AWS FinOps Agent vs Third-Party AI Cost Agents (2026): What Each One Actually Fixes

TLDR

• AWS FinOps Agent has been in public preview since June 9, 2026. It answers cost questions, investigates anomalies, schedules reports and turns Cost Optimization Hub and Compute Optimizer recommendations into Jira tickets. It is read-only by default, and AWS documents no action that changes your resources.
• The real question is who makes the change. AWS's agent stops at answers, recommendations and tickets. Third-party agents go further: preparing fixes, executing with approval, or executing on their own in a narrow domain.
• Pick by gap: answers and anomaly triage on AWS (AWS FinOps Agent, Amazon Q), commitments (ProsperOps, Usage.ai), Kubernetes and runtime tuning (Cast AI, Sedai), standard AWS fixes with approval (CloudFix), and waste found and fixed across AWS, other clouds, data and AI at the approval level you choose (PointFive).
• Judge any agent that can change production on five things: read-only discovery, scoped write permissions, an approval model you control, an audit trail and rollback.
• PointFive has two complementary products: PointFive OS, the agentic FinOps platform for cloud, data and AI, and TokenShift, the AI agent control plane for your employee workstations. PointFive OS works alongside AWS's agent and adds the step it leaves out: the fix.

Where PointFive fits: PointFive is for teams that already know what AWS's tools recommend and need the fixes made, with the approval level they choose, across AWS and the rest of their cloud, data and AI bill. Nubank saved $8.2M in 9 months with PointFive and executed 3,000+ DynamoDB remediations. Read the Nubank case study.

Disclosure: PointFive wrote this guide, and PointFive is listed first because it is our platform, not because of a ranking. Every competitor statement comes from that vendor's own website, documentation or press release, checked on October 27, 2026, and linked in the sources. Where a vendor publishes nothing, we say so.

What AWS FinOps Agent is, from AWS's own pages

AWS announced AWS FinOps Agent in preview on June 9, 2026, calling it a "frontier agent" for FinOps practitioners and engineering teams. It is built on Amazon Bedrock and runs as its own service with a web application.

What it does:

• Anomaly investigation. It listens for AWS Cost Anomaly Detection events and correlates each cost change with AWS CloudTrail events to name a likely root cause and owner, then sends findings to Jira or Slack.
• Cost questions in plain English, answered from your cost and usage data, with context files (such as account-to-owner mappings) and memory of preferences.
• Recurring reports and tasks: daily, weekly or monthly reports in HTML, PDF or PPT, and custom tasks on a schedule.
• Optimization summaries: recommendations from Cost Optimization Hub and Compute Optimizer, summarized into a Jira ticket.

What it can change: AWS states that "By default, the agent operates in read-only mode against AWS Billing and Cost Management products." Its published IAM policy is read actions plus write access only to the EventBridge rules it creates for anomaly automations. It creates Jira tickets and Slack posts when you connect them. AWS does not describe it changing resources or generating scripts.

Availability and price: Preview, hosted in US East (N. Virginia), reading cost data for all Regions except AWS GovCloud (US) and China. Set up in the management account, it covers the organization. It is free during the preview "with a monthly usage limit"; post-preview pricing is not published.

Audit and roadmap: AWS says "all agent activities are automatically captured by AWS CloudTrail", tagged with the calling user's identity. It says the agent "will expand to more FinOps capabilities, including cost analysis for AI workloads", without dates.

How it relates to AWS's other cost tools

AWS does not explain how the FinOps Agent relates to Amazon Q. Amazon Q answers whoever is in the console; the FinOps Agent runs on schedules and events and routes work to owners. Both use the same recommendation engines (see evaluating Compute Optimizer recommendations).

