The best AI SRE tools in 2026

15 products, from their vendors' documentation

• AWS DevOps Agent

  Amazon Web Services · cloud provider

  Generally available since March 31, 2026, after a preview from December 2025

  AWS DevOps Agent investigates incidents that arrive from ticketing, paging and observability tools or from a manual request, across AWS accounts, Azure resources and connected tools. It produces a mitigation plan or a spec that a coding agent such as Kiro can implement. AWS previewed it in December 2025 and made it generally available on March 31, 2026.

  Changes are off by default. With directed actions enabled, each approval covers one tool, operation and resource, for a single use or up to four hours, and CloudTrail records who approved it. A change attempted during an autonomous investigation fails instead of asking. Memory stores keep recurring root causes per alarm, standing directives and lessons from corrective feedback.

  Worth checking

  Cost follows agent time, about $30 per agent-hour, and AWS's own example of ten 8-minute investigations comes to $39.84. Queries it runs against services such as CloudWatch Logs Insights are billed separately, and each Agent Space runs three investigations at once by default.

  Sources: AWS DevOps Agent, User guide, Directed actions, Pricing, GA announcement

• Datadog Bits Investigation

  Datadog · observability platform

  Generally available since December 2025, launched as Bits AI SRE

  Datadog's investigation agent forms hypotheses about a root cause and queries Datadog telemetry to confirm or rule out each one, then posts its findings to the Slack, case or on-call destinations already set on the monitor. Datadog made it generally available on December 2, 2025 as Bits AI SRE and renamed it Bits Investigation in mid-2026, alongside Bits Code and Bits Chat.

  Datadog's March 2026 update put a typical investigation at three to four minutes. By default each monitor starts at most one automatic investigation in a rolling 24 hours, a limit admins can change, and warn, no-data, renotification and test events do not start one. Bits Infrastructure Operations, in preview, applies infrastructure fixes automatically where a guardrail allows it, for example in staging, and asks for approval elsewhere.

  Worth checking

  Investigations, chat messages (about 0.5 credits each) and code fixes (about 5) all draw on the same monthly credit bundle, and Datadog describes these figures as averages. Unused credits do not roll over, overage bills at the on-demand rate, and one-click infrastructure actions, guardrails, Bits Detection, Bits Infrastructure Operations and automatic memories are still in preview.

  Sources: Bits Investigation docs, Bits Remediation docs, AI credits pricing, GA announcement, Bits Infrastructure Operations. Full comparison with Edge Delta

• Dynatrace SRE Agent

  Dynatrace · observability platform

  Preview; the SRE Agent workflow template shipped in late July 2026, inside Dynatrace Intelligence, which launched in January 2026

  At Perform in January 2026, Dynatrace introduced Dynatrace Intelligence, the successor to Davis AI, and added domain agents and agentic workflows to its platform. The SRE Agent, a workflow template that shipped in late July 2026, is one of eleven ready-made agentic workflows in the documentation, all marked preview. It picks up a problem after Dynatrace has run its own root-cause analysis, queries logs, events, metrics and traces with DQL, and writes its analysis and recommendations back onto the problem.

  The default workflow is capped at five tool calls and a 1,000-word answer, and it does not apply a fix by itself. In July 2026 Dynatrace announced an Autonomous SRE Agent and a Cloud SRE Agent. The Cloud SRE Agents app hands problems to agents from AWS, Microsoft and Google, and its Hub listing describes it as a community-supported project.

  Worth checking

  The SRE Agent is marked preview in the documentation and coming soon in the Dynatrace Hub, so confirm what your tenant can use. We found no release note confirming that the Autonomous SRE Agent shipped after its August target.

  Sources: Dynatrace Intelligence docs, SRE Agent in the Dynatrace Hub, January 2026 announcement, July 2026 announcement, SRE Agent docs, Agentic AI licensing FAQ, Rate card

• Edge Delta

  Edge Delta · telemetry pipelines and AI SRE

  Available since October 2025, with a self-serve trial

  Edge Delta is a telemetry-native AI SRE that continuously understands, investigates, and operates production. Its AI Teammates (SRE, Security Engineer, Software Engineer and Work Tracker) reason over the logs, metrics, traces and events moving through Edge Delta's Telemetry Pipelines, so baselines and service context already exist when an incident starts.

