Datadog AI agent observability, security seek to boost trust
As AI agents mature, new tools aim to bolster their reliability and security with fresh visibility into automation workflows and more detailed troubleshooting.
NEW YORK -- As the breakneck pace of enterprise generative AI adoption continues, advancing agentic workflows comes with a host of new requirements for reliability and security.
Like its IT automation and observability competitors, Datadog has teed up a broad set of fresh AI agents as well as AI agent observability and security tools to address these challenges. Attendees at its DASH 2025 conference were intrigued by new releases such as site reliability engineer (SRE), developer and security analyst agents, and an AI experimentation tool. But some trepidation lingers about AI trust and how Datadog will price the new features, many of which remain in preview.
MyFitnessPal, makers of a health and fitness tracking app, switched from Cisco's Splunk to Datadog's security information and event management (SIEM) tool over the last year because Datadog was easier for a small security operations team of three to manage, according to a breakout session presentation by Allen Cox, senior director of security and IT at the company, based in Austin, Texas.
Datadog's new Bits AI security analyst agent, released in preview this week, could save that team even more time on SecOps tasks, but Cox said during a Q&A at the end of the session that AI agents still generally present trust and organizational issues.
"If you had asked us a year ago [if we would use an AI agent], we would have said absolutely not," he said. "The way things are advancing, we're building some confidence in a [tool] like that, [but] there's some fear and doubt around whether this could create a forensics tool situation where you're biased toward, 'Oh, my forensics tool didn't tell me this, therefore it doesn't exist,' and that may not necessarily be true."
It will also take time for organizations to adjust to new ways of working with AI agents, Cox said.
"AI may give us an opportunity to expand the total range of things that we're choosing to alert on, but we have to rethink detection and response in that context as well," he said.
Yanbing Li, chief product officer at Datadog, presents an array of AI and observability announcements during the Datadog DASH 2025 keynote session.
AI agent updates prompt pricing questions
The Bits AI SRE agent Datadog shipped this week had been promising in preview for one customer, Bert Stewart, head of the global command center at Thomson Reuters, a financial and legal software maker based in Toronto.
Bits AI emerged a year ago as a copilot that could assist with investigations into IT incidents. In this version, it has evolved to analyze more types of data, including dashboards and deployment changes; perform multistep reasoning and complex tasks that span multiple services as it conducts investigations; and learn from past investigations which steps were useful and which weren't to improve results.
It's very interesting stuff, but when things roll out from Datadog, you don't know how much it's going to cost.
Bert StewartHead of the global command center, Thomson Reuters
"We've been playing around with it in a couple of our accounts, and it's been really fast -- it's cutting the time to find an issue to [a fraction of what it was]," Stewart said during an interview with Informa TechTarget following a breakout session he presented this week.
However, pricing for the agent has yet to be finalized. It's initially available as a free trial, and Stewart said how it's eventually priced will remain a crucial factor in whether this and other Datadog AI observability tools make it into production.
"It's very interesting stuff, but when things roll out from Datadog, you don't know how much it's going to cost," he said.
Similarly, Stewart said he's excited to try a new feature launched in a limited preview for Datadog's LLM Observability product this week, Experiments, which performs comparative tests of various AI models to determine their suitability for a given task.
"I think that's a really cool feature, because doing it yourself, you have to build a lot of infrastructure," he said during his session presentation.
But generally, the decision to use a vendor-provided feature comes down to an "age-old dilemma in IT: Do we build it or do we buy it?" Stewart said during a Q&A at the end of his session. "Can we build it better, cheaper, and are we willing to support it, or do we want to buy something that works really well and deal with the tradeoffs?"
Datadog AI agent observability draws on runtime data
AI trust features are built in with Datadog's own agents. For example, the SRE Agent shows the user each of its hypotheses about the root cause of an incident and the chains of reasoning it uses to arrive at a conclusion.
More new features also shore up observability for agents that customers develop themselves or use from other vendors. Another new LLM Observability feature, AI agent monitoring, includes an agent execution flow graph that tracks how automation workloads move through an agentic system, as well as each agent's associated instructions, tools, guardrails, agent framework and model information. An AI agent troubleshooting feature automatically detects and flags errors within the execution flow graph. A separate new product, AI Agents Console, offers an overarching view of internally and externally developed agents within an organization, including AI governance information about their associated costs and security posture.
On the security front, Datadog added a Sensitive Data Scanner that can identify and prevent data leakage and data poisoning attacks in AI systems through its log management and Cloud SIEM tools. Other newly introduced features can detect malicious models and prompts, model drift, and prompt injection attacks. More AI agent security detections for agentic tool misuse and identity-based and denial-of-service attacks are coming later in 2025, according to Datadog officials during a keynote presentation.
None of these features is totally unique to Datadog, IDC analyst Katie Norton said, but as with its internal developer portal, Datadog could set itself apart from competitors with its focus on observed runtime data, rather than code or preproduction test results.
"Datadog does have code security, but I think some of their competitors are not as strong at bringing together those two areas of shift left and shift right," Norton said. "There's a pendulum that swings between shift left and shift right in the market, and right now, I would say the pendulum is a little more on the right side -- AI apps aren't deterministic, so you need to have visibility into the runtime to secure them."
Beth Pariseau, a senior news writer for Informa TechTarget, is an award-winning veteran of IT journalism covering DevOps. Have a tip? Email her or reach out @PariseauTT.
