Skip to content
Main Site News Console

$70 Billion in Funding! DeepSeek Code Is Really Coming, Led by ACM Gold Medalist Cui Tianyi

· 量子位
国内AI

DeepSeek’s funding has surged to 70 billion yuan, and its leadership has told potential investors that it is prioritizing breakthrough AI research over short-term commercialization.

Image 3

Meanwhile, DeepSeek Code is now officially a real thing!

It was even posted by the company’s own senior researcher, Deli Chen, who personally put out a recruiting notice:

Image 4

In just a few lines, three keywords were practically laid out in plain sight:

Harness team, Code Harness, DeepSeek Code.

If you were to rank the AI world’s buzzwords of the year, Harness would almost certainly make the list.

AI coding tools have long since moved beyond programmer circles and into the mainstream productivity-tool arena.

Under the post, one commenter put it bluntly:

DeepSeek Code just sounds way more fitting than Harness!

Image 5

Image 6

What’s even more worth noting is that the team lead may also be changing.

Cui Tianyi, the former co-founder of TSY Capital who joined DeepSeek in March this year, is reportedly set to become the head of the Agent Harness team.

Image 7

This is one development after another.

DeepSeek’s own Code product may truly be on the way this time.

DeepSeek Code is Coming!

DeepSeek making its own Code product no longer feels sudden at this point.

Let’s rewind a few days.

On May 15, DeepSeek started “widely recruiting talent” on its official hiring site.

Among the latest openings, a position called Agent Harness Product Manager was prominently listed:

Image 8

Three days later, on May 18.

Another role, Agent Harness R&D Engineer, also appeared on the company’s recruiting site:

Image 9

That alone is quite intriguing.

Everyone can feel that Harness has been gaining more and more presence in the AI world this year.

Especially now that large-model capabilities are approaching their ceiling, the industry’s focus is shifting from how powerful the model itself is to a much more practical question:

How do you actually turn a model into an Agent that can reliably get work done?

In essence, a “Harness” is the execution, feedback, and error-correction system wrapped around a large model.

If an Agent is going to keep moving forward on real tasks, it needs a Harness to handle workflow orchestration, tool calls, result checking, and error rollback.

DeepSeek’s official job posting summed up that relationship very directly:

Model + Harness = Agent.

An even clearer signal is hidden in the requirements for the Harness role:

Image 10

Have extensive hands-on experience with products like Claude Code, Cowork, Codex, Cursor, OpenCode, GitHub Copilot, Manus, OpenClaw, Hermes, etc., and have integrated these products into your work and daily life.

Claude Code, Codex, Cursor, and GitHub Copilot point to AI coding entry points.

Manus, OpenClaw, and Hermes lean more toward Agent execution, task orchestration, and tool calling.

Add in the three keywords Prompt Engineering, Context Engineering, and Harness Engineering, and the direction becomes very clear.

DeepSeek is not just looking for someone who understands product.

It is looking for someone who has used AI coding tools deeply, understands Agent workflows, and knows where models tend to get stuck in real tasks.

(That is an absurdly high concentration of Code-era talent, I’m telling you!)

Image 11

And that’s not all; what comes next is even more worth a close look.

Just in the past couple of days, DeepSeek’s hiring pace has clearly accelerated.

In addition to product managers and R&D engineers, roles such as Agent Deep Learning Algorithm Researcher and Agent Data Strategy Engineer have also been posted all at once.

More than 10 positions within a single day.

Of course, these roles may not all point directly to DeepSeek Code or Harness, but they at least show one thing—

DeepSeek is indeed filling in more key pieces around Agent capabilities. (doge)

Image 12

Then came the real “smoking gun.”

DeepSeek senior researcher Deli Chen personally posted a recruiting message:

Image 13

Build a Code Harness from scratch.

Which means:

DeepSeek is really going to spin up a new team, bring in a fresh batch of people, and build a new product line.

The Team Lead Has Also Come Into View

Beyond the hiring signals, another more important clue has also started to emerge.

It’s not just that roles are being opened in rapid succession—DeepSeek’s new team also seems to already have its key leader in place.

