Observability is being rewritten around AI operations

AI became the category's organising principle

The fortnight was not quiet. Datadog, New Relic, Sentry, Grafana Labs and PagerDuty all showed some form of AI-facing motion, but the more important point is that it appeared across product pages, hiring, documentation and investor messaging rather than in one launch cycle.

Datadog added a product page for “LLM as a Judge” on 5 Sep, published a blog on building workflows with AI agents on 4 Sep, and posted AI Platform hiring on 3 Sep, including a Senior Applied Scientist role. On 3 Sep it also published content on monitoring prompt caching to optimise token usage. This is a clean signal: Datadog is treating AI workloads as both a monitored surface and a workflow layer inside the platform.

New Relic moved the front door. On 4 Sep its homepage hero changed to “End visibility silos,” with new navigation for “New Relic Autopilot” and “AI Observability,” plus sections including “Observability built to understand AI.” The day before, on 3 Sep, the same tracked homepage had an empty hero and removed several of those navigation items. That one-day gap matters. The repositioning was not just a campaign page appearing in isolation. It was a live-site reset around AI and platform consolidation.

Sentry’s hiring language on 4 Sep pointed to an “AI-native future” for application monitoring, while new content on 2 Sep covered AI size estimates. PagerDuty’s Q2 coverage on 2 Sep and 4 Sep tied its AI vision to improving profitability, and Grafana Labs’ 10,000-customer milestone, announced on 26 Aug and observed in the period on 5 Sep, was reported alongside increased AI adoption across the platform.

Open source velocity is now commercial signalling

The week’s most visible product work happened in public repositories. Grafana Labs released Grafana 13.2.1 on 2 Sep with security fixes for CVE-2026-12704 and CVE-2026-14199, shipped mcp-grafana 1.3.0 on 3 Sep with an optional per-call orgId argument, and released datasource updates across Azure Prometheus, Pyroscope and other plugins between 4 Sep and 6 Sep. It also added repositories tied to faro, mimir, beyla, opentelemetry and data sources across the period.

New Relic’s engineering rhythm was similarly broad. It released nri-jmx 3.16.0 on 31 Aug with security upgrades, newrelic-bitmovin-analytics 2.3.0 on 2 Sep with API, timing, documentation and CVE work, infrastructure-bundle 3.3.42 on 3 Sep, terraform-provider-newrelic 3.97.2 on 4 Sep with dependency updates to resolve Dependabot alerts, csec-java-agent 1.8.0 on 5 Sep, newrelic-opamp-rs 0.0.42 on 6 Sep, and a documentation release on 7 Sep adding Fluent Bit and GCP service attachment material.

Sentry pushed across language and runtime surfaces. It released sentry-python 2.68.1 on 2 Sep to prevent auto-collected logs stopping when enable_logs is true, Sentry 26.8.0 on 3 Sep with detector query and alert-editing changes, sentry-java 8.55.0 on 5 Sep with unhandled session state and ANR profiling moving out of experimental status, and sentry-unreal 1.22.0 on 6 Sep with native backend and Mac/iOS attachment fixes.

PagerDuty’s smaller but cleaner signal was library maintenance. It released python-pagerduty 7.0.0 on 2 Sep and 4 Sep, moving the upstream HTTP client to HTTPX2, then 7.0.1 on 5 Sep with documentation, maintainer and dependency updates. In a market selling operational trust, these low-glamour releases are not background noise. They are part of how vendors keep credibility with technical buyers.

Reliability incidents cut through the positioning

The category’s irony was hard to miss: the companies selling production assurance had their own reliability events in the same period. Grafana Labs reported a minor US Central production incident on 2 Sep that was resolved after about 294 minutes. On 3 Sep it investigated a major incident involving high latency in ap-south-1, alongside a resolved latency event for Grafana Cloud Logs in the same region. On 4 Sep it resolved a minor Tempo and Mimir read/write failure incident after about 162 minutes.

The pressure continued on 5 Sep, when Grafana Labs reported three incidents: US Central Region Instability with minor impact resolved in about 375 minutes, Degradation of Hosted Grafana in US Central Region with major impact resolved in about 219 minutes, and degradation affecting alert rule creation, deletion and updates. For Grafana Labs, that sat beside positive commercial momentum, including the 10,000-customer milestone observed on 5 Sep.

Sentry also had a noisy stretch. On 1 Sep it reported a major incident involving 504 request timeouts under investigation, plus a minor ingestion backlog that lasted about 132 minutes. On 2 Sep its status page showed 22 incidents, including degraded span ingestion in the US and a resolved 504 incident. On 3 Sep it reported degraded span ingestion in the US for about 374 minutes, later resolved.

Datadog reported two minor incidents on 3 Sep: delayed CICD Optimization, Code Coverage, Code Security and DORA data in monitoring status, and delayed DBM Monitor Evaluations resolved after about 17 minutes. None of these events changes the strategic direction of the market. They do explain why reliability proof, incident transparency and fast post-incident product communication still matter more than AI vocabulary.

PagerDuty chose profit discipline and automation

PagerDuty’s fortnight was the clearest example of a maturing software company balancing growth narrative, cash generation and restructuring. On 1 Sep it reported ARR above $500 million, free cash flow of $32.8 million and five consecutive quarters of GAAP profitability. On 2 Sep, Q2 fiscal 2027 coverage again highlighted $33 million in free cash flow, five consecutive profitable quarters, customer growth acceleration and flat overall growth. The same day, observed earnings-call coverage said non-GAAP EPS was $0.32 and that the company raised its margin outlook.

Then came the cost action. On 3 Sep, multiple articles reported an approximately 15% headcount reduction. On 4 Sep, further coverage described a second-quarter earnings report with declining earnings amid market weakness, while fiscal 2027 guidance was lifted and ARR topped $500 million. The public facts point to a company trying to defend the operating model while reasserting product relevance.

Product motion supported that message. On 2 Sep PagerDuty published posts on Scoped OAuth for Public Apps and PD Automation Runner for on-premises automation. It also created repositories for wiz-pagerduty-automation-rules and cursor-plugin on 2 Sep. The signal is not that incident response is disappearing into AI. It is that incident response vendors are pushing automation closer to security, developer tools and controlled enterprise environments.

Datadog carried the market's multiple narratives at once

Datadog had the broadest surface area of observed motion. It added sales, engineering, leadership and marketing roles across regions between 2 Sep and 7 Sep, including channel partner roles in security, regulated-industry marketing and distributed systems engineering. It refreshed content around AWS events on 2 Sep, published prompt-caching material on 3 Sep, added CI/CD and AI workflow content on 4 Sep, and launched runtime threat response content plus “LLM as a Judge” on 5 Sep.

The same period carried market and governance noise. Press coverage on 3 Sep reported a 5.57% stock close decline on 1 Sep and a planned $6 million stock sale by CEO Olivier Pomel. On 4 Sep, articles said the stock had dropped 22% in a month, while also noting a rise after Snowflake’s strong earnings. On 5 Sep and 6 Sep, multiple articles reported insider sales by executives including the CFO, COO, CEO, CTO and legal chief, alongside purchases by investment firms.

There was also a security footnote. On 2 Sep, six CVEs were observed for the Datadog Android application, originally published on 7 Aug, covering issues including exported activities without permission guards and plaintext storage of sensitive data, with CVSS scores ranging from 4 to higher severities as reported. For a company moving hard into runtime security and code-adjacent workflows, that kind of disclosure is manageable, but it increases the importance of crisp remediation messaging.