  An investigation ends with a root cause and its evidence, and proposed fixes wait in an approvals queue. After a fix runs, the teammate compares telemetry from before and after the change, and the incident is saved to Production Memory for later investigations.

  Worth checking

  Detection on live data and the fullest context depend on telemetry passing through Edge Delta pipelines; data that stays in another backend is reached through connectors, so the teammates see what those tools return. Pro usage beyond the included $20 draws on credits whose cost per investigation is not published, and Pro keeps data and memory for 30 days.

  Sources: Edge Delta AI SRE, Edge Delta pricing

• Google Gemini Cloud Assist investigations

  Google Cloud · cloud provider

  Preview since June 2025; new investigations need Premium Support or account-team access since April 2026

  The investigations feature in Gemini Cloud Assist analyzes logs, metrics and configuration for Google Cloud resources and returns root-cause hypotheses with suggested next steps. It entered public preview in June 2025, and since April 10, 2026 creating or running an investigation requires a Premium Support contract or access through your Google Cloud account team.

  Investigations run with the permissions of the person who starts them, and the documentation states that this access is never used to change data. Proactive Mode, in private preview for Premium Support customers, investigates Cloud Alerting alerts and cost anomalies in the background and publishes results as Eventarc events.

  Worth checking

  It covers Google Cloud only, one project or App Hub application per investigation. Google advises against using it on data with residency requirements, because investigation data may be stored in any Google Cloud data center, and investigations stopped working inside VPC Service Controls perimeters on April 13, 2026.

  Sources: Investigations docs, Release notes, Gemini Cloud Assist, Gemini pricing

• Grafana Assistant Investigations

  Grafana Labs · observability platform

  Generally available since July 2026

  Grafana Assistant became generally available in Grafana Cloud in October 2025, and its Investigations mode followed on July 29, 2026. An investigation explores metrics, logs, traces and profiles, builds hypotheses, and returns a report with findings, supporting evidence and recommended next steps.

  By default, alert enrichment opens at most 50 new investigations every 20 minutes, and changes to the same alert group within six hours continue the existing investigation. An investigation can schedule its own re-checks, for example whether an error rate has recovered, and close them once the situation resolves. Metering for investigations began on October 1, 2026.

  Worth checking

  Investigations need a separate entitlement and are not available on self-managed Grafana, even when it connects to the Cloud Assistant backend. MCP tools set to auto-approve can write to or delete data in connected services without asking.

  Sources: Investigations docs, GA release note, Grafana pricing, MCP server settings

• Harness AI SRE

  Harness · software delivery platform

  Sold as an Enterprise plan module; no GA date published

  Harness AI SRE combines incident management and on-call with AI agents. AI Scribe records what happens in Slack, Teams and Zoom calls, the RCA Change Agent ranks recent deployments, pull requests and ServiceNow change records as likely causes, and runbooks can roll back a deployment by running a Harness pipeline, scale out, or flip a feature flag.

  On-call schedules, escalation policies, paging and a mobile app sit behind a feature flag, with import from PagerDuty, Opsgenie and xMatters. Alerts are enriched with similar past incidents and how they were resolved, AI post-mortems can be drafted when an incident closes, and ServiceNow change records are polled every five minutes.

  Worth checking

  Runbooks on automatic triggers run without a person, and Harness recommends running production rollbacks manually. Runbooks exposed to the AI investigator, which run automatically or with a person in the loop to feed root-cause analysis, are in early access, and correlation with feature flag, infrastructure and config changes sits behind a feature flag.

  Sources: Harness AI SRE, Harness pricing

• incident.io Investigations

  incident.io · incident management

  Generally available since August 5, 2026

  Investigations is incident.io's AI SRE, sold as an add-on to its incident management platform. When an incident is declared, it queries connected telemetry, code and past incidents, then posts a root-cause hypothesis with a confidence level and linked evidence. It runs on Nexus, which incident.io calls its production intelligence model; each customer gets a separate instance, and the underlying models come from OpenAI, Anthropic and Google under zero-data-retention terms.

  In September 2026 incident.io reported that the median time to the first accurate message in an incident channel fell from 6.7 to 3 minutes. An escalation step can hold alert-triggered paging until the first hypothesis exists, up to a time limit you set.

