I recently came across a Reddit discussion with a very direct title: why hasn’t AI made software engineers’ jobs any easier?

That question stuck with me. Working in Silicon Valley — having led teams, and now back on the front line doing applied AI — my own observation is this: tools really have made many things faster. But the scope we are expected to own, the number of things we are expected to try, and the speed at which we are expected to deliver have all been raised at the same time.

There are really two forces pushing everyone forward at once.

Force one: you work on AI, so you have to be faster

If you build AI products—or the tools and platforms behind them—for a tech company, you work in an environment that never stops moving.

Every week or two, there is a new model or capability worth evaluating, and every day brings another product, tool, or announcement. A limitation you mapped yesterday might be broken by a new model today. An approach you settled on last week may deserve a second look this week.

That pace is exciting. You can genuinely watch things that were impossible become doable overnight.

But it also means that learning, evaluating, experimenting, and changing direction have become a steadily expanding part of the job. You ship, and at the same time you have to keep asking: is this path still the right one?

Force two: even if you don’t work on AI, you are expected to do more because of it

The second pressure comes from expectations about efficiency.

If AI helps you code faster, organize information faster, and build decks faster, then surely the same amount of time can produce more output and take on more tasks.

We used to say “talk is cheap.” Increasingly, it feels like “code is cheap.”

In the teams I observe, the old pattern was to write a document first so everyone could align on the problem and the direction. Now the expectation is more often: build a prototype first, then discuss it.

That has real advantages. Abstract ideas become something you can touch. Discussions get concrete, and misunderstandings surface earlier. Ideas that were never worth the implementation cost now get a trial run.

But that is also exactly where the extra work comes from.

Prototyping got cheaper. Understanding the prototype, validating it, and deciding what to do next did not get cheaper at all. After version one, there is new feedback, new comparisons, and “what about the other version?”

You used to choose one direction, then start building. Now you can build three directions and compare. More experiments are possible, but the people who try them, discuss them, revise them, and judge them all have to invest more time too.

Every iteration is cheaper. When the number of iterations grows along with it, the total workload does not necessarily go down.

This is spreading beyond engineering

Marketing can generate ten versions quickly, but someone has to judge which ones fit the brand and which might mislead.

Design can explore more directions faster, but teams may then expect to see more options.

Product managers can organize requirements and prototype faster, yet they still have to resolve contradictions, set priorities, and decide what not to build.

Legal, finance, research — many jobs will see a version of this: first drafts become easy to get, while the volume of material requiring professional judgment grows.

A workday used to have a rhythm of thinking, executing, organizing, and checking. As the familiar execution parts get compressed, what remains concentrates on review, trade-offs, exceptions, and accountability.

Output can speed up. Human understanding and judgment cannot speed up by the same multiplier.

For managers, the pressure stacks one level higher

This is the part I think about most, from my years leading teams.

AI can organize status updates, consolidate information, and prepare meetings. So organizations may expect a manager to lead more people and drive more projects at the same time.

Meanwhile, everyone’s output and experiments have increased too. More people, more proposals, more prototypes, more decisions that need coordination — all of it lands in the manager’s scope.

Managers do not need to check every detail personally. But they still have to understand the key trade-offs, build the right review mechanisms, handle conflicts, and make sure the different directions connect in the end.

When the scope grows and everyone’s output grows with it, the amount a manager has to absorb and judge grows too.

AI can summarize for you. That does not mean you understand. Seeing the status update does not mean you have enough time to spot the risk.

What I actually worry about: the room we have left for good judgment

My own feeling is conflicted. I love what AI makes possible, and I enjoy turning ideas into reality faster.

But I also find myself asking: as everyone’s scope keeps expanding and the pace of delivery keeps accelerating, can we still protect the depth of our understanding?

If every saved minute is immediately filled by the next round of experiments, we may become busier and more productive—and still lose the room to think things through.

So the skill that matters most to me now is deciding what is worth trying, how far is far enough, and when to stop and think clearly about what we already know.

AI gives us the ability to do more. It also makes “deciding what not to do” more important than ever.

Originally published in Chinese on Facebook (2026-09-22). This English version is a translated and lightly edited edition.