If you run IT operations, someone has probably asked you this year whether an AI agent can take over incident triage. It’s a fair question. The tools got good fast, and most of the big monitoring and cloud vendors now sell one.
We’ve spent years on call for our clients’ infrastructure and databases, so we looked at these agents the way we’d look at a new hire for the night shift. Can you trust what they tell you? Can they reach the systems where your problems actually are? And who picks up the phone when they’re wrong?
Below are five questions worth asking before you choose an AI SRE agent, then a short profile of 12 tools you’re likely to come across.
Key takeaways
- Every agent on this list can investigate an alert and suggest a likely cause within minutes. Very few will change anything without your approval, and that’s a good thing.
- Most are tied to one vendor’s data. Your first question should be where your logs, metrics and tickets already live.
- Paging and escalation still belong to your on-call tool and your people.
- Self-managed Oracle and SQL Server databases and on-prem servers are the most common blind spot.
- We couldn’t find an independent accuracy benchmark for any of these tools. Every accuracy figure we saw came from the vendor.
What is an AI SRE agent?
An AI SRE agent, sometimes called a DevOps AI agent, is software that investigates production alerts for you and suggests a fix. It’s the newest branch of AIOps, the practice of using AI to run IT operations.
Think of it as a fast first responder. When an alert comes in, you get a summary of what looks wrong, the evidence behind it and a suggested next step, often before an engineer has opened a laptop. Some tools will also draft the fix as a pull request or line up a rollback for you to approve.
What you still own is everything around that. Somebody has to decide whether the diagnosis is right, wake the right engineer, keep the ticket honest and follow up until the problem is closed. In our experience that’s where most of the hours in an incident go.
Buyers seem to feel the same way. In Grafana Labs’ 2026 observability survey of 300 IT decision-makers, 85% said they were comfortable with AI doing read-only investigations. Only 52% were comfortable letting it fix things on its own. (Grafana sells one of these tools, so read the numbers with that in mind.)
Five questions to ask before you choose
1. Where does your data live? If all your telemetry is already in Datadog or Dynatrace, that vendor’s agent will know your environment best. If it’s spread across several tools, look at the agents built to read from many sources.
2. Where do your tickets live? Some agents open and update Jira or ServiceNow tickets for you. With others, you’ll build that connection yourself.
3. Can it reach your databases and servers? Many alerts in the environments we manage come down to a blocked session, a full tablespace or a disk on a server outside the cloud. Ask how the agent would see those.
4. What is it allowed to change? Look for approvals on every action, a log of who approved what and a clear list of things the agent will never do.
5. What will it cost in a bad week? Most of these tools charge by usage. A noisy week with hundreds of alerts is exactly when the bill grows.
Agents from the cloud providers
AWS DevOps Agent
A good fit if you run almost everything on AWS. You get investigations that line an incident up against your recent deployments and code changes, plus chat in Slack and Microsoft Teams. Changes to AWS resources are off by default, need an operator’s approval each time and never include deletes. ServiceNow has a built-in integration. Jira isn’t built in, although AWS has published a sample integration you can build on. You pay $0.0083 per agent-second, and queries in other AWS services are billed separately.
Azure SRE Agent
Microsoft’s agent has been generally available since March 2026. It’s the natural choice on Azure, and you can connect it to Datadog, New Relic, Splunk and Dynatrace as well. Incidents can come in from PagerDuty or ServiceNow. You choose between a review mode, where you approve each proposed change, and an autonomous mode. Check the defaults carefully when you set it up. Billing is in Azure Agent Units, with a charge for every hour an agent exists plus usage.
Agents from the monitoring vendors
Datadog Bits Investigation
You may know it by its old name, Bits AI SRE. If your monitoring is in Datadog, you can have an investigation start the moment a monitor alerts. It can open Jira tickets, and one-click actions for Kubernetes and AWS are in preview with an approval step. If you already use Datadog Database Monitoring, your self-hosted Oracle and SQL Server databases are in view too. Datadog bills in AI credits and says an investigation now costs about $4 on average.
