Why Is Full-Stack Observability Suddenly Critical for AI Agents?

CEO Helen Gu put it bluntly in the TechCrunch interview: "The biggest problem facing the industry today is not just monitoring and diagnosing where AI models go wrong—it's also diagnosing how the entire tech stack operates now that AI is part of it." This is a fundamental shift in thinking. For the past two years, the AI industry has been obsessed with model accuracy, RAG pipelines, and prompt engineering. But Gu is pointing out that an AI agent can fail because a database query timed out, because a GPU cluster had a memory leak, or because a downstream API changed its schema—and none of those are model problems.

According to a 2025 Gartner report cited in the article, 40% of AI agent failures in production are actually infrastructure-related, not model-related. That number is likely higher in practice because most teams don't have the tooling to distinguish between the two. InsightFinder's pitch is that they can correlate model-level metrics (token usage, response latency, hallucination scores) with infrastructure-level metrics (CPU, memory, network, database query performance) in a single pane of glass.

What Does This Mean for Incumbents Like Datadog and New Relic?

This is where it gets interesting. Datadog and New Relic dominate the application performance monitoring (APM) market, but their tools were designed for a world where software behavior is deterministic. You write code, it runs, you monitor it. AI agents are non-deterministic—they can produce different outputs from the same inputs, and those outputs can trigger cascading effects across the stack. Traditional APM tools can tell you a database query was slow, but they can't tell you that the query was slow because the AI agent generated a malformed SQL statement.

InsightFinder is specifically targeting this gap. Their platform ingests logs, traces, and metrics from both the AI layer and the infrastructure layer, then uses machine learning to correlate anomalies across both. The $15M raise suggests investors believe this is a $1B+ market opportunity that the incumbents are currently ignoring. I expect Datadog to acquire a startup in this space within 12 months—they cannot afford to cede this category entirely.

Who Actually Benefits From This Deal Beyond InsightFinder?

The immediate winners are enterprises deploying AI agents in production—think financial services, healthcare, and e-commerce companies that cannot afford unexplained failures. If you're a bank running an AI-powered trading agent, you need to know whether a bad trade was caused by a model hallucination or a network latency issue. InsightFinder's tooling provides that root-cause analysis.

The losers are the legacy APM vendors who will have to scramble to add AI-specific features. New Relic has already started adding LLM monitoring, but it's a bolt-on feature, not a rearchitecture. The deeper loser might be the open-source monitoring ecosystem—tools like Prometheus and Grafana are powerful but require significant customization to handle AI workloads. InsightFinder's value proposition is that it works out of the box for AI agents.

There's also a subtle winner here: cloud providers like AWS and Azure. If enterprises can reliably monitor AI agents, they'll deploy more of them on cloud infrastructure. Better observability removes a key adoption barrier.

My Predictions

1. Datadog will acquire InsightFinder by Q3 2027 for $500M-$1B because they cannot build AI-native observability fast enough internally and will pay a premium for market leadership.
2. New Relic will lose 15-20% of its enterprise AI customer base within 18 months as companies realize its bolt-on AI features cannot match InsightFinder's integrated approach.
3. AWS will launch a competing AI observability service by Q2 2027 to capture the cloud-native monitoring market, potentially partnering with or acquiring a startup in the space.