Recently our company has been pushing one thing: transitioning frontend towards full-stack.

The big boss said at the quarterly meeting:

From now on, frontend developers can’t just write frontend anymore.

When that sentence landed, my heart actually sank a little.

It’s not that I’m unwilling to learn new things. Anyone in tech knows that if you stop learning, you fall behind. What really gave me a dazed feeling was suddenly realizing that the “frontend path” I originally chose might be turning into a different road.

It’s no longer just about pages, components, interactions, and engineering.

It’s starting to become about APIs, data, deployment, closed business loops—even AI agents.

And this change isn’t happening slowly off in the distance.

It’s already right in front of us.

The era didn’t give a heads-up

The internet industry over the past decade-plus has actually reshuffled the table several times.

The earliest was the PC era. Portals, search engines, forums, backend systems—frontend started from slicing images, writing pages, and fixing compatibility. Back then, many people’s daily work was HTML, CSS, jQuery, gradually turning static pages into something interactive.

Then came the mobile internet era.

Around 2014, the number of mobile users surpassed PC. App startups exploded, and every business wanted to squeeze into that phone screen. By 2016, Douyin launched, Kuaishou rose, and mobile became the absolute main battlefield. Cross-platform frameworks, Hybrid, React Native, Mini Programs—all of these were pushed forward during this phase.

Frontend went from “writing web pages” to “writing apps.”

And later, now.

ChatGPT appeared at the end of 2022. In 2024, the concept of agents began to take shape. From 2025 to 2026, all kinds of agent frameworks, AI IDEs, code assistants, and automated development tools erupted fully.

This time, the change isn’t just “frontend needs to learn a new framework.”

This time, the change feels like it’s aimed directly at “the act of development itself.”

And today, I somehow find myself standing right on the crest of this wave.

Not because I was ready.

It’s the era that pushed people here.

I wasn’t excited; I was just scared

To be honest, I didn’t feel excited at first.

I felt scared.

This fear isn’t mine alone. It flows everywhere in the company.

Vertically, the organization is pushing: architecture restructuring, tech stack changes, headcount tilting toward AI, business units demanding higher human efficiency.

Horizontally, colleagues are infecting each other: conversations in the break room, shares on social feeds, hiring data sent to group chats, discussions at the dinner table about which department is shrinking again, which direction is heating up.

A lot of the time, anxiety isn’t something you conjure up on your own.

It’s that you might still be able to keep steady, but then everyone around you starts saying “it’s over,” and you begin wondering if you’re just reacting too slowly.

I happened to land right on this timeline myself.

In 2024, I interned at Ctrip. Back then people used ChatGPT more like a slightly smarter search engine: write some regex, explain an error, help generate a bit of boilerplate code.

By mid-2025, Cursor had spread everywhere. Fixing bugs, writing components, adding tests, reading code—many tasks no longer required you to type from first line to last.

Then in 2026, tools like Codex came out.

That felt different.

It wasn’t just that it could write code. It could read the project itself, understand context on its own, break down tasks, make decisions—even find a path through an unfamiliar codebase.

That moment I was a little terrified.

Because you suddenly realized it didn’t feel like a tool.

It felt more like a junior teammate.

And in some scenarios, even more reliable than a junior teammate.

The cruelest change happens at the entry point

If the anxiety inside the company still carries a vague sense of “transition period,” campus recruiting is another kind of naked reality.

I’ve watched three cohorts.

The class of ’25: I followed friends through the summer and autumn recruiting circuit. Back then, having one big-tech internship was already a strong plus.

The class of ’26: I entered the arena myself. By this time, one or two internships had become common. Three internships no longer sounded like a legend.

By the class of ’27, as I watch juniors rush into autumn recruiting, two big-tech internships have almost become the baseline.

Without that, many résumés don’t even reach the interview stage.

They simply never get that far.

In three years, the bar has been forcibly raised several floors.

Not because students suddenly got much stronger, nor because everyone genuinely fell deeper in love with tech. The more realistic reason: headcount is shrinking, slots are fewer, and screening standards naturally rise.

When opportunities were plentiful, companies were willing to nurture new hires.

When opportunities dwindle, companies only want people who “can hit the ground running.”

That’s brutal, but that’s today’s reality.

Employees are afraid, and companies are too

Now the scenario everyone fears most is pretty similar.

One normal workday, you’re suddenly called into a meeting room by your manager. The door closes, and the other person begins saying this year’s environment is tough, business restructure, organizational optimization—then that sentence slowly drops:

“This round happens to involve you.”

That’s not a joke. People around me have actually experienced it.

So it’s normal to be afraid.

But later I gradually realized that it’s not just employees who are afraid—companies are afraid too.

Companies are afraid of falling behind in this wave of AI.

Afraid that competitors have already reengineered their business processes with agents while they’re still piling up manpower on requirements.

Afraid that rivals have doubled their per-person efficiency while they’re still in meetings debating “should frontend learn backend.”

Afraid of turning from an internet company into a “legacy internet company.”

That’s why you notice that many changes are not the impulse of some leader—the entire environment is already forcing organizations to move forward.

Employees fear being eliminated.

Companies fear being eliminated by the era.

Everyone’s in the same boat; some stand on the deck, some stand in the cabin.

