Learn
An assistant has finished your lines for a year, and next week nothing will
You know the idea. You have shipped harder code than this every week. But the interview editor completes nothing, and your hands stop at a line you have not typed in full for months: the default for a missing key, the sort with two keys, the import.
The weakness
One engineer wrote on Bluesky this week that, about a year after they stopped writing code by hand, their hands simply would not move on interview-style problems, especially when nervous. They were careful to add that they could not tell whether the tools were the cause. That honesty matters: the fix does not depend on the cause.
A DEV essay from the same week put the wider point in one line: "Writing code has become the cheapest part of software development. Understanding it hasn't." In an interview round both are measured at once, by hand, in front of someone. The parts that slip are rarely the algorithm. They are the small lines an editor used to finish for you.
What goes wrong, on one small task
An original example. "Given a list of tags, return the tags that appear at least twice, most frequent first, ties by name."
The idea takes ten seconds. The first line typed by hand is often this:
Map<String, Integer> counts = new HashMap<>();
for (String tag : tags) {
counts.put(tag, counts.get(tag) + 1);
}
On the input ["red"] it throws a NullPointerException. The first time a tag is seen, counts.get(tag) returns null, and adding one to it unboxes null. The Python version fails the same way for the same reason:
counts = {}
for tag in tags:
counts[tag] += 1 # KeyError on the first tag
Every map you fill by hand has a first key, and the first key has no entry yet. Completion usually writes that case for you, which is exactly why it goes missing when nothing completes. A version that handles it, with the tie rule too:
def repeated_tags(tags):
counts = {}
for tag in tags:
counts[tag] = counts.get(tag, 0) + 1
repeated = [t for t, c in counts.items() if c >= 2]
return sorted(repeated, key=lambda t: (-counts[t], t))
On ["red", "blue", "red", "green", "blue", "red"] it returns ["red", "blue"]. On ["red"] it returns [], and so it does on []. On ["b", "a", "b", "a"] the counts tie, and it returns ["a", "b"]. In Java the counting line becomes counts.merge(tag, 1, Integer::sum);.
What to say out loud
Say what a missing key means before you write the loop: "The first time I see a tag there is no entry, so I start it at one. I'll write that case first."
When you do not remember a call, say the behaviour and write the plain version: "I don't remember the exact name of the method that takes a default, so I'll do the check by hand. It's two lines." An interviewer can follow that. Silence while you search your memory looks the same as not knowing the idea.
And before you run anything, name the smallest input you will try: "One tag, then no tags, then a tie."
The exercise
Seven days, starting tomorrow. Switch off completion and suggestions for all of it, or use a plain text editor. Each day has one check; write the result down in one line.
- Monday: the baseline. Solve one small problem you have not seen, aloud, timed, with nothing completing your lines. Write down every place you stopped because of a call or a piece of syntax, not because of the idea. Check: how many stops, and how many were about syntax?
- Tuesday: the first key. From a blank file, write five loops that fill a map: a count, a group of items per key, the first index of each value, the last index, and a sum per key. Check: run each on a one-element input. How many failed on the first key?
- Wednesday: the calls you lean on. Take Monday's stops. Write each library call from memory with its arguments in order, then check the documentation once. Check: how many were right before you looked?
- Thursday: a sort with two keys. Write a sort by count descending, then name ascending, in your interview language, Java or Python. Check: on an input you build with a tie, are the tied items in name order?
- Friday: say each line first. Solve one problem aloud, saying what each line is for before you type it. Check: count the silences longer than ten seconds.
- Saturday: a timed round. Forty minutes, one new problem, recorded, with someone watching or the recorder on, so it feels like the room. Check: minutes from saying the idea to the first run, compared with Monday.
- Sunday: the review. Put Monday's list of stops next to Saturday's. Check: which stops are still there? They become next week's Tuesday.
Where this fits in Gadfly
Gadfly has four rooms: algorithms, system design, behavioral and code review. In the algorithms room you code in Java or Python. Tutor mode asks the next question and never reveals the solution; saying "I'm stuck" gives one more step, up to pseudocode, never finished code, so the lines you type stay yours. Mock mode gives no coaching during the round, and a debrief at the end quoted from the record, not a grade. The plan above works with a plain text editor and a timer instead.
Published 15 November 2026 · A practice plan for the week · all notes · feed
This note was drafted by Gadfly's scout — an automated writing agent run by Dmytro Hryshchenko — and checked by rules against the product's claims registry, not read by a person before it went live. The exercise is original. Spotted a mistake? [email protected].