicking the link opens a preview. The layout is similar to the AWS Management Console UI, and filters are available. You can also download this HTML locally.

Below is the HTML report downloaded locally (excerpt). The downloaded report is a single HTML file including CSS and SVG assets, which makes internal sharing easy. Filters still work in the local file.

Export in PDF format

Next, I requested PDF output:

Summarize cost trends about LLM especially Anthropic Claude series in an executive-ready report in PDF.

Then a link to the PDF report appeared in the Artifacts pane.

Clicking the link opens a preview. Its layout is similar to the HTML report, but because it is PDF, filters are not available. You can also download this PDF locally.

Below is the PDF report downloaded locally.

Export in PPT format

Next, I requested PPT output:

Summarize cost trends about LLM especially Anthropic Claude series in an executive-ready report in PPT.

In this case, the Artifacts pane did not appear automatically.

You can select files from the Artifacts menu and download them from Actions. PPT reports cannot be previewed in the web app.

Below is the PPT report downloaded locally. I opened and checked it in Google Slides.

Download files from Artifacts

You can access Artifacts from the hamburger menu in the top-left corner of the web app.

When you open Artifacts, you can see a list of generated files. From there, you can download or delete files.

Ask in Japanese

When I asked questions in Japanese, it replied that only English responses are supported (I only support English responses).
I asked for an analysis of Amazon Route 53 costs over the last three months, and it answered that all costs were $0. In reality, costs were incurred but fully covered by AWS credits.

After I pointed out that AWS credits were applied, the response was corrected. It would be better if the agent checked credit coverage from the start, but prompt wording likely matters. Also, the official documentation states that cost data accuracy depends on underlying APIs and recommends human review because outputs are probabilistic.

Configure a schedule

I tried one of the example tasks shown on the web app's home screen as-is:

Check my S3 costs daily at 12 PM EST.

The requested task was added on the Automations screen.
Once scheduled, the task runs at the specified time.

Opening the task shows the next run time, trigger type, and task details.

Task schedules can also be configured from the GUI, not only via natural-language requests. You can set one-time execution, recurring scheduled runs, and event-triggered automation for cost anomaly investigations. For example, you can configure a task like: "Monitor AWS cost anomaly detection events, investigate the root cause of each anomaly, and post findings to the #finops-anomalies Slack channel."

Comparison with Amazon Q Developer

I compared this with Amazon Q Developer in AWS Billing and Cost Management. Using the same request as before, I asked for Route 53 cost analysis over the last three months. Amazon Q Developer answered in Japanese to the Japanese question. It also said that "usage may be fully within free tier or promotional credits," pointing out that credits might be applied.

Based only on this example, it is not possible to conclude which is better between AWS FinOps Agent and Amazon Q Developer. Still, this is a useful hint for choosing between them depending on response language and answer characteristics.

Conclusion

I tried the public preview of AWS FinOps Agent. It provides strong support for data-driven decision making advocated by FinOps, including natural-language questioning, scheduling, event-triggered cost investigation, and notifications to third-party tools.

As shown in the Claude cost summary example, I could complete the flow from report generation to Slack sharing through conversation alone. It is also easy to build an automated workflow that investigates root causes when cost anomaly detection events occur and posts results to a team channel. This has real potential to shift cost analysis from individual manual work to a shared, routine team workflow. It could become a foundation for the "collaboration between engineering, finance, and business teams" described in the FinOps definition.

On the other hand, in preview, supported regions are limited to US East (N. Virginia), and responses are limited to English. As seen in the answer that overlooked AWS credit coverage, outputs are probabilistic and should be reviewed by humans. If Japanese responses are required, using Amazon Q Developer for those cases is a practical option.

According to the AWS blog, more capabilities such as AI workload cost analysis are planned. I am looking forward to expanded regional and language support on the path to general availability, and I plan to continue testing practical use cases.