The boat is rocking; no one is steady.

Stop asking whether AI will replace you

I used to turn the same question over and over:

Will AI replace me?

Later I thought, the question itself doesn’t matter much.

Because the answer is likely: it will replace part of you, and also reshape part of you.

It won’t suddenly chop everyone off in one stroke, but it will first eat up the work with clear boundaries, high repetition, and that only requires execution.

Write an API—AI can do it.

Write a page—AI can do it.

Write a test case—AI can do it.

Even give it enough clear context, and it can write it fairly well.

So the real question isn’t “Will AI replace me?”

It’s:

If AI can complete more and more execution work, what value do I have left?

The answer I’ve arrived at for now is “independent end-to-end delivery.”

The most important capability going forward is getting things done

Before, when a requirement came in, the chain was long.

Product writes the PRD, frontend waits for API docs, backend waits for field confirmation, testing waits for dev to hand off for QA. Everyone only handled their own segment—like an assembly line.

That model made a lot of sense in the past.

Because complex systems need division of labor, and division boosts efficiency.

But after AI appeared, a lot of “execution steps within the division” are being compressed.

An API that used to take someone half a day can now generate a draft in ten minutes.

A page that used to take a day to build can now be 70% done by AI first.

Unit tests, scripts, documentation that used to require dedicated effort can all be quickly generated by tools.

That means the value of pure execution will drop.

What becomes truly valuable is the person who can start from a vague requirement, think through the problem, settle on an approach, build the system, and deliver the result.

That is the person who can independently close the loop.

You don’t necessarily have to be an expert at every layer, but you have to be able to cross them.

You need to know how frontend connects to business, how the backend organizes APIs, how data flows, how services deploy, how to troubleshoot when something breaks, and finally how to make the thing actually run.

Future R&D won’t resemble assembly-line workers so much.

It will look more like a small-system owner.

The boundary between “frontend” and “backend” will thin out

So I increasingly feel that the boundaries among frontend, backend, QA, and test development will become thinner and thinner.

This isn’t about anyone stealing someone else’s job.

It’s that “job roles” themselves are a product of the industrial-era division of labor. In the past we needed to break a complex process into many segments, with different people responsible for different parts.

But AI is changing that premise.

When tools can fill in lots of execution details for you, it becomes possible for one person to cover a longer chain.

In the future, companies may not care as much about whether you are a “pure frontend” developer.

They’ll care more: if we hand this requirement to you, can you drive it forward?

Can you troubleshoot issues on your own?

Can you collaborate with AI?

Can you string together frontend and backend, data, deployment, and monitoring?

Can you be accountable for the business outcome?

If you can, you’ll have value.

If you can’t, and you cling only to “I’m frontend, that’s not my scope,” the risk will just keep growing.

For newcomers, this isn’t necessarily all bad

For juniors who haven’t yet entered the field, this change is certainly brutal.

But it might not be all bad news.

In the past, people always obsessed over which direction: frontend, backend, client-side, test development—which has more promise? Which makes it easier to get into a big tech firm? Which sits closer to the core?

Some directions even naturally carried a hint of a superiority chain.

But looking ahead, those distinctions may gradually weaken.

Because all directions will be pulled into a single larger role: business development.

What matters in the future isn’t which position you start from, but whether you can continuously expand your capability radius.

Frontend background? Doesn’t matter.

QA background? Doesn’t matter.

Client-side background? Doesn’t matter.

As long as you can independently get things done, you’ll have opportunity.

Tech stacks will change, job titles will change, organizational structures will change.

But the ability to get things done won’t depreciate.

After the bubble, what will remain?

At this moment, I sometimes feel a collective mood that feels very familiar.

“Catch up with OpenAI.”

“Everyone must learn AI.”

“All products must be AI Native.”

These phrases appear in weekly reports, in OKRs, in company meetings, like a massive collective mobilization. Everyone knows the direction matters, but everyone also vaguely senses that along the way there’s bound to be bubbles, recklessness, misjudgments, and wasted effort.

The AI direction is probably right.

But not every charge is right.

Not every agent project has value.

Not everything labeled “AI Native” is truly native.

And not everyone who is pushed to transform can comfortably make the turn.

When the wave comes, some people get swept in, some get knocked down, and some genuinely build real skills inside it.

After the bubble clears, what remains won’t be slogans.

It will be capability.

It will be the ones who actually built things with their hands, stepped into the pits, understood the business, and assembled systems.

Scared as I am, I’ll start moving

I’m writing this not because I’ve figured everything out.

On the contrary, I’m scared too.

I also stare at the screen late at night wondering: did I choose the wrong path? Is frontend dying? Is it too late for me to pick up full-stack now? If AI takes one more step forward, where can I still stand?

But later I realized that mulling over these questions by itself doesn’t make me any safer.

What truly gives me a bit of steadiness is taking action.

Yesterday, learn a bit about agents.

Today, fill in a bit of backend.

Tomorrow, try to complete a small requirement end to end.

I might not become strong right away, but at least I won’t stay frozen in place.

The ones who can survive this round might not be the ones shouting the loudest, nor the ones who worried first.

More likely they’ll be the ones who are scared—and still began to move their hands.

That’s probably how it is for me too.

Scared as I am.

The road has already changed, so I’ll just follow it forward.