Your Coding Agent May Be Writing Yesterday’s JavaScript
Key takeaways
- Working AI-generated code can still rely on outdated implementation patterns.
- Browser features such as
<dialog>deserve consideration before adding a library. - Browser support, accessibility, and design requirements should guide implementation choices.
- Clear project constraints make a coding agent’s decisions easier to review.
A coding agent can deliver a working feature and still leave you maintaining code the browser could have handled. A November 10, 2026 Show HN post about moving agents beyond “~2019 JavaScript” drew 388 points and 123 comments. It raised a question worth bringing into your next code review: why did the agent build it that way?
Knowing a feature doesn’t mean choosing it
The “2019” label is shorthand for a lag in implementation habits. It doesn’t establish that coding models share some universal knowledge cutoff.
An approach that made sense years ago can survive as a default long after the platform changes. An agent might know a newer browser feature exists and still reach for a familiar pattern.
Training examples are one possible influence. The surrounding repository and the wording of your request may also steer the choice. If the existing code consistently solves a problem one way, an agent may follow along.
That makes “Does it know modern JavaScript?” an incomplete review question. Ask what led it to this implementation, given this project’s requirements.
Before you install, check the browser
“Modern web primitives” sounds more exotic than it is. It means capabilities the browser already provides.
Suppose you ask an agent to build a simple modal. It might propose installing a popup library. Depending on the requirements, the HTML <dialog> element could deserve the first look.
Both approaches may produce the feature. The library also brings another dependency to track and update. If the browser’s built-in behavior meets the requirements, that maintenance work may be avoidable.
Native functionality still needs scrutiny. A dialog must work with a keyboard, satisfy accessibility requirements, and fit the project’s design system. Choosing an HTML element doesn’t settle those questions.
The useful standard is straightforward: every added tool should have a job you can explain. A dependency count of zero is no substitute for meeting the requirements.
Give the agent a decision rule
“Use modern JavaScript” leaves a lot open to interpretation. An agent can use current syntax while adding an unnecessary package. A project supporting older browsers may have good reasons to keep an established approach.
More concrete instructions give both the agent and the reviewer something to work with:
Check the project’s supported browsers. First assess whether built-in browser features meet the requirements. If you recommend an external library, explain which requirement makes it necessary.
Now the agent’s rationale becomes reviewable. If it says a native feature cannot support a required interaction, you have a specific claim to check.
This is also a useful test for modern web guidance, including Google Chrome’s modern-web-guidance: does it help the agent make an appropriate choice for the project?
A list of newer features can inform a decision. Browser support, accessibility needs, and design constraints determine whether that decision holds up.
Best practices need to be explicit
The Hacker News discussion included a welcoming reaction and a much broader claim that best practices were effectively dead in the AI era.
That second view is a participant’s judgment. It doesn’t establish that proven engineering principles have stopped being useful.
My reading is more practical: teams need to state their standards clearly when delegating implementation. Which browsers matter? What accessibility behavior is required? Under what circumstances is another dependency justified?
Those answers help reviewers assess the result. They also give agents criteria beyond producing something that runs.
The thread’s popularity doesn’t demonstrate that any particular guidance improves code. That requires examining results under the same requirements: fewer unnecessary dependencies, correct browser behavior, and easier maintenance.
The review should cover both what the code does and why the implementation fits. The next time an agent proposes a library, ask which requirement the browser cannot meet on its own.
Comments
Loading comments...