The AI agent observability area is taking off — however how can enterprises ensure what observability merchandise and options they want?
Observability startup groudcover (decrease case "g" intentional) introduced this week that it raised $100 million in a spherical led by One Peak, bringing its whole funding to $160 million.
The corporate says it has greater than 250 paying prospects, tripled annual recurring income over the previous 12 months and is more and more changing established observability platforms inside enterprise environments. These are company-reported figures, however collectively they level to rising momentum in one in all enterprise software program's best markets.
That market has lengthy been dominated by firms together with Datadog, Dynatrace, New Relic, Splunk and Grafana. Between them, they signify billions of {dollars} in annual income and years of product maturity. Breaking into that group has by no means been straightforward.
groundcover's argument is that synthetic intelligence has basically modified the assumptions these platforms have been constructed on.
Relatively than competing function for function, the four-year-old firm is attempting to persuade enterprises that the structure underpinning observability itself wants to alter as AI techniques grow to be extra autonomous, produce vastly extra telemetry and more and more take part in software program operations. Whether or not that thesis proves appropriate stays an open query, however it gives a compelling lens by way of which to look at how observability is evolving alongside enterprise AI.
AI is popping telemetry into an infrastructure downside
Observability has historically been considered as a post-production self-discipline. Engineers deploy functions, monitor logs, metrics and traces, examine incidents, and enhance reliability over time.
That workflow is altering.
AI-assisted software program growth has dramatically accelerated deployment cycles. Coding assistants generate extra code, infrastructure evolves extra quickly, and organizations are deploying more and more advanced distributed techniques that mix microservices, Kubernetes clusters, APIs and enormous language fashions. On the similar time, enterprises are starting to function AI brokers that execute multi-step workflows, name exterior instruments and work together with manufacturing techniques.
Every of these actions generates telemetry.
The result’s an explosion of operational information that organizations more and more wish to retain reasonably than discard. AI functions introduce further layers of observability past conventional infrastructure monitoring, together with immediate execution, mannequin latency, token consumption, retrieval pipelines, device invocations and agent habits. As enterprises experiment with autonomous techniques, that telemetry turns into more and more invaluable as a result of it supplies the context wanted to know what an AI system truly did and why.
For a lot of organizations, this creates pressure with pricing fashions that cost in response to the quantity of knowledge ingested.
Traditionally, engineers have typically responded by sampling traces, shortening retention intervals or limiting which information is collected. These approaches scale back prices, however additionally they scale back visibility exactly when AI-driven techniques demand extra full operational context.
"We've seen telemetry exploding," groundcover co-founder and CEO Shahar Azulay stated throughout a latest media briefing. "Customers are annoyed by not getting all the worth from Datadog and related platforms. They're limiting the information, siloing it, sampling it."
Whether or not that frustration is widespread sufficient to reshape the market stays to be seen, however the underlying development is tough to disregard. AI is making observability much less about amassing sufficient information and extra about amassing the whole lot organizations might finally want.
Relatively than including AI, groundcover argues the structure itself has to alter
Many observability distributors have launched AI assistants, AI-powered root trigger evaluation and AI observability options over the previous two years. Datadog, Dynatrace, New Relic and Grafana have all introduced merchandise aimed toward serving to enterprises monitor AI functions or automate operational duties.
groundcover acknowledges these developments however argues they don’t handle what it sees because the extra elementary difficulty: the place telemetry lives and the way prospects pay for it.
As an alternative of working a traditional SaaS platform that shops buyer telemetry in vendor-managed infrastructure, groundcover makes use of what it calls a bring-your-own-cloud (BYOC) structure.
Prospects hold the information airplane—together with telemetry storage and processing—inside their very own AWS, Microsoft Azure or Google Cloud environments, whereas groundcover supplies a managed management airplane and person expertise. A completely self-hosted deployment choice can also be obtainable.
Whereas some opponents, together with Datadog and some different observability distributors, do provide restricted hybrid or customer-controlled information residency choices, these are typically not equal to a full BYOC mannequin. Usually, telemetry continues to be processed and saved inside the vendor’s managed infrastructure, with solely partial controls (equivalent to regional information residency, personal hyperlinks, or selective log forwarding) obtainable.
That architectural choice influences almost each side of the corporate's technique.
As a result of prospects already pay for their very own cloud infrastructure, groundcover argues it might probably keep away from charging primarily based on telemetry ingestion. As an alternative, pricing relies totally on monitored hosts, no matter telemetry quantity.
The corporate believes this modifications buyer habits.
Relatively than deciding which logs or traces are too costly to maintain, organizations can theoretically retain full telemetry and use it for operational evaluation, compliance and AI-assisted troubleshooting.
"We don't worth by information quantity," Azulay stated. "We worth by the dimensions of the infrastructure."
The excellence issues as a result of AI workloads have a tendency to extend telemetry far quicker than infrastructure itself.
That doesn’t essentially make host-based pricing universally cheaper. Organizations with comparatively mild workloads unfold throughout many hosts might discover completely different economics than dense Kubernetes environments producing huge quantities of telemetry. The corporate's personal briefing notes that per-host pricing is most advantageous for organizations with excessive telemetry density and could also be much less compelling for evenly utilized fleets.
Nonetheless, the broader argument is much less about price alone than predictability. Enterprise infrastructure groups typically wrestle with observability payments that fluctuate alongside utility progress. groundcover's mannequin makes an attempt to align pricing extra intently with infrastructure planning reasonably than information era.
eBPF sits on the heart of the corporate's technical differentiation
The second pillar of groundcover's technique is eBPF, a Linux kernel know-how that has quickly grow to be some of the necessary constructing blocks for contemporary cloud observability.