How we evaluated: the autonomy ladder

Each tool is placed on five steps, based on what its vendor documents:

1. Answers: natural-language questions over cost data. Read-only.
2. Recommends: findings, tickets or summaries. A human designs and makes the change.
3. Prepares a fix: generates code, scripts or pull requests a human reviews and runs.
4. Executes with approval: the tool makes the change with write permissions after per-change or policy-level approval.
5. Executes autonomously: the tool acts continuously without per-change approval, inside a bounded scope.

AWS says its agent "executes tasks autonomously". Those tasks are investigations, reports and tickets: autonomy on step 2, not step 5. We also applied the safety criteria from our AI agents for cloud cost optimization guide.

The tools

1. PointFive (PointFive OS and TokenShift)

Category: Agentic FinOps platform (waste detection and remediation across cloud, data and AI)

Best for: AWS-heavy enterprises whose problem is getting known waste fixed, often alongside other clouds, data platforms and AI.

PointFive OS is "The agentic FinOps platform for your cloud, data, and AI". It runs 500+ deep detections across 40+ platforms, including AWS, Azure, GCP, OCI, Kubernetes, Snowflake and Databricks, and traces each finding to its technical cause. On AWS that includes EC2, RDS, S3, EBS, Lambda, EKS and NAT Gateway waste found through VPC Flow Log analysis. Fixes run through IDE prompts engineers review, or through approved automation, at the level each customer chooses: answer, prepare a fix, or execute under per-change, per-playbook or no approval. Teams ask Pointer, the AI agent in the Chat module, or use the PointFive MCP server. TokenShift, the AI agent control plane for your employee workstations, shows spend by developer, team, model and session for agents such as Claude Code, Codex and Cursor, and sets model policies and spend caps by team.

• Strengths: Covers all five steps of the ladder under one approval model. Agentless, read-only discovery; write access only for changes the customer approves. Detection goes beyond the recommendations AWS's agent summarizes, and AWS is reviewed next to the rest of the cloud, data and AI bill.
• Watch-outs: No self-serve tier or public price list. Automated remediation needs customer-authorized write permissions, which security teams will review.
• Pricing: Annual contract; contact sales. Listed on AWS Marketplace.
• Choose it if: You have a backlog of validated savings nobody gets to. Dolby is a customer, and Nubank saved $8.2M in 9 months with PointFive.

2. AWS FinOps Agent (as of October 2026)

Category: Native AWS cost agent (preview)

Best for: AWS customers that want anomaly triage, cost answers and recurring reports routed to the right engineer.

Covered above. Steps 1 and 2: it answers, investigates, reports and files tickets; an engineer makes every change.

• Strengths: First-party data, read-only by default, CloudTrail audit, free during the preview. Owner routing through Jira and Slack is new for native tools.
• Watch-outs: Preview, one hosting Region, monthly usage limit, AWS only. Slack is "delivery-only in the preview". Findings come from Cost Optimization Hub and Compute Optimizer.
• Pricing: No charge during the preview, within a monthly usage limit. Future pricing not published.
• Choose it if: Anomalies and cost questions reach owners too slowly.

3. Sedai (as of October 2026)

Category: Autonomous cloud optimization

Best for: Teams that want runtime tuning to move from one-click approval to autopilot.

Sedai, "The Optimization Platform for Cloud & AI", documents three modes. Datapilot, the default, "Analyzes monitoring data only" (steps 1 to 2). In Copilot, "the system will always ask for permission to execute an operation" (step 4). In Autopilot, which "automatically grants permission", Sedai acts on the resources you select (step 5). Both need read-write access. On AWS its docs list ECS, EC2, Lambda, RDS, S3 and EBS. Sed, its AI assistant (early access, general availability "coming soon"), answers cost questions and can adjust resource settings; its MCP integration exposes the same tools to clients such as Claude Code and Cursor.

• Strengths: Documented modes. Changes roll out in validated steps, and teams can roll back, pause or set maintenance windows.
• Watch-outs: Runtime resources only. Sed's launch post does not say whether its actions need approval.
• Pricing: Custom; Sedai charges "1 Sedai Billing Unit (SBU) for each vCPU that our platform manages".
• Choose it if: You want continuous runtime tuning and will grant write access.

