AI SRE Agents 2026: Five Products, Five Different Meters
One bills agent-seconds, one bills AI credits at roughly 6.5 per investigation, two bill seats, and one will not tell you. The word agent got stretched across five product shapes, and the meter is the fastest way to work out which one you are looking at.
Five vendors sell something called an AI SRE. They do not sell the same thing, and the feature grids are built to obscure that. The quickest way through the fog is not the feature list. It is the meter, because what a vendor decided to count tells you what they think they are selling.
AWS DevOps Agent counts agent-seconds. Datadog counts AI credits, about 6.5 per autonomous investigation. PagerDuty counts seats and hands you a bundle of AI Actions with them. incident.io counts seats and does not meter the AI at all. Resolve AI counts something, but you will need a sales call to find out what.
Two of those bars rest on assumptions and two do not. The seat-based numbers are what you will actually pay: PagerDuty Business at $41 per user per month on annual billing is $820 for twenty, and incident.io Pro at $25 plus its $20 on-call add-on is $45 a head, or $900. The usage-based ones move with how you use them, which is either the honest model or the unpredictable one depending on your mood.
The five archetypes under one label
Sort each product into one of these before reading a single feature page.
The agentic investigator takes an alert, forms hypotheses, pulls logs and metrics and traces and deploys, and tells you what probably broke. AWS DevOps Agent and Resolve AI live here. Most useful category, most oversold.
The telemetry-bound RCA engine does the same job but can only see inside one observability platform. Datadog’s Bits Investigation is the clearest example. Excellent if everything you care about is already in Datadog, blind to whatever is not.
The on-call orchestrator runs the human side: who to page, what to tell stakeholders, keeping the timeline straight. incident.io and PagerDuty sit here, though PagerDuty has been pushing into the first category.
The postmortem generator writes the summary and the stakeholder update. PagerDuty’s Scribe Agent and incident.io’s postmortem editor are this. Useful, low-risk, and not worth enterprise money on its own.
The remediation agent takes action. Almost nobody ships this fully autonomous, and the ones that gesture at it hedge with a human in the loop. That, more than any roadmap slide, tells you where the market actually is.
Comparing a postmortem generator against a multi-agent investigator on one feature grid is a buyer trap. They are not competing for the same job.
What each one can reach
The archetype matters less than the blast radius of what the agent can see.
AWS DevOps Agent is wired into CloudWatch, X-Ray, and your AWS topology, and it does reach outside AWS through observability integrations and MCP servers. It is worth reading its billing carefully before you model anything, because the agent-second rate is not the whole bill and the support credits that offset it expire monthly. I went through that arithmetic in more detail in the piece on its pricing mechanics.
Datadog Bits Investigation is as good as your Datadog coverage and no better. It is now one of a family alongside Bits Chat, Bits Agent Builder, Bits Code, and Bits Security Analyst, all drawing on the same AI Credits meter at different rates: about 0.5 credits for a chat, 3 for an agent-builder run, 5 for a code fix, 6.5 for an autonomous investigation. Credits are $500 per 500 per month on annual billing, or $1.30 each on demand. So an investigation is roughly $6.50, and sixty of them fit inside one bundle.
That correction matters, because the number that circulates for this product is $500 per twenty investigations. It is not. The agent license was never going to be the expensive line anyway. The Datadog ingest it reasons over is, and no comparison that leaves the ingest out is telling you anything.
PagerDuty sells seats at $25 a month on Professional, $49 on Business, or $21 and $41 on annual billing, with Enterprise quoted. Every paid plan includes PagerDuty Advance as an allowance of one-time AI Actions: 1,000 on Professional, 5,000 on Business, 20,000 on Enterprise, with each action consuming tokens according to its type. The SRE Agent is generally available, in Slack through Advance and in the Operations Console through AIOps and Advance.
The interesting part of PagerDuty’s position is structural rather than technical. It knows who is on call and what the escalation policy says, which no observability vendor knows, and its Spring 2026 direction was to put the SRE Agent onto the schedule itself as a responder that triages before a human is paged. Whether that is a good idea depends entirely on how much you trust it, but it is a genuinely different asset from knowing why the database is slow.
incident.io is per seat, not the flat platform fee it sometimes gets described as. Free on Basic, $15 a user annually on Team, $25 on Pro, with on-call as a separate per-user add-on at $10 or $20. The AI features, Scribe and the AI agent and the postmortem editor, are gated to Pro and Enterprise and bundled into the seat with no separate meter. That makes it the easiest of the five to forecast: seats times a number, and nothing moves when you have a bad month.
Resolve AI is the tool-agnostic one, reading across observability, infrastructure, and source control rather than requiring one pipeline. It has kept raising: $125M Series A at a $1B valuation in February 2026, then a $40M extension in April at $1.5B led by DST Global and Salesforce Ventures. It publishes no pricing, which in this market means sales-led and priced accordingly.
What the MTTR percentages are measuring
Every page in this category waves a percentage. Before any of them move you, check which of two numbers it measures.
Time-to-root-cause and time-to-resolution are not the same thing. Cutting investigation time by 80% is real value. If a human still writes the fix, gets the change approved, and deploys it, end-to-end resolution drops by considerably less. Vendors quote whichever is larger.
And every reference number comes from a team with clean telemetry and well-instrumented services, because those are the teams that agree to be case studies. An agent inherits the patchiness of what it reads. “Handles the first 80%” assumes the first 80% is the easy, documented part, which on your worst incidents is exactly what it is not.
The only useful evaluation is a shadow-mode pilot. Run it against real incidents for two weeks, log every hypothesis, and grade it against what your engineers concluded. Count how often it was right, how often it was confidently wrong, and what it saved on the incidents that actually hurt rather than the ones it will obviously nail.
Pick by where you stall
If you are AWS-heavy, start with DevOps Agent, and check what support credits your plan already generates before you treat it as a new line item.
If your telemetry is all in Datadog, Bits Investigation is the path of least resistance because it already sits on your data. Price the ingest, not the credits.
If you are multi-tool or multi-cloud, that is the case Resolve AI is built for, and you will be having an enterprise conversation about it.
If your bottleneck at 3am is human orientation, working out who owns what and who to page and what to tell leadership, then incident.io or PagerDuty solve your actual problem, and a deep investigator would be solving one you do not have.
Pull last quarter’s incidents and ask which archetype would have helped on the three worst ones. The answer is usually not the tool with the best demo.
Prices and product statuses came from the AWS, Datadog, PagerDuty, and incident.io pricing pages and support documentation on 23 August 2026. Resolve AI’s figures are funding announcements, not prices, because there are no prices to read.