As an alternative of requiring builders to manually instrument functions, eBPF permits software program operating contained in the working system kernel to look at community visitors, system calls and utility habits with minimal code modifications.
That permits quicker deployment and broader visibility throughout infrastructure.
For organizations working Kubernetes clusters and cloud-native functions, decreasing instrumentation complexity can considerably shorten deployment instances whereas growing telemetry protection.
Azulay argues this turns into particularly necessary as AI techniques generate more and more advanced interactions throughout companies.
"Our sensor permits us to look at techniques very deeply from infrastructure to utility to AI workloads with out builders needing to instrument code," he stated throughout the briefing.
eBPF itself is hardly distinctive. Many observability distributors now incorporate it into their platforms.
What groundcover argues differentiates its method is combining computerized eBPF assortment with customer-controlled storage, OpenTelemetry compatibility and unified pricing inside a single platform.
The corporate's personal analysis briefing acknowledges that none of those applied sciences individually represents a aggressive moat. The claimed differentiation lies within the mixture of eBPF-first assortment, managed BYOC structure, host-based economics and full-stack observability delivered collectively.
AI brokers have gotten each prospects—and customers—of observability
Maybe essentially the most fascinating side of groundcover's technique extends past conventional monitoring.
The corporate more and more describes observability as infrastructure for autonomous software program growth.
Traditionally, observability platforms have served human operators investigating manufacturing incidents.
groundcover believes future observability platforms will more and more serve AI brokers as effectively.
Its Agent Mode product permits engineers to research incidents utilizing pure language throughout logs, metrics, traces and Kubernetes occasions. Extra importantly, Azulay envisions observability turning into the suggestions mechanism that informs coding brokers about what truly occurred in manufacturing.
Relatively than merely detecting failures after deployment, observability turns into steady operational context that autonomous techniques can use to judge modifications, determine regressions and finally suggest or implement fixes.
"We're seeing observability transferring from being a post-production device… to individuals taking context from manufacturing and feeding it again to their coding brokers to allow them to write code higher," Azulay stated.
Right now, the corporate emphasizes that people stay within the loop.
Agent Mode investigates incidents and surfaces suggestions, however manufacturing modifications nonetheless require human approval. Azulay expects autonomy to extend step by step as organizations grow to be extra snug permitting AI techniques to take part in operational workflows.
That imaginative and prescient displays a broader development rising throughout enterprise software program, the place AI brokers more and more span growth, testing, deployment and operations reasonably than functioning as remoted assistants.
Why some enterprises are contemplating alternate options
groundcover is coming into an intensely aggressive market populated by distributors with a long time of enterprise expertise.
Datadog alone generated greater than $3 billion in annual income in 2025. Dynatrace, Cisco's Splunk enterprise, Grafana Labs and New Relic all preserve in depth companion ecosystems, mature integrations and enterprise assist organizations that newer entrants can not simply replicate.
groundcover is just not trying to outscale these incumbents in a single day.
As an alternative, it argues that AI creates an architectural inflection level much like earlier transitions from on-premises infrastructure to cloud-native computing.
In keeping with Azulay, many purchasers initially undertake groundcover to cut back observability prices however more and more stay as a result of they need unrestricted entry to richer telemetry and AI-native workflows.
He says deployments sometimes substitute incumbent platforms reasonably than function alongside them, though the corporate has not publicly disclosed buyer migration information or unbiased research validating that declare.
The corporate's journalist briefing additionally urges warning round some efficiency claims.
Income progress, buyer counts and enterprise adoption figures originate from groundcover itself. Revealed buyer case research reporting important price financial savings are vendor-authored and shouldn’t be handled as unbiased validation with out further proof. The briefing additionally recommends scrutinizing precisely what metadata leaves buyer environments in commonplace BYOC deployments, reasonably than assuming that no operational information ever reaches vendor infrastructure.
These caveats are necessary as a result of the observability market has grow to be crowded. Gartner at present tracks multiple hundred observability merchandise, and almost each main vendor now markets AI-powered operational capabilities.
Success will probably rely much less on whether or not AI issues—which more and more seems inevitable—and extra on whether or not enterprises conclude that current architectures stay enough.
The bigger query buyers are betting on
Seen narrowly, groundcover's Sequence C is one other massive infrastructure funding spherical.
Seen extra broadly, it displays a rising debate about what observability turns into in an period the place software program more and more writes, exams and operates itself.
If AI continues producing exponentially bigger volumes of operational information, conventional assumptions about telemetry assortment, pricing and storage might come below growing strain. Distributors that constructed companies round charging for information ingestion might have to evolve their economics alongside buyer expectations. New entrants, in the meantime, have a chance to design round these altering assumptions from the outset.
groundcover believes that chance lies in combining customer-controlled infrastructure, computerized telemetry assortment and AI-assisted operations right into a platform designed for autonomous software program reasonably than merely including AI options to current observability merchandise.
Whether or not that architectural guess proves sturdy will rely on enterprise adoption over the subsequent a number of years.
However the firm's newest funding spherical suggests at the least some buyers consider the subsequent battle in observability won’t be fought over dashboards or alerts. It is going to be fought over who builds the operational information layer that more and more clever software program depends upon to know—and finally handle—the techniques it runs.
