TL;DR
Take-home coding projects no longer show who wrote the code, because an AI assistant can finish most of them in minutes. Here is how to build a junior developer test that measures real ability and still treats candidates fairly.
A junior developer test still works in the ChatGPT era if it measures what a candidate can do in the moment. That means a short, timed coding task scored against hidden test cases and drawn from a randomized pool. Paste and timing signals are recorded for a person to review. An unsupervised take-home project, on its own, no longer shows who wrote the code.
For a small business hiring its first or second developer, that shift matters. A take-home that once filled a weekend can now be finished in minutes with an AI assistant, and the polished result says little about the person who submitted it. This guide explains why take-home tasks lost their signal and what a fair coding assessment for juniors should measure. It shows how to design a test that discourages outside help without pretending to be foolproof, and how to read the results without treating any single number as a verdict.
Why take-home coding tasks no longer prove ability
For years, the take-home project was the default coding test for junior roles. You sent a brief, the candidate built a small app over a few days, and you judged the code. The format felt fair because it was relaxed and realistic, and the person submitting the work had to write it.
That assumption no longer holds. AI coding assistants such as ChatGPT and GitHub Copilot can produce a working to-do app, a REST endpoint or a data-cleaning script from a one-paragraph prompt. Stack Overflow's annual Developer Survey has reported widespread and growing use of AI tools among developers, so these assistants are now ordinary equipment, not a niche trick.
The real problem is missing evidence, not AI use
Working developers use AI tools every day, and a junior you hire will too. The issue with an unsupervised take-home is that it gives you no way to separate three very different candidates:
- One who wrote the code and understands every line
- One who used AI well, then reviewed and fixed what it produced
- One who pasted a prompt and submitted the output unread
All three can hand in the same clean repository. Only the first two can maintain that code on your team, and the take-home alone cannot tell you which candidate you have.
What a take-home still tells you
Take-home tasks are not worthless. A short, paid take-home can still show how a candidate structures a project, writes a README and names things. Its best use now is as a discussion prompt in a later round, where the candidate walks you through the decisions. Used as the main proof of skill, it measures access to tools more than ability.
What should a junior developer test measure?
A junior developer test should measure whether the candidate can read unfamiliar code, write small working functions, find and fix bugs, and explain their reasoning in plain language. Those four skills predict whether a junior can contribute in the first few months far better than whether they can build a complete app alone.
The U.S. Bureau of Labor Statistics describes software developers' work in its Occupational Outlook Handbook as including designing, testing and modifying programs. At junior level, most of that work happens inside code someone else wrote. A practical coding assessment for a junior role usually covers:
- Code reading: predicting what a short function returns, or spotting why it fails
- Small implementation tasks: writing a function that handles normal input and edge cases
- Debugging: fixing a broken snippet and describing the cause
- Framework basics for your stack, such as writing a simple Flask route that validates input and returns JSON
- Written explanation: a few sentences on why they chose an approach
Notice what is missing: long algorithm puzzles that few junior roles use day to day. Keep the test close to the actual job.
Is an AI-proof coding test possible?
No coding test is fully AI-proof, and any product that claims otherwise deserves skepticism. A determined person can use a second device or another person's help. The realistic goal is different: make outside help harder and slower, and make what happened during the test visible so a person can review it fairly.
If you searched for an AI-proof coding test, that reframing is useful. You are not building a fortress. You are designing a coding assessment where genuine skill shows up clearly and where unusual patterns leave evidence for you to discuss with the candidate.
Four design choices that show who can actually write the code
No single feature does the job alone. These four choices work together, and each has a trade-off worth understanding.
Short, timed tasks sized for a junior
A timed coding task limits how long a candidate can spend prompting, copying and adjusting outside output. Keep each problem small enough that a capable junior finishes comfortably, often in 15 to 25 minutes, so time pressure does not become the thing you are measuring. The trade-off is stress: some strong candidates perform worse under a clock. Offer extra time as an accommodation when a candidate asks, as covered in our guide to pre-employment test accommodations.
Hidden test cases
Hidden test cases are automated checks the candidate never sees. The candidate sees a few example inputs and outputs, but the code is scored against additional cases such as an empty list, a negative number, a very long string or a missing field.