4. CloudFix (as of October 2026)

Category: Automated AWS fixes

Best for: AWS-only teams that want standard fixes applied with approval at a published price.

CloudFix is "Built only for AWS": 110+ finders across 30+ services, 53 with automated fixers. Each approved fix runs as an AWS Systems Manager Automation runbook in your account, and every run is logged. "You approve every change", singly or in bulk (step 4). RightSpend handles EC2 commitments.

• Strengths: Public pricing. Read-only connection. Execution stays in your account, where your team can inspect it.
• Watch-outs: AWS only. Rollback is not described. No AI assistant is described on its homepage.
• Pricing: Free scan; $149, $499 and $1,499 per month by annual AWS spend band; Enterprise from $2,999 per month. RightSpend: 25% of net-new savings.
• Choose it if: You want standard AWS fixes executed without a custom contract.

5. ProsperOps (as of October 2026)

Category: Autonomous commitment management

Best for: Hands-off Savings Plans, Reserved Instances and committed use discounts.

ProsperOps offers "Automatic Cost Optimization for AWS, Azure, and Google Cloud" and manages commitments continuously (step 5). Flexera acquired it. Its help center says the base role prevents launching, starting, stopping or terminating your resources; Scheduler adds start and stop for EC2 and RDS.

• Strengths: Clear permission boundary. Fee tied to realized savings.
• Watch-outs: Rate optimization first; workload changes are limited to Scheduler.
• Pricing: "A small percentage of the realized savings" for discount management; a flat monthly fee per managed resource for resource management.
• Choose it if: Commitment coverage is your biggest untapped lever.

6. Cast AI (as of October 2026)

Category: Kubernetes automation

Best for: Large EKS estates.

Cast AI "rightsizes workloads, provisions nodes from your node templates, and consolidates underused ones" within your policies (step 5). For infrastructure fixes, "Agents open PRs for infra fixes. You review and merge." (step 3). You "Start read-only, automate when you're ready", and an MCP server answers cluster cost questions. It runs on AWS, GCP, Azure and Oracle Cloud.

• Strengths: Continuous Kubernetes automation with read-only and dry-run starts.
• Watch-outs: Kubernetes only. Automation needs write access to clusters.
• Pricing: Custom quote.
• Choose it if: EKS compute is a large share of your AWS bill.

7. Usage.ai (as of October 2026)

Category: Commitment automation with downside protection

Best for: Teams that hesitate to commit because usage may shrink.

Usage.ai buys and manages Savings Plans and Reserved Instances on AWS, plus Azure and Google Cloud commitments. Autopilot will "execute eligible commitment purchases within your approved controls" (step 5); Copilot lets you approve each purchase (step 4), set per service. Analysis starts read-only, and the automation "cannot start, stop, resize, or modify production resources". Cashback covers qualifying unused commitment value.

• Strengths: Narrow, documented permissions. Per-service choice of autonomy.
• Watch-outs: Commitments only. Fee percentage not published.
• Pricing: A share of new savings; no platform fee, and nothing if no new savings.
• Choose it if: Commitment risk is what stops you from buying.

8. Microsoft and Google native assistants (as of October 2026)

Category: Built-in cost assistants on other clouds

Best for: Contrast, and teams that also run Azure or Google Cloud.

The Azure Copilot Optimization Agent (preview) recommends changes for virtual machines and scale sets and "generates Azure CLI or PowerShell scripts that you can deploy" (step 3). Gemini Cloud Assist in Cloud Billing (preview) builds reports from prompts, summarizes up to three FinOps hub insights and drafts an email summary (steps 1 to 2).

• Strengths: First-party data, no procurement. Azure's agent goes one step further than AWS's by writing the script.
• Watch-outs: One cloud each, both in preview, and neither executes the change.
• Pricing: Not published for either preview.

