Sawmills is a prime example of an agentic application in the data infrastructure layer. Unlike AI copilots that provide recommendations for a human to execute, the Sawmills platform, 'Mills,' operates as an autonomous agent with specific domain expertise in telemetry and OpenTelemetry standards. It monitors live data streams and makes real-time decisions—such as filtering, redacting, or deduplicating—to optimize system health and costs.
For the broader agent ecosystem, Sawmills represents the move toward 'Agentic Engineering' where infrastructure management is delegated to specialized AI agents. Their focus on natural language interaction via Slack and Teams demonstrates how agents can integrate into existing developer workflows to resolve complex infrastructure issues without requiring the user to leave their primary workspace.
Most modern engineering teams suffer from a phenomenon often called the observability tax. As systems scale, the volume of logs and metrics grows exponentially, often leading to a situation where the cost of monitoring a service rivals the cost of running it. Sawmills, founded in 2024 and based in Los Altos, CA, entered the market with the specific goal of breaking this cost curve. The company is led by Ronit Belson, Amir Jakoby, and Erez Rusovsky, and recently emerged from stealth with $10 million in seed funding from Team8 and Mayfield.
At the core of their offering is Mills, a platform they describe as the first agentic telemetry management system. While the tech industry is saturated with "AI copilots" that wait for human prompts, Mills is designed to act autonomously as a domain expert. It sits between telemetry sources—such as Kubernetes clusters or serverless functions—and observability backends like Datadog, New Relic, or Splunk.
Sawmills operates across the entire development lifecycle, rather than just acting as a filter at the end of the pipe. In the IDE and CI phase, Mills analyzes telemetry as it is written. It flags high-cardinality data, redundant metrics, or PII before the code ever reaches production. This shift-left approach aims to prevent bad telemetry from shipping in the first place, rather than cleaning it up after the ingestion bill arrives.
In the production environment, the platform uses AI to analyze live data streams in real-time. It identifies spikes, runaway logging, and duplicate events. Instead of just alerting a DevOps engineer to a problem, Mills can automatically apply remediations within pre-defined guardrails. This might involve sampling low-value logs, aggregating high-frequency metrics, or redacting sensitive information. For issues that require a human touch, Mills integrates with Slack and Microsoft Teams, allowing developers to approve optimizations or adjust filters directly from their chat interface using natural language.
Sawmills is built entirely on the OpenTelemetry (OTel) standard. This is a critical architectural choice that allows it to function within existing enterprise stacks without introducing a new form of vendor lock-in. Because it uses the OTel Collector, it can ingest from and export to virtually any major observability or security tool.
Competitively, Sawmills sits in a unique space. It doesn't compete with Datadog or Splunk; it makes them more affordable and useful by ensuring only high-signal data is stored. By claiming cost reductions of 60% to 80%, Sawmills is making a purely economic argument to CTOs while solving a technical headache for SREs. The challenge for the company will be proving that their AI can consistently distinguish between "noise" and the "rare signal" required for debugging complex, novel outages.
An agentic telemetry management platform that optimizes observability costs and data quality.
Sawmills is hiring.