Observability startup groundcover argues that the rise of AI agents necessitates a fundamental architectural shift in how enterprises monitor their systems. Instead of the traditional vendor-managed model, the company is pushing a bring-your-own-cloud (BYOC) approach where data stays in your VPC and costs align with your infrastructure size rather than the sheer volume of telemetry generated. (source)
Why AI is turning telemetry into an infrastructure problem
Traditional observability followed a post-production workflow: engineers deployed code, then monitored logs and metrics to fix issues. AI-assisted development has shattered this cycle. Coding assistants generate massive amounts of code, and organizations are now operating autonomous AI agents that execute multi-step workflows and interact with live production systems.
Each agent action creates a trail of telemetry—prompt executions, model latency, and tool invocations. For most teams, this data is gold because it provides the context needed to understand what an AI system did and why. However, this creates a massive tension with standard SaaS pricing models that charge per gigabyte of ingested data. To save money, engineers often resort to sampling traces or shortening retention periods, but this happens exactly when you need the most complete visibility.
"We've seen telemetry exploding," Shahar Azulay, co-founder and CEO of groundcover, said during a recent briefing. "Users are frustrated by not getting all the value from Datadog and similar platforms. They're limiting the data, siloing it, sampling it."
How the BYOC model changes your cost structure
Most major players like Dynatrace and New Relic manage your data in their own cloud. Groundcover's BYOC architecture flips this: you keep the data plane—storage and processing—inside your own AWS, Microsoft Azure, or Google Cloud environment. Groundcover provides the managed control plane and user experience on top of it.
Because you are already paying for the cloud infrastructure where the data lives, groundcover argues they can eliminate telemetry ingestion fees. Instead, they price primarily based on the number of monitored hosts. This shift is critical for AI teams because:
- Predictable billing aligns with your existing infrastructure planning rather than fluctuating data spikes.
- Full retention allows you to keep every log and trace for compliance or AI-assisted troubleshooting without fear of a massive bill.
- Data residency ensures your sensitive AI telemetry never leaves your controlled environment.
However, this isn't a universal win. The company notes that per-host pricing is most advantageous for high-density workloads. If you have a lightly utilized fleet of many hosts, a traditional volume-based model might still be cheaper. You should evaluate your host-to-telemetry ratio before switching.
The technical edge: eBPF and Agent Mode
To make this architecture work without manual overhead, groundcover relies on eBPF—a Linux kernel technology that allows the platform to observe network traffic and system calls with minimal code changes. This means you get deep visibility into AI workloads without requiring developers to instrument every single application.
The platform also introduces Agent Mode, which allows engineers to investigate incidents using natural language. Azulay envisions a future where observability serves as a feedback loop: instead of just detecting a failure, the platform feeds production context back to coding agents so they can learn from mistakes and write better code in the next iteration.
Verdict: When to switch to BYOC
Groundcover isn't trying to out-feature the established giants; they are trying to out-architect them. Whether this is the right move for you depends on your specific deployment profile:
- Switch to BYOC if your AI agents are generating massive telemetry volumes that make your current SaaS bills unpredictable or impossible to manage.
- Stick with SaaS if you have a lean infrastructure with low telemetry density, where the simplicity of vendor-managed ingestion outweighs the need for host-based pricing.
- Wait and watch if you are currently satisfied with your sampling strategy and don't yet have a requirement for complete operational context for autonomous agents.
