From Using DBs to Building Them: A 17-Year DTCC Journey on Technical Growth and AI-Era Data Innovation

Abstract

Over 17 years, the DTCC stage provided six speaking opportunities, with topics ranging from database performance, PostgreSQL distributed systems, Oracle migration to cloud, to lakehouse, AI Agents, and Semantic Storage. This mirrors the author's database career journey: no one can predict the future, and the path has been a continuous process of adjusting capability boundaries and following the market to find new problems.

Background: DTCC 17 Years, 1,606 Talks, Five Shifts in Chinese Database Landscape

The previous article "DTCC 17 Years, 1,606 Talks, Witnessing Five Shifts in Chinese Database Home Turf" showed a clear curve: from using international databases, to domestic substitution and compatibility, and now pushing "innovation" to the forefront in the AI era. But for practitioners in the moment, each present is fuzzy — you only know what you can do now, not which trends will last or which skills will be obsoleted.

The author is grateful for the DTCC stage. Six talks covered: second-level cache (2010), PostgreSQL distributed solutions (2015), breaking Oracle migration barriers on Alibaba Cloud (2018), building a cloud-era DBA knowledge system (2019), unified data service platform for trillion-scale lakehouse (2021), and how humans and AI Agents understand the same data with Cortrix (2026).

Figure 1: Six DTCC appearances, evolving identities and topics. Images from DTCC 2010, 2015, 2018, 2019, 2021, 2026 official speaker pages; text based on six official agendas.

This article discusses a practical question: how can a technical person connect their current capabilities with ongoing market changes to explore long-term development space?

1. Master Using Databases Before Changing Them

In 2010, China had 457 million internet users and 303 million mobile internet users. The state launched cloud computing service innovation pilots, mentioning distributed storage and massive data management. Yet for most database practitioners, the most pressing issues were not that grand.

The first DTCC's 29 sessions: 34.5% focused on high availability and operations. All 16 sessions naming specific database products featured international vendors. Core concerns were performance, HA, backup, recovery, and monitoring — how to use databases well.

The author was an architect and China product manager at EnterpriseDB (a PostgreSQL commercial distribution). EDB's strongest selling point was Oracle compatibility. The boss challenged the talk title "Reasonable Use of 'Second-Level Cache' Technology to Improve Overall Database Performance", asking why not promote Oracle compatibility to attract customers.

The author insisted: compatibility can be explained in one page — it only solves migration, not technical innovation. Compatibility shouldn't be a DTCC technical discussion topic. But the performance gap between memory and disk, and whether a cache layer can better utilize limited resources, is a discussable technical problem that earns peer recognition and sharpens one's own skills.

Figure 2: First DTCC official key visual. Source: DTCC 2010 official page.

A technical person's first step: deepen the immediate problem, build a capability that withstands real-world testing. Without this foundation, all development is empty talk.

2. Pushing PostgreSQL Is Not Just Learning Another Database

In 2011, the author co-founded the PostgreSQL Chinese Community. The debate centered on "autonomous and controllable". Many sought a lower-cost, more compatible replacement, questioning if open source could handle enterprise workloads. The author's view: open source is a development model; enterprise-grade support requires distribution vendors to solve via technical services.

More importantly: if we only discuss compatibility and replacement without understanding database design principles, participating in community, reading source code, or building engineering capabilities, switching products brings no new value.

By 2015, "Internet+" drove cloud, big data, and traditional industry integration. DTCC that year: 22.8% big data content, 14 distributed sessions. The author presented "PostgreSQL Distributed Solutions". He didn't chase the big data hype but extended existing capability: from running one database well to coordinating multiple machines; from personal usage to community understanding, dissemination, and improvement.

Summary of those years: Technical growth follows an I → T → π trajectory. [I] Deepen a specialty (I am PG); [T] Build cross-disciplinary skills on that base (community management, product management); [π] Find a second growth curve (now leveraging AI to shift from databases to data).

In today's AI acceleration: if you know databases, you don't have to immediately train models; first see how models access databases and handle unstructured data. If you know operations, don't abandon accumulation for Agents; first see which actions automation takes over and what new responsibilities remain. Transformation isn't wiping the slate clean — it's leveraging existing capabilities to pry open new ecosystems and opportunities.

3. From Community to Vendor: Understanding Why Customers "Dare Not Switch"

2018: cloud databases moved from concept to real business pressure. PolarDB launched in 2017, commercial in March 2018. That year's Double 11 hit 213.5 billion yuan GMV, over 1 billion logistics orders.

The author was at Alibaba Cloud responsible for database products, presenting "How Alibaba Cloud Breaks Oracle Migration-to-Cloud Barriers". Technically, migration seems like syntax conversion, data replication, and app refactoring. But working with real customers reveals: technical feasibility is just the start. Customers ask: how to rollback on cutover failure? What about existing ops systems? Who owns business interruption? Does the new cost model really hold?

Moving from community/user to vendor forced confrontation with these. Previously the question was "how to implement a feature"; as a product person, must understand why customers hesitate, how enterprises decide, and who bears risk.

By 2019, CAICT's cloud whitepaper noted fast-growing DBMS demand. The author's DTCC topic became "How to Build a Cloud-Era DBA Knowledge System". Industry anxiety: will DBAs disappear after moving to cloud? Judgment: cloud reduces repetitive ops, but never eliminates human judgment and responsibility. This holds in the AI era too. Patching, backup, scaling, monitoring can go to the platform; migration decisions, performance boundaries, fault accountability, cost, and business risk still need human understanding. AI provides candidate solutions and efficiency, but responsibility always rests with the AI user.

