{"id":30039,"date":"2025-10-27T07:58:32","date_gmt":"2025-10-27T14:58:32","guid":{"rendered":"https:\/\/www.pingcap.com\/?p=30039"},"modified":"2026-01-07T04:41:16","modified_gmt":"2026-01-07T12:41:16","slug":"rethinking-scale-tidb-evolution-ai-agent-database","status":"publish","type":"post","link":"https:\/\/www.pingcap.com\/ko\/blog\/rethinking-scale-tidb-evolution-ai-agent-database\/","title":{"rendered":"Rethinking Scale: TiDB\u2019s Evolution Into an AI Agent Database"},"content":{"rendered":"

Recently, I\u2019ve been meeting customers across industries including AIaaS, Web3, and FinTech. No matter where I go, the conversation always turns to one thing: AI.<\/p>\n\n\n\n

Everyone\u2019s asking the same question: “How will AI transform our business, and how does TiDB embrace this change?”<\/p>\n\n\n\n

So, I started sharing what had been on my mind lately.<\/p>\n\n\n\n

\u201cThe era of AI is here,\u201d I said. \u201cBut the real shift is that the main users of databases aren\u2019t humans anymore, they\u2019re AI agents.\u201d<\/p>\n\n\n\n

That\u2019s usually when customers stop for a moment. You can see the idea land.<\/p>\n\n\n\n

Then I add, \u201cAI agents are going to change how we build systems. They’re moving us from predictable, rule-based design to something far more dynamic and non-deterministic.\u201d<\/p>\n\n\n\n

At that point, the room always goes quiet. You can almost watch people thinking it through.<\/p>\n\n\n\n

<\/span>From Determinism to Uncertainty<\/span><\/h2>\n\n\n\n

In the old days, systems were predictable. Developers designed schemas, wrote queries, and controlled every detail.<\/p>\n\n\n\n

Building an e-commerce site?
Start with a products<\/code> table \u2014 id<\/code>, name<\/code>, price<\/code>.
Need a list page? Just run:<\/p>\n\n\n\n

SELECT * FROM products ORDER BY id;\n<\/code><\/pre>\n\n\n\n
Everything was deterministic, by design. But in the AI era, those rules collapse. Now you tell an agent: \u201cBuild me a website.\u201d<\/p>\n\n\n\n
It invents tables and SQL queries you\u2019ve never seen. A minute later, it deletes them and tries again. This shift from deterministic logic to adaptive experimentation doesn\u2019t just change how we build systems, it redefines what \u201cscale\u201d even means.<\/p>\n\n\n\n
And that\u2019s usually when someone asks the big question: \u201cSo what does this mean for TiDB?\u201d<\/p>\n\n\n\n
\n
It’s the right question to ask because the move from human-driven to AI-driven systems redefines scalability at every level.<\/p>\n<\/blockquote>\n\n\n\n
<\/span>The Three New Scalability Challenges<\/span><\/h2>\n\n\n\n
To work well with AI agents, databases must now handle three new dimensions of scalability:<\/p>\n\n\n\n
    \n
  1. Data volume & concurrency scalability<\/strong>: Millions of AI agents read and write simultaneously, generating petabytes of data. What used to be \u201cOLTP scale\u201d is now AI agent scale.<\/li>\n\n\n\n
  2. Schema scalability<\/strong>: Each agent defines its own schema. A single system may need to manage millions of unique tables. Traditional databases fail not on data, but on metadata management.<\/li>\n\n\n\n
  3. SQL scalability<\/strong>: AI-generated SQL is unpredictable, messy, and ever-changing.
    The challenge isn\u2019t just can it run?<\/em> But can it run efficiently and affordably?<\/em><\/li>\n<\/ol>\n\n\n\n
    These are problems traditional databases were never designed to solve.<\/p>\n\n\n\n
    <\/span>A Happy Accident: How TiDB Evolved into the AIaaS Database<\/span><\/h2>\n\n\n\n
    When customers ask why TiDB can handle all this, I can\u2019t help but smile.<\/p>\n\n\n\n
    The funny thing is that TiDB wasn\u2019t built for AI-as-a-Service from day one. It just evolved<\/em> into the perfect fit.<\/p>\n\n\n\n
    Years ago, we received a message from Databricks<\/a>: \u201cCan you support one million tables?\u201d<\/p>\n\n\n\n
    We thought, That\u2019s wild. Who would ever need that many tables in an OLTP database?<\/em><\/p>\n\n\n\n
    Not long after, \uc544\ud2c0\ub77c\uc2dc\uc548<\/a> reached out with an even bigger ask: \u201cCan you handle ten million tables? We\u2019re building Forge<\/a> \u2014 a platform where third-party developers can create their own apps.\u201d<\/p>\n\n\n\n
    Atlassian explained that tens of thousands of developers would be building apps, each with its own schema, all running inside the same database.<\/p>\n\n\n\n
    \"\"<\/figure>\n\n\n\n
    A diagram showing a typical representation of a one-schema-per-tenant (database = schema) model for Atlassian before Forge<\/em><\/p>\n\n\n\n
    At first, it sounded crazy. But the more we thought about it, the more it made sense. So we built it.<\/p>\n\n\n\n
    And somewhere along the way, we realized something interesting: what looked like a quirky SaaS scaling problem was actually a preview of the future.<\/p>\n\n\n\n
    Each app in Forge behaved just like an AI agent, with its own schema, its own workload, its own data.<\/p>\n\n\n\n
    By solving that SaaS challenge, we had unknowingly built the foundation for AI-driven scalability, years before the AI era officially even began.<\/p>\n\n\n\n
    <\/span>Why TiDB Fits the AIaaS World<\/span><\/h2>\n\n\n\n
    Here\u2019s the short answer: TiDB\u2019s design naturally aligns with the new scalability model of AI.<\/p>\n\n\n\n