Safety and Observability Move to the Center of AI Agent Funding
AI observability platforms and voice agent safety tools are emerging as critical infrastructure that lets enterprises deploy autonomous agents at scale with confidence, by making their behavior traceable, testable, and accountable in production environments where every interaction can carry compliance, cost, and brand risk.
Two fresh Series A rounds show where serious AI agent funding is heading: into the plumbing that keeps agents safe and observable, not only into bigger models. Tsuga has grown to several millions in revenue, with average contract values in six figures, only six months after leaving stealth, serving frontier model labs and enterprises alike. Coval has raised USD 28 million (approx. RM129 million) in new capital, bringing its total to USD 31 million (approx. RM143 million) since its 2024 launch, to expand evaluation for voice and chat agents. Together, they point to a simple truth: if agents are going to run businesses, the real bottleneck is not creativity—it is control.
Tsuga Bets on AI-Native Observability Over Legacy Telemetry
Tsuga is making an explicit bet that the old observability model—ingest everything into a third-party cloud and charge more as volumes grow—cannot survive the agent era. Every agent loop, autonomous deployment, and token interaction now emits telemetry at volumes legacy tools were never built to handle. As costs spike, teams start sampling, and the result is the worst of both worlds: high bills and blind spots.
Tsuga’s answer is an AI observability platform that deploys entirely inside the customer’s own cloud accounts, across major hyperscalers and sovereign clouds, so telemetry never leaves their environment. By design, there is no infrastructure tax, no duplication cost, and no sampling, which means AI agents can run on complete, unsampled data. Pricing is a single rate per GB of consumption, with costs expected to fall as Tsuga’s engineers tune each environment over time. This is not a cosmetic tweak; it is a new architecture for autonomous agent infrastructure, built around data gravity, governance, and predictable cost rather than vendor lock-in and ingestion fees.
Coval Targets Voice Agent Safety as a First-Class Problem
If Tsuga owns the observability data plane, Coval is going after the safety and reliability layer for voice agents. Its full-stack evaluation platform is designed for simulation, observability, and labeling across voice and chat agents, allowing enterprises to move from experimentation to reliable production at scale. The new funding is explicitly aimed at the growing reliability and compliance challenges that come with autonomous voice agents.
Coval already runs tens of millions of evaluations and serves more than 60 enterprises; companies like Zoom rely on it to cut manual QA by up to 30x and speed voice agent deployment by up to 10x. In the words of Zoom’s CX AI product lead, “Reliability and observability are a top priority for us as voice AI moves into customer-facing production environments.” Coval’s architecture is purpose-built for voice: audio processing quality, telephony latency, transcription error analysis, and workflow evaluation all sit in one autonomous agent infrastructure stack. This is what voice agent safety looks like when treated as engineering discipline, not a checklist.
From Pilots to Production: Why Safety and Observability Are Now Non-Negotiable
Enterprises are rapidly adopting voice AI for customer service, sales, financial services, and healthcare, but many still rely on manual QA that breaks under real-world complexity and scale. At the same time, AI-native applications throw off staggering amounts of telemetry, making legacy observability architectures financially and operationally untenable. This is the gap Tsuga and Coval are rushing to fill: reliable, explainable, observable AI agents as a default, not an upgrade.
Coval plans to use its new capital to grow sales and solutions engineering, deepen simulation, expand integrations, and enhance human review and monitoring features—turning evaluation into a continuous feedback loop across the whole voice agent lifecycle. Tsuga, meanwhile, leans on an architecture that scales with consumption and gives customers direct control over their telemetry, with prices that can decrease as environments are tuned. Together, they make an implicit argument: in the agent era, safety and observability are table-stakes. Any enterprise rolling agents into production without an AI observability platform and voice agent safety stack is not moving faster; it is flying blind.
