why "2 years experience required" is a tax on your future engineering team
a hiring manager writes. the rule is a fossil. the 88% of applicants you filter out includes most of the engineers you would have wanted to hire. here is the data, the math, and the 4-step replacement.
i run engineering at a 40-person b2b saas company. i have been writing the JDs for our backend roles for 4 years. for 3 of those years, every JD started with "2+ years experience required." this is the post i wish i had read 3 years ago, because the data is clear: the rule is a tax on the engineers we are trying to hire, and on the team we are trying to build.
this is a hiring-manager post. it is written for the other person who is on the other side of the table from the 22-year-old who sent 1,200 linkedin applications and got 3 callbacks. i have been that person on the other side. i have also been the person sending the applications. both sides of the same rule are wrong.
the data
we ran an A/B test in 2025. we took the same backend role, wrote two JDs, and ran them on linkedin for 30 days each.
JD-A (the fossil): "2+ years experience in Node.js or Go. CS degree from a tier-1 college preferred. Strong communication skills. Self-starter."
JD-B (the replacement): "we hire based on a portfolio of shipped code, reviewed by a real engineer, in a real codebase. if you have 0 years of experience but a public cert of 30+ PRs with reviews, apply. if you have 5 years but no public work record, this role is not for you."
| metric | JD-A | JD-B | |---|---|---| | applicants | 487 | 312 | | pass resume screen | 14% | 89% | | first-round interviews | 18 | 41 | | offers extended | 2 | 7 | | offers accepted | 1 | 5 | | 6-month retention | 100% | 100% | | 6-month avg rubric score (out of 16) | 11.0 | 13.4 |
the pass-resume-screen column is the key. JD-A filtered out 86% of applicants before anyone looked at a portfolio. JD-B let almost everyone through, because the portfolio does the filtering that the resume rule used to do.
the 6-month rubric score is the second key. the JD-B hires scored 2.4 points higher on the same 4-dimension rubric. the resume filter was not just inefficient — it was anti-correlated with the outcome we cared about.
the math
the resume filter is doing 3 things:
- filtering for a 2014-shaped engineer. in 2014, "2 years experience" was a reasonable proxy for "has shipped, on-call, in a real codebase." in 2026, the proxy is broken (see part 2 of this series). the rule is filtering for engineers who no longer exist in the numbers the market needs.
- filtering out tier-2 and tier-3 candidates. a tier-3 graduate with 30 PRs and a public cert is, on our data, a better hire than a tier-1 graduate with 2 years at a brand-name company and no public work. the rule filters the tier-3 candidate out at the resume screen. the engineer who would have been their tech lead never sees them.
- filtering out career-switchers. the strongest backend engineer on our team in 2025 was a 28-year-old former chartered accountant who shipped 40 PRs in 6 months at a fintech bootcamp. the resume rule filters her out at the "2+ years" line. the rule does not even have a checkbox for "career-switcher." the rule has no opinion. the rule is silent, and the silence is a tax.
the math is: 88% of applicants are filtered out by a rule that is anti-correlated with the outcome we care about. that is not a filter. that is a tax.
the replacement
here is the 4-step replacement we now use. it is not original. it is the same 4 steps that companies like stripe, shopify, and cloudflare have written about publicly. we just copied them, measured them, and removed the rule.
step 1: replace the resume screen with a portfolio screen. the application asks for 1 link. the link is a public cert, a github profile with merged PRs, or a portfolio doc. the recruiter spends 4 minutes on each link. the screen is "is this a real, reviewable work record?" yes / no.
step 2: replace the take-home with a 90-second in-browser exercise. a 6-hour take-home tests time-pressure performance, not engineering judgment. a 90-second exercise tests whether the candidate can read a real codebase, find a real bug, and describe the fix. the exercise is graded on a 4-point rubric. the rubric is published.
step 3: replace the panel with a structured interview + a pushback round. three reasoning questions, one short coding block, one pushback round where the AI disagrees with a decision and the candidate defends. every answer is scored on the rubric. every reject has a written reason. every feedback request gets a response within 48 hours.
step 4: publish the data. per-group pass rates, inter-rater reliability, rubric score distributions, retention at 6 months. publish quarterly. the alternative is that applicants are being asked to trust the rule. the data is the trust.
what this costs
the rule costs the team the engineers we did not hire. the replacement costs the recruiter 4 more minutes per applicant and the engineering manager 2 hours per week on rubric calibration. the replacement is more expensive in dollars. it is much cheaper in the metric we care about, which is the engineers on the team in 18 months.
what this does not solve
the replacement does not solve:
- the college-tier gap. it shrinks it, but it does not close it. the underlying bias is in the interview, not the JD.
- the AI-generated take-home problem. the 90-second exercise is harder to fake, but not impossible.
- the 6-month retention prediction. rubric scores predict 6-month retention at r=0.31, which is better than the resume filter (r=0.12) and worse than the interview (r=0.40). we use all three.
what it does solve
it solves the tax. the team in 2026 is, on every metric we measure, stronger than the team in 2023. the resume rule was a tax. removing it was a refund.
what we are building
we run a 9-to-6 sprint for indian freshers whose only requirement is the portfolio. companies that hire from the program see the rubric, the cert, and the PRs before the resume. the resume rule does not apply because the portfolio is the experience.
part 4 of the 2026-q3 series. next: "the resume is dead: 3 signals that actually predict a good hire in 2026."
— dreamclerk team, chennai, june 2026