Explainability will drive observability
Gartner predicts explainable AI (XAI) will push observability spending into half of GenAI deployments by 2028 — a major shift from today’s ~15% focus on XAI. That implies platform teams must bake session-level traces, explainability dashboards, and automated governance into SDKs and orchestration layers to meet rising enterprise and regulatory expectations. (cxotoday.com)
Gartner published the prediction on March 30, 2026 and separately forecast the global GenAI models market will top $25 billion in 2026 and reach $75 billion by 2029. (gartner.com) Gartner explicitly frames XAI as capabilities that “describe a model, highlight strengths and weaknesses, predict likely behavior and identify potential biases,” and it says LLM observability must surface non‑traditional metrics such as hallucinations, bias and token utilization. (gartner.com) Commercial observability tooling is already offering session‑level traces and SDKs: Datadog maintains an LLM Observability GitHub repo with notebooks and examples for tracing agent and workflow spans. (github.com) Datadog’s official docs show Python, Node.js and Java instrumentation options and an LLM Observability API for capturing traces, metrics and evaluations with minimal code changes. (docs.datadoghq.com) OpenTelemetry has published GenAI semantic conventions and a gen_ai attribute registry (including gen_ai.conversation.id and gen_ai.evaluation.explanation) plus agent span conventions to standardize session, token and tool‑call tracing across providers. (opentelemetry.io) Vendor platforms are pairing explainability dashboards, replay/playground tools and programmable hooks: Arize advertises trace-based replay and evaluation-driven CI/CD for agents, Fiddler markets LLM observability with explainability and governance workflows, and WhyLabs provides explainability tooling plus an open‑source LangKit for LLM signal extraction. (arize.com) Regulatory pressure is tightening: the EU AI Act sets transparency and explainability obligations for providers with phased enforcement dates (transparency rules already in force for some model categories and wider obligations by August 2026), and U.S. regulators such as the FTC have launched enforcement initiatives targeting deceptive AI practices. (digital-strategy.ec.europa.eu) Gartner and the vendor landscape converge on concrete platform patterns enterprises are adopting now—instrument SDKs and orchestration layers with OTEL‑compatible traces, embed explainability hooks for post‑hoc and human‑in‑the‑loop validation, and surface governance metrics (citation accuracy, hallucination rates, token costs) in dashboards for SREs and compliance teams. (gartner.com)
