Why Traditional Ops Stalls and How AI‑Driven Solutions Can Revitalize It

Background

When operations teams find themselves in a bind, should they cling to old habits or embrace new approaches? Below are several typical situations they face:

Release changes become a formality.

Deployment steps are overly complex, with hundreds of systems and inconsistent documentation.

Hundreds of systems, dozens of OSes and databases lack unified standards.

Security controls are insufficient.

Management views problems only from a top‑down perspective.

In response, various suggestions emerge from the field:

Low standardization makes automation effort‑heavy.

Standardized processes need system support, e.g., change requests submitted and reviewed in a system.

Management chaos cannot be solved by operations alone.

Regulatory pressure, especially in finance, demands zero tolerance for violations.

Without addressing the first issue, the others persist.

Fundamentally, standardization is lacking.

Focus on solid change management first; CMDB and automation can follow later.

Change processes are unlikely to move forward without clear boundaries.

Some prefer to maintain the status quo.

Some follow leadership directives to complete daily inspections.

Operate within defined responsibilities and adhere to change scripts.

Assign problem owners to solve their own issues.

Planning vs. Failure

Identifying problems is easy; solving them is hard. Key keywords from the suggestions include:

Operations automation

Standardized processes

CMDB

Operations inspection

Governance policies

Industry regulations

These raise the question: can a single operations person handle all these tasks? Since many issues span multiple departments and industries, it is essential to define clear boundaries and focus on what is feasible.

Why are these valuable operations initiatives rarely implemented? They require long‑term investment and slow‑to‑show results, often needing a 1‑2 year commitment to build solid foundations.

AI‑Powered Operations: A Real‑World Example

The 2023 CCF International AIOps Challenge champion, ByteDance’s SRE‑Copilot, demonstrates how large language models can transform operations:

Operations Planning : Parse user requirements, generate natural‑language workflows, and select appropriate system components to create executable workflows.

Operations Visualization : Use natural‑language interaction to execute simple data queries and visualize fault data.

Anomaly Detection : Support multimodal data, orchestrate multiple agents, and integrate diverse platform data to dramatically reduce MTTR.

Root‑Cause Analysis : Unsupervised approach that leverages expert knowledge and historical faults, achieving high accuracy on known issues and reasoning on unknown ones.

Fault Classification : Classify incidents based on expert experience and historical data to aid post‑mortem analysis.

Automated Remediation : Recommend self‑healing actions after fault classification, allowing operators to focus on higher‑level tasks.

Code Generation : Generate scripts from user prompts, cutting development time from hours to minutes.

Fault Reporting : Automatically produce diagnostic reports in natural language covering the 5W (When, Where, Who, What, Why).

Knowledge‑Base Q&A : Provide private‑domain answers from internal knowledge bases, improving response accuracy and reducing on‑call workload.

Large language models are poised to become more persuasive than seasoned experts, clarifying the boundaries of what operations can achieve.

Key Takeaways

Strive for solutions that serve the current situation, even if they are not perfect.

Balance in‑house development with external solutions based on resource constraints.

Maintain a solid foundation of asset management, monitoring, and configuration management to adapt to change.

Conclusion

When operations are stuck, the decision to stay the course or innovate depends on corporate culture and team dynamics, but recognizing the right direction is often more crucial than knowing every implementation detail.