The more the platform automates, the less a technical person's value can stay at "I know how to operate this product."

Figure 3: 2019 DTCC on-site sharing. Original with ITPUB watermark. Source: Alibaba Cloud Developer Community DTCC 2019 talk coverage.

Figure 4: 2019 slide "Cloud Service Support Boundary": after platform handles basics, business DBAs still own schema design, standards, optimization, cost, and fault diagnosis. Source: Alibaba Cloud Developer Community.

4. From Product Expansion to Convergence: Why I Started Looking Beyond Databases at Data

2021: China's database industry was booming. CAICT counted 135 domestic database products: 81 relational, 54 non-relational; among relational, 58.03% were based on MySQL or PostgreSQL secondary development. Same year, Data Security Law passed, effective Sept 1.

Afterward, product count kept rising: 269 in 2024, down to 164 in 2025, adjusted to 182 in 2026. Numbers don't equal industry health, but show a shift from concentrated emergence to differentiation and convergence.

At that stage, more database products didn't automatically yield unified data capabilities. The author's 2021 DTCC talk: "Unified Data Service Platform Practice under Trillion-Scale Lakehouse Integrated Architecture". The topic no longer centered on a single database product, but on how to find, manage, and serve data uniformly after it enters multiple engines.

From using databases, to building databases, to stepping outside databases to view data — not leaving databases, but seeing their capability boundaries more clearly because of deeper involvement.

5. AI Makes "Innovation" a Database Opportunity Again

The previous article ended with a personal bias: "Xinchuang" (IT application innovation) shouldn't equal "substitution"; "innovation" should be the true main theme.

2025: AI, vectors, and Agents became DTCC's strongest annual theme signal for the first time; 2026 they occupied 55.1%. After model capabilities spread rapidly, the database industry faces more than adding a vector type or supporting more data models.

Agents read documents, call tools, save memories, and modify business state. Data infrastructure must answer not just "which record is most similar", but: where does information come from, when is it valid, what does it conflict with, can it serve as basis for next action?

2026, the author's sixth DTCC, brought the new open-source semantic storage project Cortrix ( https://github.com/cortrix/cortrix) with the topic "Not Only Vector: How Humans and AI Agents Understand the Same Data". This time the discussion wasn't about a database feature, but how data responsibility behind Agents should be divided. Vectors find similar content, but similarity ≠ current validity. After chunking and vectorization, are original source, version, applicability scope, and substitution relationships preserved? When multiple Agents use the same data, do they each maintain conflicting interpretations? When answers change, can we trace back to the original data and processing path?

Figure 5: From 2021 unified data service to 2026 semantic storage. Speaker images and topics from DTCC 2021 and 2026 official pages.

Cortrix is the productized answer to these questions. Not only open-sourced code, but opened issues and discussion spaces for community participation like PostgreSQL mailing lists. Also released reproducible benchmark methods and a full demo.

From database user, to community participant, to industry vendor and open-source project co-founder, the position keeps changing. The driver isn't a higher title, but the desire to explore new problems and domains. Used to care about making databases faster, then distributed scaling, cloud migration, serving more data; today's problem: how to let humans and Agents form a shared, understandable, usable, and traceable basis for the same data.

AI is a chance for the database industry to re-discuss "innovation". Don't have to just follow foreign product feature additions, nor reduce AI to a compatibility checklist item. Can rethink: in the Agent era, which data infrastructure problems remain undefined?

6. How Does Technical Growth Connect with Market Changes?

Six DTCC talks aren't a replicable career path. Everyone's starting point, resources, and risk tolerance differ.

Looking back, the author has driven himself with four questions:

Foothold : Where is my most confident capability starting to lose effectiveness? That spot is often closer to the real next step than the hottest buzzword.

Trend : Is the market change a short-term hype or a shift that will permanently reshape future development? Example: distributed systems offload consistency and fault tolerance to the system; cloud services offload ops and delivery to the platform; Agents are offloading semantics, provenance, and action basis to the data layer.

Reuse : Have I accumulated product-specific operation steps, or cross-product problem-analysis skills? If I switch databases, cloud platforms, or Agents, can my past experience still apply?

Trajectory : After each pivot, can I leave something verifiable? A talk, a document, a real case, a piece of code, an open-source project, even a community ecosystem — these prove I truly took that step more than a new title.

Figure 6: Foothold, Trend, Reuse, Trajectory — technical development path based on the four self-check questions.

Technical people can't accurately predict products ten years out, nor chase every daily hotspot. A more realistic approach: use your current most solid capabilities to catch the problems the market has just surfaced but hasn't solved well.

From using databases, to building databases, to stepping outside databases to view data — not saying goodbye to the past. It's bringing accumulated capabilities into a larger problem space.

References

[1] Scott's six DTCC official agendas: 2010, 2015, 2018, 2019, 2021, 2026

[2] CNNIC 27th Statistical Report on Internet Development in China (Central Cyberspace Administration PDF mirror)

[3] NDRC: Notice on Doing Well in Cloud Computing Service Innovation Development Pilot Demonstration Work

[4] State Council: Guidance on Actively Promoting "Internet+" Action

[5] Alibaba Cloud PolarDB Development History, PolarDB Commercial Launch Timeline

[6] Alibaba Group: 2018 Double 11 Data

[7] CAICT: Cloud Computing Development White Paper (2019)

[8] CAICT: Database Development Research Report (2021)

[9] Central Cyberspace Administration: Full Text of PRC Data Security Law

[10] Cortrix GitHub

[11] SequoiaDB Interview: Scott Xiao on PostgreSQL Chinese Community and Database Industry

[12] Database Development Research Report (2026) Release Interpretation