  Worth checking

  It works only on incidents, so an alert reaches it only after a workflow or a person turns it into one. The product page says sensitive data is redacted before it reaches model providers, while the documentation describes pattern-based redaction that stays off until you ask incident.io to enable it.

  Sources: incident.io AI SRE, Triggering docs, incident.io pricing, GA changelog

• Microsoft Azure SRE Agent

  Microsoft · cloud provider

  Generally available since March 10, 2026, after a preview from May 2025

  Microsoft's agent takes incidents from Azure Monitor, PagerDuty or ServiceNow, runs scheduled and webhook-triggered tasks, and investigates across Azure and connected tools before proposing or applying a mitigation. Microsoft previewed it at Build in May 2025 and made it generally available on March 10, 2026.

  By default the managed identity gets Reader and Monitoring Contributor roles, with a Privileged level offered at creation; delete and Key Vault commands are blocked outright, and administrators can add tool policies and hooks on top. Review mode pauses only Azure infrastructure writes, and new incident response plans and scheduled tasks default to Autonomous, including the quickstart plan it creates for high-severity alerts.

  Worth checking

  The always-on charge runs from creation until deletion, even while the agent is stopped, which comes to about $292 a month per agent in US regions before any investigation. A 30-day trial for new customers waives it for up to three agents.

  Sources: Azure SRE Agent overview, Azure SRE Agent pricing, GA announcement

• New Relic Autopilot

  New Relic · observability platform

  Generally available since July 2026; previewed as SRE Agent from February 2026

  New Relic previewed an SRE Agent in February 2026 and made it generally available as Autopilot in July 2026. It attaches an analysis to alerts routed to it through a notification workflow and answers questions in the New Relic AI panel or in Slack, using telemetry from the New Relic accounts you select.

  Memories, added in July 2026, are off by default and kept for 90 days unless you extend them to as long as a year. Each organization runs one Autopilot, limited to 100 requests an hour, and an investigation looks back over the last 24 hours by default.

  Worth checking

  Autopilot does not create alerts, so its coverage follows the alert conditions you already maintain in New Relic. The GitHub connection is in public preview, and Jira, named in the launch announcement, does not appear in the documentation.

  Sources: New Relic Autopilot, Autopilot docs, GA release note, Memories docs

• PagerDuty SRE Agent

  PagerDuty · incident management

  Generally available since October 30, 2025; called Paige in current docs

  PagerDuty's SRE Agent, now called Paige in its documentation and pricing, works inside the incident. It pulls logs, metrics, runbooks and incident history from connected tools, suggests likely causes, and recommends an incident workflow for a responder to run. PagerDuty made it generally available on October 30, 2025.

  Memory is kept per service. Paige saves what it learned from an incident once the incident resolves, keeps uploaded runbooks for later conversations, and a Memory API can view, update or redact what it holds. When an escalation policy triggers the agent, it works alongside the paged human and does not delay the page.

  Worth checking

  Each request, nudge or automatic trigger uses 4 AI Actions and per-tier allotments are not published, so estimate a month of usage before committing. The agent reads only the first 2,000 characters of an alert's custom details and of each incident note.

  Sources: SRE Agent product page, SRE Agent docs, PagerDuty pricing, GA changelog

• Resolve AI

  Resolve AI · dedicated AI SRE

  Sales-led; the trial button opens a demo request

  Resolve AI was founded by Spiros Xanthos and Mayank Agarwal, who co-created OpenTelemetry and led Splunk Observability. It picks up alerts and incidents from the tools you already run and investigates them with teams of agents that pursue several hypotheses at once. In deep investigations, a separate verifier checks each finding against production evidence before it is shown.

  Alerts run in one of three modes: triage, an adaptive mode that sets the depth from the alert's impact, runbooks and past handling, or a full investigation with proposed fixes. Background agents watch the deploy stream and flag regressions in error rate, latency or saturation before an alert fires. The company has raised more than $190 million, most recently at a $1.5 billion valuation in April 2026.

  Worth checking

  Each chat, triage, investigation, incident and background task draws credits at the rates in your contract. Reaching a monthly cap pauses automatic work, while work a person starts continues.

