The Experience Penalty: What the Data Actually Says About Age Bias in Senior Engineering Hiring

You have eighteen years of experience, four of them at staff level. You have run migrations that touched every service in the company, and you can still explain why the 2019 outage happened in a way that makes a room go quiet. You applied to forty-one roles this quarter. You got six screens and one onsite.

The instinct is to read that as a verdict on your résumé, so you rewrite it. Then as a verdict on your interviewing, so you grind algorithm problems. Both may be true. But there is a third possibility nobody says out loud in advice written for senior engineers, and the research on it is a lot better than the research on résumé fonts.

Here is what the callback studies actually measured, where the effect comes from, what changed when screening moved into software, what the law does not cover, and — most usefully — which of it you can route around.

The Shape of the Penalty, and Where the Evidence Stops

The strongest causal evidence on age and hiring comes from correspondence studies: researchers send otherwise-identical fictitious résumés that differ only in implied age, then count callbacks. The largest one ever run sent more than 40,000 applications in response to 13,371 unique job ads across 12 cities in 11 states, as matched triplets aged 29–31, 49–51, and 64–66.

Here is the honest caveat, and it matters: that study tested administrative assistant, retail sales, janitor, and security roles. It did not test staff engineers. Neither did the Swedish field experiment that randomized ages between 35 and 70 across 6,000+ résumés and found each additional year cut callbacks by about half a percentage point. Anyone claiming a rigorous field experiment on age bias in senior software hiring is selling something.

What we do have is a meta-analysis. Lippens, Vermeiren and Baert pooled nearly every correspondence experiment published between 2005 and 2020 and found a discrimination ratio of 0.58 for older applicants — they received about 58% as many positive callbacks as younger counterparts. That places age alongside the most-studied forms of hiring discrimination, not below them.

The downstream cost shows up in government data. Among long-tenured workers displaced between 2021 and 2023, only 55.3% of those aged 55–64 were reemployed as of January 2024, versus 74.5% of workers aged 25–54. Tracking Americans from age 50 onward, ProPublica and the Urban Institute found 56% are pushed out of a longtime job at least once — and only about one in ten ever earns as much again. Meanwhile 64% of workers age 50+ report seeing or experiencing age discrimination at work, a figure AARP has measured as flat for three straight years.

The Ladder Runs Out Before Your Career Does

Before blaming bias for everything, account for a structural fact that gets misread as bias constantly: engineering ladders are pyramids, and they get narrow fast. Levels.fyi's own leveling standard describes a typical company as having fewer than 30% of engineers at Senior, fewer than 10% at Staff, and usually under 3% at Principal.

That geometry produces the plateau you feel in your comp. In Dice's most recent salary data, tech professionals with 15+ years of experience averaged $133,047 and received a 0.5% raise, while the 3–5 year cohort took nearly 6%. Zoom out and it is bleaker: the average U.S. tech salary of $112,521 in 2024 was worth less than the inflation-adjusted $114,648 earned in 2005.

The demographics compound it. Only 19.1% of professional developers in Stack Overflow's 2025 survey were 45 or older. That is a self-selected sample, not proof that engineers get pushed out — but it does mean that when you walk into a loop at 47, the median person evaluating you has likely never worked for someone your age.

So if your comp has been flat for four years, the first question is not "am I facing bias?" It is "am I at the top of a band?" Those need completely different responses, and we cover the second in the promotion versus job-hop math.

The Bias Is Not About Your Age. It Is About Three Inferences.

This is the most actionable finding in the literature, and almost nobody cites it. Van Borm, Burn and Baert ran a scenario experiment with real recruiters to decompose why older candidates get fewer interviews. Their answer: assumptions about technological skill, flexibility, and trainability account for roughly 41% of the total age penalty.

The effect runs both directions across the funnel. Job ads containing ageist stereotype language independently cut the hiring rate of workers over 40 from about 20% to about 15% — and applications from workers over 40 fell 11.7 percentage points for those ads. Half that damage is self-selection: older candidates read "digital native," "high-energy team," "fast-paced startup culture," and do not apply.

Then there is fit. In a seven-country survey of 1,404 hiring managers, 44% named candidates aged 35–44 as the best fit for their team versus only 15% for candidates 45 and older. The same managers reported that 87% of the 45+ hires they actually made performed as well as or better than younger colleagues. The prediction is wrong and the outcome data is sitting right next to it.