Hidden test cases reward understanding over pattern matching. Code that only handles the visible examples, or that hard-codes their outputs, fails the hidden ones. They also allow partial credit, so a candidate who handles most edge cases scores above one who handles none. The trade-off is clarity: write the task description carefully so a failed hidden case reflects a skill gap, not an ambiguous instruction.
Randomized question pools
A randomized question pool gives each candidate a different set of comparable problems. If one candidate's questions are posted online or shared with a friend, the next candidate still faces different ones. Serving questions one at a time also stops a candidate from pasting the entire test into an AI assistant at the start. The trade-off is consistency: every problem in a pool should be similar in difficulty, or scores become harder to compare.
Paste, tab-switch and timing signals
Integrity signals record what happened during the test: pastes, switches away from the tab, and how long each answer took. A 40-line solution pasted in five seconds after a tab switch looks different from code typed over eight minutes with corrections along the way.
These signals are evidence, not proof. A candidate might paste a helper snippet from their own notes, switch tabs because of a notification, or simply type fast. Treat signals as the start of a conversation. Our guide on how to read an Integrity Score fairly explains how to weigh this kind of evidence, and the 2026 guide to detecting AI cheating in online hiring tests covers the wider set of signals.
Coding assessment formats compared
Most small businesses combine two of these formats rather than relying on one.
| Format | What it shows well | Exposure to outside AI help | Candidate time | Your time |
|---|---|---|---|---|
| Unsupervised take-home project | Project structure, documentation, code style | High, with little visibility | Hours to days | High to review |
| Timed online coding test with hidden test cases and signals | Core skill under consistent conditions | Reduced, with signals recorded for review | 30 to 60 minutes | Low, scoring is automated |
| Live code walkthrough or pair session | Reasoning, communication, depth of understanding | Low | 30 to 45 minutes | Moderate, one person per session |
A practical pattern for a junior role is a timed online coding test first, then a short live walkthrough for the strongest few candidates.
How to run a junior developer coding test, step by step
- List the real work. Write down the five tasks a junior will do most in the first three months, such as fixing small bugs, adding an endpoint or writing tests.
- Match test tasks to that list. Choose code reading, debugging and small implementation problems in the language and framework you use. Drop anything the job never requires.
- Set fair time limits. Time each task yourself or with a current developer, then allow roughly double for candidates.
- State the rules up front. Tell candidates the test is timed, whether AI tools are allowed, and which signals are recorded. Clear rules reduce honest confusion.
- Review scores and evidence together. Look at the overall score, which hidden test cases passed, and any integrity signals. Note questions to ask, not conclusions.
- Hold a short walkthrough. Ask shortlisted candidates to explain one of their solutions and change it live, for example adding a new edge case. Someone who wrote or truly understood the code can usually do this easily.
- Let a person decide. Combine the test, the walkthrough and the interview. Never reject anyone on a score, band or automated output alone.
Should candidates be allowed to use AI on a coding test?
That depends on the job. If your team uses AI assistants daily, a fully closed-book test measures only part of the role. If you need to confirm fundamentals, a closed-book section is the clearest way to see them.
A split design often works well for juniors. Keep a short closed-book section for core skills, then add one task that reflects modern work: give the candidate a function generated by an AI assistant that contains a subtle bug, and ask them to find it, fix it and explain the risk. That task tests the judgment a junior needs most in the ChatGPT era, which is knowing when generated code is wrong. Whatever you choose, say it plainly in the instructions so no candidate is penalized for following a rule they misunderstood.
Example: a small agency hiring a junior Flask developer
Consider a hypothetical five-person web agency that builds client dashboards in Python and Flask. The owner needs a junior developer and receives 60 applications, many with impressive portfolio projects.
The owner sets up a 45-minute assessment: a Python fundamentals section, one Flask task to write a route that validates a form field and returns JSON, a debugging problem, and two short written answers about their choices. Each candidate gets different questions from the same pools.
Two results stand out. Candidate A scores near the top, but the evidence timeline shows several large pastes right after tab switches. Candidate B scores a little lower, passes most hidden edge cases, and shows steady typing with corrections. The owner invites both to a 30-minute walkthrough. Candidate A explains the code fluently and adds an edge case in minutes, which suggests the pastes came from personal notes. Candidate B also does well. The owner now has two strong finalists, chosen on evidence rather than on a portfolio that anyone could have generated.