Dynatrace Intelligence SRE Agent
Built for teams that already rely on Dynatrace. When a problem is detected you get the likely cause, the business impact and the team that owns it, and a ticket can be opened in Jira or ServiceNow. Fixes are prepared for your approval. Pricing is public: $0.10 per AI unit, and Dynatrace puts a typical investigation at around 30 units, or about $3.
Grafana Assistant Investigations
Generally available since July 2026 and the most open of the monitoring vendors’ agents. You can point it at any data source Grafana can query, including Splunk, Datadog, SQL Server and Oracle. Fixes arrive as pull requests, and you decide tool by tool what needs approval. You pay $20 per active user a month with a token allowance, then $2 per million tokens.
New Relic Autopilot
Formerly called SRE Agent. You get triage and root cause analysis on your New Relic data, in the product or in Slack. Autopilot recommends actions and leaves every change to a person, which some teams will prefer. Pricing is based on compute units, and New Relic doesn’t publish the rate.
Splunk AI SRE
Part of Splunk Observability Cloud and generally available since June 2026. Today you get grouped alerts, a root cause with a confidence level and a step-by-step plan for application and Kubernetes alerts. You run the fix yourself. Splunk doesn’t publish a separate price for it.
Agents from the incident management vendors
PagerDuty Paige
If PagerDuty already runs your on-call, Paige has a head start, because the people side is covered. You can connect more than 20 tools, from Datadog and Grafana to CloudWatch and Azure Monitor. Putting Paige on an escalation policy next to your engineers is in early access. You get it with PagerDuty’s platform plans, and usage is metered in AI Actions.
ServiceNow Otto for ITOM
The new name for Now Assist. Its AIOps AI Specialist became generally available this month. The draw is your CMDB: you get root cause and service impact based on how your systems are actually connected. ServiceNow’s own pages describe the approval step for fixes in different ways, so ask how it works in your release. Pricing isn’t published.
incident.io Investigations
Generally available since August 2026. An investigation starts when an incident is declared, and the findings land in your incident channel. You can connect a long list of monitoring tools, plus MySQL and PostgreSQL databases in your own network. The only change it can make is a pull request that you review and merge. It’s an add-on to the Pro and Enterprise plans, with no public price.
Independent AI SRE agents
Resolve AI
One of the best-funded newcomers. You can connect most popular monitoring, cloud and code tools, and keep the on-call rotation you already have in PagerDuty. Pricing is credit-based and set in your contract, so you’ll need a sales conversation to get a number.
Cleric
A simpler option with prices on the website. Access to your infrastructure is read-only, fixes come as pull requests and you can create Jira tickets from its findings. Plans start at $700 a month, and an investigation works out to about $5.
How the 12 tools compare
| Tool | Best fit | Who makes the change | Published price |
|---|---|---|---|
| AWS DevOps Agent | AWS-native teams | You approve each AWS action | $0.0083 per agent-second |
| Azure SRE Agent | Azure teams | You approve, or autonomous mode | Azure Agent Units |
| Datadog Bits Investigation | Datadog customers | You approve (preview) | About $4 per investigation |
| Dynatrace Intelligence SRE Agent | Dynatrace customers | You approve | About $3 per investigation |
| Grafana Assistant Investigations | Mixed data sources | You merge the pull request | $20 per user a month, plus tokens |
| New Relic Autopilot | New Relic customers | You do | Not published |
| Splunk AI SRE | Splunk Observability customers | You do | Not published |
| PagerDuty Paige | PagerDuty customers | You approve | Included in platform plans |
| ServiceNow Otto for ITOM | ServiceNow customers | Ask ServiceNow | Not published |
| incident.io Investigations | Slack-based incident teams | You merge the pull request | Not published |
| Resolve AI | Multi-tool environments | You approve | Not published |
| Cleric | Smaller teams that want clear pricing | You merge the pull request | From $700 a month |
| Boston SoftDesign AIOps Platform (ours) | Cloud, hybrid and on-prem environments that want a 24/7 team included | Our engineers, with your approval | By quote |
What you still need people for
After reading through all 12, the same gaps kept coming up, and they’re the ones we deal with every night:
- Paging and follow-up. Unless your on-call tool is also your agent vendor, somebody still has to route the alert, wake an engineer and make sure the ticket gets closed.
