Translated from the Korean original. Some screenshots show the original Korean interface.
A seminar that started with a DBA's question
If you're a DBA who has ever traced an execution plan line by line to find why a single query slowed down, you've probably had the same thought at some point.
"Couldn't AI do this analysis for me?"
The "MS SQL Server Tuning with AI" Tech Seminar, held at EXEM headquarters on June 25, started from exactly this question. Focused on using GitHub Copilot chat as a hands-on AI collaboration partner, it ran in 3 sessions: the basics, using Foundry Local, and real-world SQL performance optimization cases. The room was full of DBAs and developers running SQL Server, each with similar concerns.
Those concerns came down to one question: "Do we use AI only as a code generation tool, or can we trust it with SQL tuning and database operations too?" The speaker, Seongsik Kim, a director at EXEM, went through the same question himself. After 19 years of database operations and consulting work, he says, "I felt both doubt and fear about AI, so I wanted to see for myself how it would change my work."
The pre-seminar attendee survey also showed AI still at an early stage of adoption. Asked about their experience using AI tools for DB work, 71% chose "never used it" or "only use it for code generation." Even those who had tried it for code generation had not yet reached the point of handing it performance analysis and tuning.
1. How much of SQL tuning can AI take over?
SQL tuning has long depended on a DBA's experience and know-how. When a slow query showed up, you had to analyze the execution plan, review the index design, find the bottleneck, and work out a fix. Experienced DBAs could do this quickly, but it still took a lot of time and focus.
Recently, AI has started to take over part of this process. According to Microsoft's documentation, GitHub Copilot in SSMS (SQL Server Management Studio) is available in SSMS 22 and later. Through natural language chat, you can ask database questions and get help writing, modifying, and optimizing T-SQL.
This seminar covered exactly this shift. In a live demo, the speaker asked for an analysis of the SQL being written in SSMS, and GitHub Copilot chat reviewed the query structure and execution plan together, then worked out an improvement direction. Work that DBAs used to interpret themselves is now done first by AI, and the engineer's role is shifting to verifying the result.
One thing needs to be clear, though. AI is good at producing a fast first diagnosis, but its answer is not always right. Microsoft also states that responses from GitHub Copilot in SSMS may contain inaccurate or low-quality content, and that a person should review the output before using it in their work.
The speaker, Seongsik Kim, also stressed, "Rather than applying what AI suggests as is, you must verify it with an engineer's experience and analysis." AI is a powerful collaboration tool that greatly speeds up analysis and widens its scope. But verifying the suggested results and making the final decision is still the engineer's job.
💡 In the end, to use AI well, you need to be someone who can verify its answers.
This topic is covered in more detail in the new book by the seminar speakers, director Seongsik Kim and consultant Byeongchan Lee: "SQL Tuning Guide with GitHub Copilot Chat."
Read the book announcement article👇
2. Beyond individual SQL tuning to DB operations as a whole
AI can now analyze slow SQL and suggest how to improve it. But in real operations, the harder work comes one step earlier: finding what slowed down and figuring out why.
The problem is that the cause of an incident is not always in the database. Today's databases run tightly connected to applications, servers, networks, and other systems, so even when a single slowdown occurs, it is often hard to pin the cause on the database alone. As microservices and hybrid cloud environments spread, this complexity has grown even further. The limits of people checking every log and performance metric by hand to find the cause are becoming clear.
That is why AI's scope is gradually expanding to operations as a whole. The role of operations platforms is also changing, moving beyond simply collecting and displaying data toward detecting anomalies and analyzing their causes.
In fact, the attendee survey for this seminar named "automatic incident and anomaly detection and root cause analysis (RCA)" as one of the areas of greatest interest for AI adoption. Attendees' interest is extending beyond individual SQL tuning into DB operations automation.
You can see this trend in exemONE, EXEM's observability platform. exemONE detects anomalies based on time-series data patterns, and its AI-based RCA (Root Cause Analysis) presents the root cause, scope of impact, and recommended actions. It analyzes applications, servers, and networks along with the DB, helping operators find the cause of a problem faster.
If AI-based SQL tuning answers "How do I improve this SQL?", AI-based observability is closer to tracing the root cause: "What is the problem, and why did it happen?" AI is expanding from a tool that helps individuals with their work to a tool that supports an organization's IT operations.
3. A requirement for organization-wide AI: closed networks
For AI to go beyond helping individuals and support an organization's operations, companies also need an AI environment that fits how they work. In particular, environments handling sensitive data, such as finance, the public sector, and manufacturing, often cannot freely use external generative AI services for security and privacy reasons.
GitHub Copilot chat, covered in this seminar, is also a cloud-based service by default. So its use may be restricted in security-sensitive environments such as finance and the public sector. Regulatory changes reflect this trend too. Korea's Financial Services Commission, through its "Roadmap for Improving Network Separation in the Financial Sector", set out a path to gradually allow generative AI in the financial industry. Cloud-based generative AI must meet additional security requirements, while AI running on the internal network can be used more broadly within existing security frameworks. That is why interest is growing in technology that lets organizations use AI safely even in closed networks.
As one answer to this need, the seminar introduced Microsoft Foundry Local in a separate session. Foundry Local is a local AI runtime that lets you run AI models in a local environment such as a personal PC, so you can use generative AI without sending data outside.
If Foundry Local lets individuals or developers run AI locally, EXEM's enterprise LLM platform eXemble extends this to on-premises and closed-network AI use across an organization. eXemble runs LLMs in on-premises and closed-network environments and brings together internal manuals, technical documents, and operational knowledge so they can be used through natural language. It also lets users query and analyze real-time performance data and operational information in natural language to get the insights they need.
Just as GitHub Copilot chat lets you work with SQL in natural language, eXemble lets you use internal knowledge and IT operations data in natural language. The difference is that Copilot is an AI collaboration tool centered on code and queries, while eXemble uses internal knowledge documents and operational data together as context. The scope expands from AI that boosts individual productivity to AI that connects an organization's knowledge with its operational data.
In the end, what matters for organization-wide AI is not simply a better AI model, but having a runtime environment that fits the company and a reliable operating framework alongside it.
4. Time to prepare for organization-wide AI
This seminar was not just about how to use GitHub Copilot chat or SQL tuning techniques. It looked at how AI can expand from SQL tuning to DB operations as a whole, and on to an AI environment for the entire organization.
AI has not reached the point of taking over all operations work. But as the seminar showed, AI is not a replacement for skilled engineers. It is closer to a collaboration tool that widens the speed and scope of analysis. What matters next is fitting AI to the nature of the work and the company's environment, so it grows from a personal productivity tool into something that supports operations across the organization.
That is the direction this seminar laid out: start with SQL tuning, support DB operations as a whole, and then build an AI foundation that fits the organization. This is where AI-based DB operations automation is heading.
EXEM plans to continue hands-on Tech Seminars every quarter on a range of DBMSs, including Oracle, PostgreSQL, and MS SQL Server.
Curious about everything from individual SQL tuning to organization-wide AI?
Check out EXEM's AI-based IT operations platform.
Sources
SQL Server Management Studio (SSMS) 22.4.1 and GitHub Copilot in SSMS (Generally Available) - Microsoft Fabric Blog (2026.03) Generative AI use in finance rises as regulators announce network separation roadmap (Korean) - Asia Economy (2024.08)
Related articles