The lesson: the signals did not decide anything. They told the owner which questions to ask.
Keep your coding assessment fair and lawful
In the United States, pre-employment tests fall under federal laws such as Title VII of the Civil Rights Act of 1964 and the Americans with Disabilities Act (ADA). The U.S. Equal Employment Opportunity Commission (EEOC) enforces these laws and explains its view of testing in its guidance on employment tests and selection procedures. In general terms, a test should be related to the job, applied consistently, and adjusted with reasonable accommodations where needed.
The Uniform Guidelines on Employee Selection Procedures describe the four-fifths rule, a rough check for adverse impact on any group. Our explainer on the four-fifths rule for small employers walks through the calculation. For questions about your own hiring process, speak with a qualified employment attorney or HR advisor.
Where a skills testing platform fits
You can build all of this yourself with a code sandbox, a spreadsheet and a lot of review time. A testing platform saves most of that work. HireValid gives small businesses coding and other skills tests in which each candidate gets a different, timed test served one question at a time from the server, with copying blocked and signals such as pastes, tab switches and timing recorded. An Integrity Score from 0 to 100, in the bands High confidence, Review suggested and Serious concerns, summarizes what was observed alongside an evidence timeline, for a person to review.
Nothing automated rejects a candidate: scores and AI-scored written answers inform you, and you decide. For a walkthrough step, an optional spoken AI follow-up interview can be read beside the test, and it never changes a score. The Free plan shows the whole product on 3 candidates in total, and the Starter plan at $49 a month suits an owner hiring for one or two roles.
Frequently asked questions
How long should a coding test for a junior developer be?
Most junior coding tests work well at 30 to 60 minutes in total, split into several short problems. Longer tests tire candidates and discourage strong applicants who have other offers. Time each task yourself, allow roughly double for candidates, and offer extra time as an accommodation when someone asks.
Can a coding test detect ChatGPT use?
No coding test can reliably prove that a candidate used ChatGPT. What a test can do is record signals such as pastes, tab switches and answer timing, then present them for review. Those signals have innocent explanations too, so use them to guide follow-up questions in a walkthrough, never as an accusation or an automatic rejection.
Are take-home coding assignments still worth using?
Take-home assignments still have value as a later-stage discussion prompt, especially if they are short and paid. On their own they no longer show who wrote the code. Pair any take-home with a live walkthrough where the candidate explains decisions and makes a small change on the spot.
Should a junior developer test include algorithm puzzles?
Only if the job truly requires them. Most junior roles involve reading existing code, fixing bugs and adding small features, not inventing complex algorithms. Puzzles unrelated to the work can screen out capable people and are harder to justify as job-related. Base each problem on tasks from the actual role.
Is it legal to give coding tests to job applicants?
Yes, employers in the United States commonly use skills tests, but federal laws such as Title VII and the ADA apply. In general, tests should be job-related, given consistently, and adjusted for reasonable accommodations. The EEOC enforces these laws. For your specific situation, consult a qualified employment attorney or HR advisor.
What score should a junior developer need to pass?
There is no universal passing score for a coding test. Compare candidates against each other and against what the role needs, and look at which hidden test cases they passed rather than the total alone. Use scores to decide who moves forward to a conversation, and make the final hiring decision as a person.
Conclusion
The ChatGPT era did not make coding tests useless. It made unsupervised take-home projects weak evidence. A stronger junior developer test is short and job-related, scored against hidden test cases, drawn from randomized pools, and paired with recorded signals that a person reads carefully. Add a brief walkthrough, decide openly whether AI tools are allowed, and give candidates clear rules and fair accommodations. The result is a hiring process that shows who can actually write and understand the code you need.
To try this approach on your next developer hire, start free with HireValid on 3 candidates, or compare the monthly plans on the pricing page.
Does a low Integrity Score reject a candidate?+
No. Integrity signals may have innocent explanations, including connection issues or accessibility needs. A person should review the evidence and speak with the candidate before deciding.
What will candidates need?+
A browser and a reliable connection. Typing, spreadsheets and code tasks are intended for desktop. If an employer enables camera checks, candidates must receive a clear notice and a route to request an alternative.
Are the scores official qualifications?+
No. These are tools for hiring decisions. English results are CEFR-aligned level estimates, not official certificates. Cognitive scores are not clinical IQ results.