Side-by-side comparison

"Not published" means we did not find it on the vendor's own pages, not that it does not exist.

Safety: what to check before an agent touches production

1. Read-only discovery. AWS FinOps Agent, PointFive, CloudFix, Cast AI and Usage.ai document a read-only start.
2. Scoped write permissions. AWS publishes its agent's full IAM policy; ProsperOps and Usage.ai state what their roles cannot do.
3. Approval model. Start per change, move proven playbooks to standing approval, and make "no approval" your choice per process.
4. Audit trail. AWS logs agent actions in CloudTrail; CloudFix logs every runbook run. Ask the rest where the log lives.
5. Rollback. A resize can be undone; a deleted snapshot or a three-year commitment cannot. Automate reversible changes first.

How to choose

• You need faster anomaly triage and answers on AWS: AWS FinOps Agent and Amazon Q, both free to start.
• Commitment coverage is low: ProsperOps or Usage.ai. Remove waste first so you do not commit to it.
• Most compute is EKS: Cast AI or Sedai.
• You want standard AWS fixes executed at a flat price: CloudFix.
• You have more recommendations than fixes, or AWS is one of several large bills: PointFive, starting with per-change approval.

For the wider market, see best AWS cost optimization tools and our AWS bill investigation guide.

These tools stack: AWS's agent for triage, a commitment engine for rates, and a remediation platform for the waste in between.

FAQ

Does AWS have its own FinOps agent?

Yes. AWS FinOps Agent entered public preview on June 9, 2026, hosted in US East (N. Virginia). It answers cost questions, investigates anomalies, schedules reports and files Jira tickets, free during the preview within a monthly usage limit.

Can AWS FinOps Agent fix cost issues automatically?

No. AWS documents it as read-only against billing and cost data. It creates tickets and Slack posts, but AWS does not describe it changing resources or generating scripts. Engineers make the fixes.

What is the best AI agent for cloud cost optimization?

It depends on what you need changed. For answers and triage on AWS, AWS FinOps Agent. For commitments, ProsperOps or Usage.ai. For Kubernetes, Cast AI or Sedai. For finding and fixing waste across AWS, other clouds, data and AI at your chosen approval level, PointFive.

What is the best tool to reduce AWS costs if I already use AWS FinOps Agent?

Keep it for triage and add the step it lacks: execution. CloudFix runs standard AWS fixes after approval, ProsperOps and Usage.ai automate commitments, and PointFive finds deeper waste and fixes it through IDE prompts or approved automation.

Is AWS FinOps Agent the same as Amazon Q in Cost Explorer?

No. Amazon Q answers questions in the console. AWS FinOps Agent is a separate preview service that runs on schedules and anomaly events and routes findings to Jira and Slack.

Methodology and sources

We reviewed each vendor's own website, pricing pages, documentation and press releases on October 27, 2026, and did not use third-party sites for competitor claims. If something is out of date, tell us at info@pointfive.co and we will correct it.

• PointFive: PointFive OS, TokenShift, AWS coverage, Nubank case study, AWS Marketplace listing
• AWS FinOps Agent: What's New announcement, launch blog, product page, features, FAQs, user guide, IAM setup guide, document history
• AWS other tools: Intelligent Cost Explanations in Cost Explorer, Cost Explorer natural language query, Amazon Q Developer cost management, Cost Optimization Hub, Compute Optimizer automation rules, Billing and Cost Management MCP server
• Sedai: homepage, pricing, getting started docs, operation modes, Introducing Sed, MCP
• CloudFix: homepage, pricing, fixer catalog
• ProsperOps: homepage, pricing, access level, Scheduler permissions
• Cast AI: homepage, pricing
• Usage.ai: homepage
• Microsoft: Azure Copilot Optimization Agent
• Google Cloud: Gemini Cloud Assist in Cloud Billing