Cui Tianyi, the former co-founder of TSY Capital who joined DeepSeek in March, is reportedly set to lead the Agent Harness team.

Image 14

That name carries real weight in the AI Code space.

Cui Tianyi earned his bachelor’s degree from the Department of Computer Science at Zhejiang University.

He was admitted to Zhejiang University through NOIP/informatics competition honors and won gold medals at the ACM International Collegiate Programming Contest Asia regional finals six times.

After graduation, he spent nine years working in Jane Street’s Hong Kong and New York offices before co-founding TSY Capital.

From competition prodigy, to core member of a quant firm, to DeepSeek Harness Team lead—his résumé itself is highly telling.

Image 15

On one side, the name DeepSeek Code is beginning to surface.

On the other, the Harness team, Code Harness, and key leadership are appearing in quick succession.

The signal that DeepSeek Code is Coming is genuinely strong this time…

Image 16

Some Alternatives to DeepSeek Code

At this point, many of you are probably already itching to ask:

When exactly will this so-called “DeepSeek Code” be available?

To be realistic, we’ll still need to wait a bit…

Based on what has surfaced so far, DeepSeek Code looks more like a project that has just entered the stage of taking shape.

Before it is truly opened to ordinary users, it may still take some time.

But that doesn’t stop us from trying out some “alternatives” in the meantime!!

Image 17

I’ve also put together a quick list of Code-style alternatives that you can use right now.

The first one worth mentioning is the GitHub sensation of the month, now sitting at 32.9k Stars: DeepSeek-TUI.

Created by American indie developer Hunter Bown, it has a very straightforward positioning—

A DeepSeek version of Claude Code. (Sounds familiar…)

Image 18

You can think of it as a TUI coding tool written in Rust.

It runs in the terminal just like Claude Code, but it has also been specifically optimized for DeepSeek.

It can handle tasks like reading and writing files, running Shell commands, browsing the web, managing Git, orchestrating sub-Agents, and connecting to MCP servers—basically everything Claude Code can do.

It also supports Skills, except it runs on DeepSeek V4 underneath.

On top of that, the author has prepared a Chinese README and dedicated configuration paths for users in China, with support for the TUNA Cargo mirror.

So if you want to get an early taste of the DeepSeek + Claude Code-style terminal coding experience, DeepSeek-TUI is definitely worth trying first~

Image 19

In addition, the official DeepSeek-maintained awesome-deepseek-integration repository also includes quite a few Code-related tools that are worth checking out~

For example, AIMatrices below.

It’s a low-barrier framework written in Rust that’s well suited for quickly building AI applications, especially for developers who want to create app tools fast on their own.

After packaging, the single executable is only about 40MB and works across platforms (Windows/macOS/Linux).

It supports quickly building, deploying, and maintaining all kinds of AI applications without writing complex code from scratch~

Image 20

Beyond desktop AI code editors, some mobile coding assistants built specifically for DeepSeek are also worth mentioning~

For example, 无思微程序 below is positioned as a DeepSeek mobile coding assistant.

Its biggest feature is support for generating mini programs, micro-games, and micro-apps from a single sentence.

Meanwhile, mainstream AI code editors like Cursor and WindSurf can also integrate DeepSeek models very well, making them suitable for more professional coding work on desktop.

Image 21

Beyond what was mentioned above, the Awesome list also contains plenty of Code-related alternatives.

You can go directly to the repository’s VS Code Extensions category, where there are many useful tools—pick whichever fits your needs~

Image 22

(There are plenty of alternatives, so no need to worry about choice~)

Finally.

It still has to be said that with players like Claude Code, Codex, and Cursor already having trained users to expect a lot, and a wave of open-source projects and community contributors rushing in behind them, DeepSeek is not exactly early to the table with DeepSeek Code.

But that doesn’t stop us from looking forward to it, or from wanting to use it.

After all, with code tools, there really is no such thing as too many…

I’m definitely going to keep my expectations high.

Image 23

References:

[1]DeepSeek-TUI:https://github.com/Hmbown/DeepSeek-TUI

[2]Some alternatives to DeepSeek Code:

https://github.com/deepseek-ai/awesome-deepseek-integration

[3]https://www.linkedin.com/in/tianyicui/