Sherlocks AI vs Datadog Bits Investigation

If you are evaluating AI SRE tools and you already use Datadog, Bits Investigation is one of the first options you will hit. This is a straight comparison of the two, including where Datadog is genuinely the better choice. A comparison page that pretends the competitor has no strengths is not worth reading, so this one does not do that.

Bits Investigation, formerly announced as Bits AI SRE, is Datadog's AI investigation agent. It went generally available on December 2, 2025, as Datadog's first generally available AI agent, after testing across more than 2,000 customer environments, per Datadog's GA announcement.

The way it works is clean. When a Datadog monitor fires, Bits automatically launches an investigation without waiting for an engineer to ask. It builds multiple hypotheses about the cause, checks dependencies, and exposes its reasoning in an Agent Trace view that shows each tool it called and the intermediate analysis behind every step. It reasons over the full breadth of Datadog platform data: metrics, logs, traces, dashboards, deployment changes, source code, RUM, Database Monitoring, Network Path, and Continuous Profiler.

Datadog has kept shipping since GA. A March 2026 update made investigations roughly twice as fast, with most now finishing in three to four minutes depending on complexity, and widened the data Bits reads to the set above.

Give Datadog real credit here. If your telemetry already lives in Datadog, Bits is one of the easiest ways to add autonomous investigation to your stack. There is no new tool to learn, no context switching, and it sits on top of best-in-class high-cardinality data handling. For a team that is genuinely all-in on Datadog, that native access is a real advantage.

Sherlocks AI is a purpose-built AI SRE that investigates production incidents autonomously and tells your team what broke, why, and what to do next. It is Slack-native, so the investigation happens where your team already works, and it runs the moment an incident fires rather than waiting for an engineer to start digging.

Under the hood, it runs 16+ domain-specialized agents that work in parallel, a Database agent, a Kubernetes agent, and more, each an expert in one slice of the stack rather than one general model trying to cover everything. During an incident, those specialists investigate at once and their findings are correlated into a single timeline: logs, metrics, traces, deploys, and Kubernetes events pulled together into one clear picture of the failure.

What it offers comes down to three things. First, autonomous investigation to a confirmed root cause, not just an alert or a symptom. Second, deployment inside your own VPC, so raw telemetry never leaves your network. Third, a persistent memory of your environment, so recurring incidents get faster to resolve as it learns how your systems fail and how your team fixes them. It is built investigation-first, to reach the cause the way a senior SRE would, without requiring you to move your data anywhere.

This is the core difference, and it decides most of the rest.

Bits runs in Datadog's cloud. For it to investigate, your telemetry has to be in Datadog. That is fine for many teams and a real problem for others, specifically anyone with data residency requirements, strict security review, or a policy that raw production telemetry should not leave their environment.

Sherlocks AI runs the opposite way. It sits inside your own VPC with read-only access, so raw telemetry never leaves your controlled network. The investigation comes to your data instead of your data going to the investigator. For regulated industries and security-conscious teams, that single architectural choice is often the deciding factor, and it is one Bits cannot match by design, because Bits is a SaaS product living in Datadog's cloud.

Bits is deep on Datadog data and still narrow outside it. It reasons powerfully over the Datadog platform, but pulling context from outside tools such as GitHub, ServiceNow, Grafana, Splunk, Dynatrace, and Sentry remained a Preview capability rather than generally available as of Datadog's March 2026 update. So the honest summary is that Bits is excellent within Datadog and still maturing across a mixed toolchain.

Sherlocks AI is tool-agnostic by design. It investigates across your existing stack rather than requiring everything to funnel through one platform first. If you run Datadog for some things, Grafana for others, and cloud-native tooling elsewhere, that neutrality matters, because incidents rarely respect tool boundaries. Open standards like OpenTelemetry make that cross-tool reach practical.

Datadog moved Bits onto a consumption-based AI Credits model, drawing from a pool shared across the Bits agents. Datadog puts average consumption at roughly 6.5 credits per investigation and bills only investigations that reach a conclusion, though actual consumption varies with complexity. Credit Datadog here too: this replaced a much more expensive per-investigation model and cut the effective cost per investigation substantially. The structural point still stands, though. Whatever the rate, the charge stacks on top of your existing Datadog bill for hosts, APM, logs, and RUM, and it scales with incident volume, so spend climbs during exactly the bad month when you are running the most investigations.

Sherlocks AI uses flat, predictable plans rather than per-investigation billing, so a bad week does not produce a surprise invoice. We are not going to pretend Sherlocks AI is always cheaper in absolute terms, that depends on your volume and your existing Datadog spend, but the shape of the pricing is easier to forecast because it does not move with your incident count.

Bits deepens your commitment to Datadog. It is most valuable when you are fully invested, which is also what makes it a lock-in decision: the more you rely on Bits, the more the whole observability stack has to stay in Datadog.

Sherlocks AI does not require you to adopt or stay on any observability platform. It works with what you have. If you switch telemetry tools later, the investigation layer does not have to move with them.

To be fair and direct: choose Bits if your observability stack is fully consolidated in Datadog, you want zero context switching, and you are comfortable with consumption-based pricing that scales with incidents. For a Datadog-native shop that values one vendor and one pane of glass, Bits is a strong, well-executed option with the reliability and support of a large platform behind it.

Choose Sherlocks AI if any of these are true: you run a mixed toolchain and do not want investigation gated on one platform, you need the AI to run inside your own VPC so raw telemetry stays in your network, you want predictable pricing that does not spike with incident volume, or you want an investigation-first tool with real, auditable investigations rather than an AI layer added onto an observability product. Faster, more reliable diagnosis is the goal, and where the AI runs and what it can see are what get you there. For the metrics behind that goal, see our guide on MTTR, MTTD, MTTA and MTTF, and for the deeper investigation method, our writeup on root cause analysis. Reliability benchmarks like the DORA and Google SRE references are a useful frame for both tools.