Datadog Builds an Observability Tollbooth for Autonomous AI Agents

When your food delivery app crashes, an engineer gets a ping, checks a stack trace, and fixes the line of broken Python. But when an autonomous AI agent books the wrong flights across three corporate cards because an external booking API returned an unexpected currency format, nobody wrote that mistake into the codebase. The model reasoned its own way into failure.

That is the mess Datadog sells software to untangle. If Nvidia sells the engines for the AI factory floor, Datadog builds the instrument cluster that tells you when the machinery is overheating. The New York company started sixteen years ago as cloud dashboard software for sysadmins. Today, more than 33,000 businesses pipe their server logs, network traces, and application errors into Datadog to keep their digital storefronts alive.

Now artificial intelligence is breaking the basic rules of enterprise software engineering. In classic cloud architecture, software is deterministic: if code receives input A, it returns output B every single time. Modern AI agents do the exact opposite. They plan multi-step workflows, query vector databases, call third-party APIs via protocols like Anthropic's Model Context Protocol, and invoke secondary sub-agents. A single hallucinated parameter in step three can silently corrupt financial ledgers or trigger runaway loops that burn thousands of dollars in cloud tokens before anyone notices.

Datadog is turning that chaotic non-deterministic behavior into high-margin telemetry billings. Through its Agent Monitoring and LLM Observability tools, the platform records every decision branch an autonomous bot takes. Engineers can inspect the prompt, verify tool parameters, measure latency at each reasoning node, and track token spend across model providers. On Datadog's latest earnings call, management disclosed that tool calls across its Model Context Protocol servers quadrupled quarter-over-quarter, expanding more than twenty-fold from late last year.

The company's core financial engine reflects that expanding footprint. Second-quarter revenue surged 36% year-over-year to $1.12 billion, generating $279 million in free cash flow on a 25% margin. Remaining performance obligations jumped 43% to $3.47 billion, showing enterprise customers are locking in multi-year commitments rather than testing toy pilots. Meanwhile, platform stickiness continues to climb: 58% of customers now deploy four or more Datadog products, and large accounts generating over $100,000 in annualized revenue expanded 23% to 4,720 customers.

For the stock to sustain its current valuation, however, the market must re-rate Datadog from a standard software utility into a compound tollbooth on enterprise AI compute. The equity trades near an eye-watering 93 times forward earnings. For that multiple to hold, Datadog must prove that agentic workflows generate non-linear telemetry volume that far outpaces traditional human web traffic. When an autonomous shopping agent queries fifty travel endpoints in seconds instead of a human clicking three links, log volume explodes. If Datadog can capture that surge without pushing customers into manual data throttling, its cash generation can compound into its multiple.

The short sellers, however, have a clean, mathematical rebuttal. Datadog's pricing model bills directly on ingested data volume, host counts, and active spans. When enterprise software leaders look at their cloud invoices, Datadog is frequently one of their three largest external line items alongside AWS and Snowflake. During the second quarter, Datadog experienced a sharp usage optimization from its single largest AI customer, prompting management to de-risk full-year guidance. If competing open-source standards like OpenTelemetry make switching easier, or if foundation model providers package native telemetry for free, Datadog will face severe downward pricing pressure that crushes a triple-digit forward multiple.

What we're watching into the next earnings report is sequential usage expansion across large enterprise accounts, specifically whether non-AI customer net expansion continues to re-accelerate toward 30% alongside agentic adoption. If autonomous bots are genuinely running the next generation of business software, someone has to meter the road. Datadog has already built the tollbooth.

Disclosure: The Signal holds no position in DDOG. Positions may change. This is not financial advice.