Enterprise AI Evolves: Explainable Troubleshooting, Context Engineering, and Composable AI

AI has moved from the hype cycle into everyday operations for large-scale software systems. A vivid illustration is Chronosphere, the New York observability startup valued at about 1.6 billion dollars, which this week rolled out AI guided troubleshooting designed to help engineers diagnose and fix production failures. The promise isn’t only speed; it’s understandability: AI that explains its reasoning and shows the evidence behind each suggested next step. Central to this is Chronosphere’s Temporal Knowledge Graph, a living map that ties together services, infrastructure dependencies, and system changes across time. It enables engineers to see how a failure unfolded, what changed, and why, so they can trust the guidance rather than merely react to signals.

In a market where data volumes grow and AI-generated code accelerates, troubleshooting remains stubbornly manual. Chronosphere positions its solution as four core capabilities: automated Suggestions that propose investigation paths backed by data; the Temporal Knowledge Graph itself; Investigation Notebooks that document every troubleshooting step for future reference; and natural language query building that makes the system approachable. The practical value emerges most clearly in real scenarios—when an SLO alert fires and the system immediately surfaces a top-ranked path, such as a likely root cause in a dependent service. Engineers can inspect the charts, see the reasoning, and decide whether to dig deeper, all within a single view rather than bouncing across tabs and tools.

What makes Chronosphere’s approach noteworthy is its emphasis on “showing the work” instead of making automatic decisions behind the scenes. Its CEO argues that transparency and control are essential to avoid confident but wrong guidance. This contrasts with older forms of AI for observability that stopped at correlating anomalies or producing fluent explanations but lacked causal reasoning. Chronosphere’s stance aligns with broader enterprise trends: to trust AI, teams need visibility into how conclusions were reached, and they want to see the exact data points and change events that fed the model’s conclusions. The company also highlights a practical strategic point—custom telemetry matters. By normalizing non-standard telemetry and connecting changes to incidents, the platform aims to close gaps that standard service maps miss. The result is not only faster incident resolution but a knowledge base that travels with the organization across future outages.

Beyond Chronosphere, the enterprise AI narrative is increasingly about building a composable, trusted stack rather than chasing a single all-in-one platform. Baseten’s pivot toward training alongside inference embodies this shift. Baseten Training lets enterprises fine-tune open-source models without the headaches of managing GPU clusters, multi-node orchestration, or capacity planning. The company emphasizes ownership of weights—customers can download and move their fine-tuned models—while Baseten handles low-level infrastructure, multi-cloud GPU scheduling, and automated checkpointing. Real customers report impressive gains: cost reductions, latency improvements, and faster iteration cycles as teams craft domain-specific models that outperform generic baselines. The broader takeaway is clear: control over weights and data, paired with robust inference infrastructure, can unlock competitive advantages without surrendering agility or price discipline.

The idea of context engineering also appears in developer tooling designed to keep humans in the loop. A case in point is Qodo, which helps large teams review code at scale by teaching models to understand the context around a pull request—the code diff, prior discussions, tests, and architectural conventions. At monday.com, Qodo has reduced unknowns in code reviews by catching subtle issues such as insecure configurations and architectural violations, all while surfacing tailored suggestions that reflect a team’s actual practices. The integration into GitHub pull request workflows demonstrates how context-aware AI can act as a teammate rather than a distant autopilot: developers retain final control, but benefit from smarter prompts, deeper analysis, and faster feedback cycles. This is the kind of in-workflow AI that enterprises say they want—assistive, accountable, and tightly integrated into existing processes rather than a disruptive bolt-on.

In parallel, Celonis at Celosphere 2025 reframed enterprise AI as “process intelligence” that lives inside the organization’s actual workflows. The platform’s living digital twin of operations, anchored by a powerful data core and a Process Intelligence Graph, enables companies to analyze where processes stall, design future states with guardrails, and operate with AI copilots that align with human workers and systems. Mercedes-Benz described how process intelligence unified data across plants, suppliers, and logistics, enabling faster action during a semiconductor crisis. Vinmar highlighted automation of order-to-cash processes and the pursuit of non-algorithmic problems, while Uniper demonstrated maintenance prediction for hydropower assets with Microsoft’s AI stack. The thread across these stories is striking: AI works best when it is grounded in process context and orchestrated across ecosystems, not when it exists as a standalone feature. Celonis’ emphasis on composability, cross-vendor integrations, and agent-enabled orchestration underscores a broader industry move toward open, interoperable AI that can scale across complex enterprises.

Taken together, these developments sketch a pragmatic path for the AI era. Enterprises are moving from lightweight automation to trustworthy, context-rich intelligence that helps humans make better decisions. They are embracing training and inference as parts of a continuum, not separate products to be locked in a single vendor. And they are demanding transparency, control, and meaningful coverage of custom telemetry so that AI guidance remains grounded in reality rather than fashionable demos. As the landscape expands—from AI-guided troubleshooting and context-aware code reviews to open-weight training platforms and process-centric enterprise AI—the most durable winners will be those who show their work, respect human judgment, and enable teams to govern AI-driven outcomes with confidence.

Sources

1. https://venturebeat.com/ai/chronosphere-takes-on-datadog-with-ai-that-explains-itself-not-just-outages
2. https://www.theguardian.com/society/2025/nov/10/ai-chatbots-stop-prisoner-release-errors
3. https://aibusiness.com/generative-ai/ibm-and-andre-agassi-sports-firm-team-up-on-ai-tennis-app
4. https://www.theguardian.com/technology/2025/nov/10/sam-altman-can-openai-profits-keep-pace
5. https://www.theguardian.com/artanddesign/2025/nov/10/can-art-enhance-your-life-heres-what-i-learned-from-ali-smith-tracey-emin-claudia-winkelman-and-more
6. https://venturebeat.com/ai/how-context-engineering-can-save-your-company-from-ai-vibe-code-overload
7. https://venturebeat.com/ai/baseten-takes-on-hyperscalers-with-new-ai-training-platform-that-lets-you
8. https://www.theguardian.com/lifeandstyle/2025/nov/10/chatgpt-dating-ick
9. https://www.theguardian.com/film/ng-interactive/2025/nov/10/morgan-freeman-interview-nelson-mandela-six-decades-on-screen
10. https://venturebeat.com/ai/celosphere-2025-where-enterprise-ai-moved-from-experiment-to-execution