- Databases and servers outside the cloud. Only a few of these tools can see a self-managed Oracle or SQL Server database, and usually only if you already run that vendor’s monitoring on it.
- Trust. A confident answer can still be wrong. You’ll want the evidence next to every finding and an engineer who approves each change.
The managed option: Boston SoftDesign AIOps Platform
We should be upfront that this one is ours. Boston SoftDesign is a managed services provider, and those three gaps are the work our team does every night for client infrastructure, databases and applications.
Our AIOps Platform is the tool our engineers use to do it, and you get both together:
- Every alert is investigated the moment a ticket is created. You keep the monitoring tools and Jira you already have.
- One team for your whole environment. We manage cloud, hybrid and on-prem environments. That means cloud services, clusters and containers, and Oracle, SQL Server, PostgreSQL and MySQL databases running on VMs and servers.
- You don’t have to modernize first. If part of what you run is older infrastructure, you still get modern DevOps practices and AI across all of it, at the pace that suits you.
- You can check every finding. Each one comes with the metrics, logs or knowledge base article behind it.
- Nothing changes without an engineer’s approval. The platform is read-only by default.
- A 24/7 team owns the incident. Our engineers handle paging, escalation and follow-up, so a diagnosis never sits unread.
“AI triage and investigation cut the time to resolution on every incident. The engineer sees the full picture the moment an incident appears and can start fixing it right away. Combined with proactive monitoring, it lets us stop problems before our clients’ end users ever notice them.”
Gleb Klimov, Infrastructure and Support Operations Lead, Boston SoftDesign
Here’s what that looked like for one client, a US education company whose AWS, Oracle and SQL Server environment we support around the clock. Over the first few months:
- Median time to a root cause went from 5.3 hours to under one hour.
- Average time to resolve an alert dropped 49%, from 17.9 hours to 9.1.
- Critical alerts were resolved 84% faster, in 3.4 hours on average, down from 21.5.
- 61% of alerts were investigated within one minute, even with 20% more alerts coming in.
If you’re weighing one of these agents and want a second opinion from people who do this work every day, talk to our team. You can also read more about our managed services.
Frequently asked questions
What is an AI SRE agent? An AI SRE agent, also called a DevOps AI agent, is software that investigates production alerts for you. You get a likely root cause, the evidence and a suggested fix, usually within minutes of the alert.
Which AI SRE agent is best? It depends on where your data is. If you’re all-in on one monitoring or cloud vendor, start with that vendor’s agent. If your data is spread across tools, look at Grafana Assistant, PagerDuty Paige, incident.io, Resolve AI or Cleric, which are built to read from many sources.
Can AI agents fix incidents on their own? A few can, in limited cases. Most tools either leave the fix to you or ask for approval before every change, and we’d recommend keeping it that way until you’ve seen how accurate the agent is in your own environment.
How much do AI SRE agents cost? Most charge by usage. Published prices range from about $3 to $5 per investigation (Dynatrace, Datadog, Cleric) to per-second billing (AWS). Several vendors don’t publish a price at all.
Do AI SRE agents work with on-prem servers and Oracle or SQL Server databases? Sometimes. Datadog and Grafana can see self-managed Oracle and SQL Server databases through their own monitoring and data sources. Several others reach private systems through a proxy you run in your network, and some don’t document it at all. Ask before you buy.
What is the difference between AIOps and an AI SRE agent? AIOps is the broader practice of using AI in IT operations, such as cutting alert noise and spotting patterns. An AI SRE agent is a newer kind of AIOps tool that works through a single incident step by step, the way an engineer would.
Is there a managed AIOps service? Yes. Boston SoftDesign offers its AIOps Platform together with a 24/7 operations team for cloud, hybrid and on-prem environments, so the investigation, the escalation and the fix are all covered by one partner.
Will an AI SRE agent replace our on-call engineers or our MSP? No. You’ll spend less time on the first investigation, and that’s a real gain. On-call coverage, escalation, approvals and the fix itself still need people.
Product names are trademarks of their respective owners. Boston SoftDesign is not affiliated with or endorsed by the vendors mentioned. Product details and pricing were checked on November 17, 2026 and may change.