Before deleting a utility call, I would look at this example. Both expressions resemble “get the value, otherwise use 3,” but their results differ:

const settings = { retries: null };
 
// _.get uses its default only when the result is undefined.
_.get(settings, 'retries', 3); // null
 
// ?? applies its default to both null and undefined.
settings?.retries ?? 3; // 3

If null means explicitly disabled, replacing the call with ?? changes a business rule. Lodash get uses its default only for undefined; ?? covers both null and undefined. Shorter syntax does not guarantee equivalent behavior.

That difference sets the order of my review. I first find the behavior callers rely on, which tells me how much work the platform can take over. A package's age and downloads help explain its maintenance context. The examples below keep those two questions separate.

Downloads cannot decide whether you should migrate

Request has been deprecated since February 11, 2020. inflight's package metadata says it is unsupported and leaks memory. Moment remains maintained without pursuing new user-facing features. Those are different states. Request's notice, inflight metadata, Moment's policy

All three still had substantial downloads in the saved September 2026 npm API data:

Deprecation and maintenance mode can coexist with substantial downloads

npm Downloads API · 2026-09-01–2026-09-30

All three packages still received substantial downloads in September 2026. These bars do not measure new-project adoption.

million downloads (approx.) / September 2026
  • request (deprecated)59.699
  • moment (maintenance mode)149.03
  • inflight (deprecated)392.617
View npm download data
Deprecation and maintenance mode can coexist with substantial downloads · million downloads (approx.) / September 2026
Package and maintenance statusmillion downloads (approx.) / September 2026
request (deprecated)59.699
moment (maintenance mode)149.03
inflight (deprecated)392.617

Official npm API, September 1–30, 2026, inclusive UTC dates. Values are in millions, rounded to three decimal places; the source retains exact counts. This is a one-month snapshot, not evidence of growth or decline. Downloads are not people or projects; the data cannot distinguish new adoption, transitive dependencies, or automated installations, or quantify their respective contributions.

Source

An old tool might depend indirectly on Request. A build environment without a cached copy can download it without anyone choosing Request again. That explains a possible mechanism, but the totals cannot estimate its share. npm's statistical definitions

The chart can therefore compare maintenance status with download traffic. It cannot estimate new-project adoption or rank every utility for retirement. Deciding whether to remove a tool from your own project requires returning to its call sites. Start with cases where the platform now offers an alternative.

When the platform takes over, one layer can go

node-fetch brought the Fetch API to Node. Node's own fetch has been non-experimental since v21; this article uses Node 24 as its baseline. For basic requests in a fixed runtime, installing that compatibility layer is harder to justify. Node Fetch documentation

Existing code still needs comparison. node-fetch bodies use Node streams, while native Fetch uses Web Streams. Its agent option and Node Fetch's dispatcher are also different interfaces. Proxy connections, cancellation, and stream handling need more than removing an import. node-fetch documentation

Even after those interface differences are resolved, someone still has to own the application's HTTP rules. When fetch receives an HTTP response, a 4xx or 5xx status alone does not reject its Promise. Callers must inspect the response and decide how to classify errors, retry, and avoid duplicate writes. Node Fetch is also based on Undici. Removing the compatibility package leaves both shared request policy and underlying implementation maintenance in place. Fetch standard

File operations show a similar shift. Node 24 includes recursive mkdir, cp, and glob, covering this build-script example:

import { mkdir, cp, glob } from 'node:fs/promises';
 
await mkdir('dist/assets', { recursive: true });
await cp('assets', 'dist/assets', { recursive: true });
 
for await (const file of glob('content/**/*.mdx')) {
  console.log(file);
}

cp has been stable since v22.3 and glob since v24.0. Capability does not establish option-for-option equivalence with third-party tools. Check symlinks, overwriting, exclusions, hidden files, and platform paths. Published packages also need to honor users' minimum Node versions. File-system documentation

I would start with built-ins in small scripts with fixed runtimes and controlled inputs. Replacing a whole legacy application carries a separate migration cost; a new API does not make that cost zero.

A small function can preserve a useful contract

A file script can have tightly controlled inputs; an interface interaction may depend on several different moments. Lodash's debounce, for example, supports leading and trailing calls, maxWait, cancel, and flush. A search panel may need cancellation when closed, immediate execution on submission, and eventual invocation during continuous typing. Like the opening null, these are behaviors to identify before replacing the call. debounce documentation

A ten-line implementation may satisfy simple requirements. If all those behaviors get added back, however, maintenance has merely moved into your repository. Removing a dependency should simplify something beyond package.json.

Date libraries pose the same question. Moment's August 17, 2026 policy retains its mutable API and excludes new user-facing features, while allowing technical maintenance and no longer categorically ruling out a breaking 3.0 release. Policy update

Code expecting a call to mutate the original object can break when a replacement returns a new object: another part of the program may read the old date. Intl.DateTimeFormat handles many display needs, but formatting is not arbitrary input parsing or a policy for appointments across daylight-saving transitions. Date-formatting specification

In the browser, native DOM APIs can select elements, change classes, and attach events. Yet jQuery released 4.0 on January 17, 2026. A use case no longer needing a library does not mean the library stopped being maintained. DOM standard, jQuery 4.0 announcement

These maintenance statements and documents were checked again on October 27, 2026; the chart retains its September period. Current policy and historical traffic have different dates.

AI may remove a dependency—or install it again

AI adds another option for maintaining a local implementation. If the job only wraps a few stable APIs, inputs are clear, and results are easy to check, generating a small replacement may be easier than before. Producing that code does not tell us whether people adopt it, however. The download data here does not measure that substitution effect.

The opposite can happen too. Moment's maintainers attribute some new use to agents' familiarity with established examples. That is their interpretation; aggregate downloads cannot verify AI's causal effect. At project level, at least one rule is practical: familiar generated code still needs a reason for each new dependency.

I would therefore start with the dependency tree: does the tool enter through direct use or an upstream package? Then I would find the functions and semantics its callers need. Where replacement makes sense, migrate one small path and compare null, cancellation, errors, and important edge inputs. An indirect dependency requires work on the upstream package; editing a lockfile does not change its API calls.

Some work remains worth delegating. Time-zone data changes with UTC offsets, boundaries, and daylight-saving rules. Getting an initial implementation right does not finish that maintenance. IANA time-zone database

Back at the opening call, if null means disabled, that rule needs to survive. Another call that only deduplicates simple strings may already be served by native Set. A package can stay in one place and leave another. I want to remove work that is no longer needed, without aiming for an ideal dependency count.