How Large Language Models Are Transforming Data Development and Developer Roles

Hello everyone, the new year has passed and we meet again as 2025 is already one‑sixth over.

During the holidays, large‑model tools like DeepSeek dominated the headlines, and many have seen them.

This post rides that wave, addressing frequent questions from advanced‑class students about the impact of AI and large models on the data‑development field.

Personally, I have been using these models for a long time; the most common tools are Cursor (Claude 3.5/GPT‑4), as well as Doubao and DeepSeek, which I rely on heavily at work, and our development platform also integrates some self‑built models.

Indeed, I have been using DeepSeek and Doubao for almost half a year.

If you have ever asked me for model tool recommendations, I have suggested them; some students asked what Cursor/DeepSeek can do, so I will give you a quick assessment.

In current development, over 50% of code writing, more than 40% of task‑optimization suggestions, and over 70% of documentation drafting and polishing can be assisted by large‑model tools.

So, what impact can this AI capability bring?

It gives the strong an extra set of wings, while the weak tread on thin ice.

Take Cursor as an example: its code completion, hints, and optimization abilities far exceed the average developer, and its Composer feature can generate whole projects and refactor code with a single click.

SQL conversion tasks, such as rewriting Spark SQL to Flink SQL or generating Doris SQL code, can be handed over to Doubao/DeepSeek or other internal assistants and completed effortlessly.

Previously manual analysis of job failures (e.g., data skew, OOM) can now be done by feeding logs to DeepSeek, which quickly pinpoints root causes and suggests optimizations like adjusting shuffle partitions or improving join strategies.

Of course, the generated code may have performance issues, and because frameworks evolve rapidly, cloud‑dependent model tools might not always provide the best advice, so developers must stay up‑to‑date and critically review suggestions.

Consequently, some low‑end development positions will inevitably shrink; the "AI‑assisted development + human deep optimization" model is already happening.

Challenges and opportunities coexist, and new roles will emerge much like the mobile‑era did, but only those with strong learning ability, motivation, and intellect will seize them.

In summary, if you haven’t tried the tools mentioned above, you should start immediately.

Moreover, if you are not yet an expert in data development, you should feel a sense of urgency because the window of opportunity is narrowing.