What Is It Like to Work in Datadog’s AI Research Lab?

What makes working at Datadog’s AI Research Lab unique? According to Chief Scientist Ameet Talwalkar, who leads the AI Research Lab, it’s the fact that his team at the cloud-based observability and security platform is building real things to solve real problems. 

 “We’re doing cutting-edge AI, but focused on real translational impact and working with engineering and product teams to actually get our research over the wall and into production,” Talwalkar said in a Datadog video.

Take ARFBench and Toto 2.0, for example. 

What Does Datadog’s AI Research Lab Work On?

ARFBench, which stands for “Anomaly Reasoning Framework Benchmark,” is a time series question-answering benchmark released by Datadog in April 2026. Built from Datadog’s own production telemetry and internal incidents, the open benchmark gives researchers a real-world way to evaluate AI systems designed for observability and incident response.

Engineers responding to outages need to interpret complex time-series data to answer questions like: When did this anomaly start? Which metrics changed? What could be causing this problem? With ARFBench, incident-response agents with these reasoning capabilities can be tested to see if they’re good at these tasks. 

“These time series question-answering tasks are essential for engineers, and present challenging and necessary tasks for SRE models and agents to perform,” Talwalkar and his co-authors wrote in a blog post. 

Ultimately, ARFBench will help engineers build and evaluate AI that helps engineers diagnose incidents faster, reduce the time to resolution and make AI-assisted observability more reliable. 

But for Datadog’s AI Research Lab, the release of ARFBench was just the beginning.

In May, Talwalkar and his peers announced Toto 2.0, a family of open-weights time-series foundation models ranging from 4 million to 2.5 billion parameters. Toto 2.0, whose models improve predictability as they scale, brings Datadog closer to AI-powered observability that can predict system behavior — not just detect problems after they occur. And according to Datadog’s blog, Toto 2.0’s largest versions now lead the benchmarks Datadog tested for observability and general-purpose forecasting.

“Toto 2.0 is the first model family for which simply making the model bigger reliably makes it better,” Talwalkar said in a LinkedIn post.

With Toto 2.0’s release, Datadog’s AI Research Lab showed that it can build AI models for more than just observability — it can also compete with broad, state-of-the-art forecasting models. 

“Our longer-term goal is to develop a full-fledged world model for observability, extending to all telemetry types, unlocking capabilities such as proactive incident detection, root cause analysis, counterfactual analysis, simulation and agent training,” Talwalkar and his co-authors said in a Datadog blog post about Toto 2.0’s release.

What Is It Like to Work in Datadog’s AI Research Lab?

ARFBench and Toto 2.0 show that Datadog’s AI Research Lab isn’t just adding AI features to observability; the team is building AI that could understand the state and behavior of increasingly complex software systems. For AI research scientists and engineers, working at Datadog means contributing to foundational AI research and working on direct applications to real production systems. 

“The things that you’re building are the direct models or the direct agents that are going into production,” Talwalkar said in a video about building Datadog’s AI Research Lab. “What sets Datadog apart and what gets me really excited is that we’re building real things to solve real problems. I’m very excited to see where we go.”

If helping invent AI that determines what next-generation observability looks like sounds like your kind of challenge, check out Datadog’s open roles. 

Frequently Asked Questions