  Sources: Resolve AI product overview, Mitigation actions docs, Billing and credit controls, Background agents, Agent Teams. Full comparison with Edge Delta

• Rootly AI SRE

  Rootly · incident management

  Enabled per team by Rootly's account team; no launch date published

  Rootly AI SRE investigates alerts and incidents through the observability, cloud and code tools connected to Rootly. It reports a root cause with a high, medium or low confidence level only when Rootly's checks on the recorded evidence pass; otherwise the result is labeled inconclusive, and teams on Rootly's six-stage flow can also see contributing factor or blocked.

  It uses Claude Sonnet 4.6 by default with GPT-5 as a fallback, scrubs personal data from outgoing prompts, and runs inference in the US. Investigation rules can run manually or automatically, or be paused, and a 90-day preview shows which past alerts a rule would have matched.

  Worth checking

  Rootly's marketing says every change needs human sign-off, but its documentation says not to treat AI SRE as read-only, because connector and custom MCP tools can change data without pausing for approval. The AI SRE tab has no cancel control, and a run cannot be started from a workflow or the API.

  Sources: Rootly AI SRE, AI SRE docs, Running an investigation, Rootly pricing

• Splunk AI SRE

  Splunk, a Cisco company · observability platform

  Generally available since June 2026

  Splunk's name for the product is AI SRE in Splunk Observability Cloud. When someone opens a supported alert, its troubleshooting agent ranks suspected root causes with a confidence level and evidence, and for Kubernetes alerts it adds a step-by-step remediation plan. The agent reached general availability in June 2026, first in the us1 realm.

  Analysis runs once per user, so returning to an alert shows the earlier result immediately. In September 2026 Splunk added an alpha, built on Claude Managed Agents, that proposes a code fix, opens a pull request for review, confirms afterward that the fix worked, and keeps reviewer feedback in memory for later incidents.

  Worth checking

  Root-cause analysis covers APM service, business transaction and Kubernetes alerts. Alert grouping into incidents is in controlled availability in the us1 realm, and the troubleshooting agent is not offered in government realms.

  Sources: Splunk AI SRE, Troubleshooting agent docs, September 2026 update

• Traversal

  Traversal · dedicated AI SRE

  Sold since its June 2025 launch; Incident Workers generally available since September 2, 2026, Alert Workers in public beta

  Traversal keeps a Production World Model of your system, rebuilt continuously from baselines, dependencies, changes and team knowledge. Its Causal Search Engine tests thousands of root-cause hypotheses in parallel and discards any that do not fit the dependencies, timing and spread of the failure. Workers join Slack or Teams channels to investigate incidents and triage alerts.

  Traversal's documentation describes the platform as always read-only, though its AWS role is not limited to reads and write-capable MCP tools are yours to block. Its homepage presents self-healing as automated remediation, which the documentation does not describe. Named customers include American Express, DigitalOcean and Cloudways, and Traversal reports 38% lower MTTR at DigitalOcean.

  Worth checking

  Kubernetes and private data sources need the outbound-only Traversal Connector running in your environment, deployed with Helm on Kubernetes or as a container elsewhere. Alert Workers are still a beta you request through Traversal.

  Sources: Traversal docs, Architecture, Workers GA announcement, Traversal pricing, Causal Search Engine. Full comparison with Edge Delta

How this guide was put together, and what it does not show

1. 01Edge Delta publishes this guide and sells one of the products in it. Products are listed alphabetically, and every entry links to the vendor pages it was written from.
2. 02We read each vendor's product pages, documentation, pricing pages and release notes from September 30 to October 2, 2026. Where marketing pages and documentation disagreed, the entry follows the documentation and says so.
3. 03The guide covers AI SRE products from established observability, incident management, cloud and software delivery vendors, plus Resolve AI and Traversal, the two dedicated AI SRE companies Edge Delta compares against in depth. Earlier-stage startups are not covered.
4. 04A hollow mark in Figure 1 means we could not find the capability in public documentation. A product may do more than its documentation says. Features in preview, early access or alpha count as documented, and where a marketing page claims something the documentation contradicts, the mark follows the documentation.
5. 05We have not run most of these products. The one hands-on measurement is AI SRE Arena v1, which Edge Delta built and ran: on 21 injected Kubernetes incidents, Edge Delta opened investigations for 18 and Grafana for 12.
6. 06Prices are as published on the retrieval dates. Datadog's per-feature credit figures are averages, PagerDuty does not publish per-tier AI Action allotments, and Edge Delta does not publish the credit cost of an investigation.