Taste is the new seniority test

Workers crowd a workbench admiring glowing holographic prototypes while one engineer studies his own apart

Claude Code and Codex, especially with the new Claude Fable model, changed what "entry level" means. You describe a product, go back-and-forth for a few minutes and you have a product! Things that once took years of learning now start with a sentence and take a few minutes.

The people who already write code are leaning on it just as hard: 84% of developers now use or plan to use AI tools, and among professionals, 51% use them every day, according to Stack Overflow's 2025 survey.

For example, youtuber PewDiePie showed off Odysseus, Sara Dietschy built StudioInspo. That used to be the kind of software you hired engineers for, now people build them in their free-time.

The first implementation is cheap, now. Anthropic studied how people use Claude Code and found that in 79% of sessions, the person just handed the task over and let the AI finish it. A Base44 product manager used Claude Fable 5 to build a native mobile development environment that was about 90% ready for production in 2.5 hours. But 90% ready is not production!

An engineer shines a flashlight into a translucent wireframe structure, exposing a red crack hidden inside

Taste is the new seniority test.

A senior developer can tell the difference between good software and bad software. That sounds obvious until both of them compile. Choosing the good product over the one that merely works IS seniority, especially now that bad software is so cheap to produce.

Bad software can have clean code, pass every test. It can demo beautifully and it can still confuse the user, solve the wrong problem, expose private data, bury the one useful action under five clever features, or introduce bugs that will only show up six months later.

Taste is choosing to shape the software around how the user thinks instead of making the user bend around it.

Taste is also noticing that a technically correct screen can ask the wrong question. It is cutting three features because the fourth one is the reason the product exists. It is knowing that today's clean abstraction is next quarter's trap. The model can produce code all day, but it can't decide what is a good output vs a bad output. That call is the job senior developers are paid to make, and own those decisions.

Developers already sense the trust gap. The same Stack Overflow survey found only 32.7% trusted the accuracy of AI output, and 66% said their biggest frustration was an answer that is almost right, but not quite. Every almost-right answer is a judgment call handed back to a human — and the human had to make it.

The person who clicks merge still owns what happens next. Anyone can build a one-person app now. Scaling it once real users depend on it still needs senior judgment. AI building does not replace code ownership. If you let AI write everything but cannot explain or repair it, you are not covering your own ass — at some point the code will break, and you will not be able to figure it out. The customer will not care whether Claude wrote it. The person who shipped it owns the failure, and the 2am fix!

Security is another important point to be covered by a senior. Veracode tested more than 100 models and found that 45% of generated code samples failed security tests. Forty-five percent failing means, sooner or later, someone's data will end up where it should not — and catching that before it ships is exactly the call the machine cannot make. "It works" and "we should ship it" are two different decisions.

A veteran engineer guides a finished machine off the assembly line to waiting customers

If you are a junior developer whose plan is to start at zero, join a company, and wait for a senior to hold your hand until you are useful, you are cooked. Semi-joking, but only semi.

Build your own apps. Put them online. Make them your portfolio. Show that you can talk to users, choose what matters, cut what doesn't, and repair what breaks. You do not need a company to hand you "business experience" before you can prove that you have taste. That's how you land on a dev job in the post-AI era.

I have spent more than ten years writing software, and I now write less code by hand than ever. The experience did not evaporate. It moved to the three jobs the machine cannot take: deciding what is worth building, recognizing what is wrong, and taking responsibility after it stops typing.

Build the app, prove your taste, and own what happens next.

References:

Taste is the new seniority test

Keep reading

Nate Lemos

Get new posts in your inbox