Finally, "overqualified." Research published in Administrative Science Quarterly found across four experiments that candidates who appeared more capable than a role required were penalized because managers inferred lower commitment, not lower ability. Nobody doubts you can do the job. They doubt you will stay.

The 2026 Market Is Worse and Better Than You Think

The macro picture is genuinely bad. As of June 2026, U.S. software development job postings on Indeed remained 27.5% below their pre-pandemic level, even as overall postings had essentially recovered to February 2020 levels. The FRED series tracking that index sat at 75.48 in late July 2026 against a 100 baseline. Roughly 95,667 U.S. tech workers were laid off in 2024 and about 127,000 in 2025.

Now the part that cuts against the doom narrative. In that same Indeed analysis, 71% of the growth in software development postings between May 2025 and May 2026 came from senior roles, with 37% of the gains coming from postings that mention AI in the title. The recovery, such as it is, is concentrated at the senior end. Junior hiring is what stayed frozen.

That reframes the strategy. You are not competing for a shrinking pool of senior roles against a flood of senior candidates. You are competing in the one segment where demand is actually returning — which means the binding constraint is getting seen, not the existence of the job. And it means screening out ghost postings matters more than volume.

One more structural note. Median tenure for workers aged 55–64 is 9.6 years versus 2.7 years for workers aged 25–34. Long tenure feels like loyalty from the inside. From the outside it reads as a thin recent-interview muscle and an untested network — which is exactly the mechanism that turns one layoff into a nine-month search.

The Algorithmic Layer: Screening at a Scale Nobody Audits

The first documented case of age discrimination executed in code settled three years ago. In September 2023 the EEOC entered a consent decree in which iTutorGroup paid $365,000 after the agency alleged its application software was programmed to auto-reject female applicants aged 55+ and male applicants aged 60+, screening out 200+ qualified people. The company admitted no wrongdoing; a consent decree is a settlement, not a finding of liability.

The larger case is still running. In Mobley v. Workday, Judge Rita Lin (N.D. Cal.) granted preliminary certification of a nationwide ADEA collective in May 2025 covering everyone 40 and over denied a recommendation through Workday's platform since September 2020, and in June 2026 granted in part and denied in part Workday's motion to dismiss. Read that posture carefully, because it gets misreported constantly: conditional certification and a partial dismissal ruling test whether allegations are legally sufficient to proceed. Neither is a finding that Workday discriminated. There has been no merits determination.

Regulation has been uneven at best. New York City's Local Law 144 requires an independent bias audit, a public summary, and 10 business days' candidate notice before an automated employment decision tool is used. A Cornell and Data & Society field study found that of 391 employers examined, just 18 had posted a bias audit and 13 the required notice. A State Comptroller audit released in December 2025 found the enforcing agency received two complaints and issued zero penalties in two years.

Elsewhere: Illinois HB 3773 took effect January 1, 2026, making it a civil rights violation to use AI with a discriminatory effect in employment decisions. Colorado's much-discussed AI Act never took effect at all — delayed, then repealed and replaced by SB 26-189 in May 2026 with a narrower regime starting in 2027. What Colorado does have is the Job Application Fairness Act (SB 23-058), in force since July 2024, barring employers from requesting age, date of birth, or graduation dates on an initial application. Connecticut, Delaware, and Oregon have comparable rules.

The operating takeaway is tactical, not legal: a meaningful share of your applications are ranked by systems with no audit, no notice, and no realistic enforcement behind them. Spend your effort on channels that bypass them.

What the Law Protects, and the Three Gaps That Surprise People

The Age Discrimination in Employment Act covers workers and applicants 40 and older at private employers with 20 or more employees. Age charges are not rare — 16,223 were filed with the EEOC in fiscal 2024, roughly one in six of all charges received. But three gaps are worth understanding before you plan around the statute.

Gap one: the causation bar is higher than for race or sex. In Gross v. FBL Financial Services (2009), the Supreme Court held 5–4 that an ADEA plaintiff must prove age was the but-for cause of the decision. Title VII permits a "motivating factor" showing; the ADEA does not, because Congress amended one statute in 1991 and not the other.

Gap two: outside applicants may have no disparate-impact claim at all. In Kleber v. CareFusion (2019), the Seventh Circuit en banc held 8–4 that the ADEA's disparate-impact provision protects employees but not external job applicants; the Supreme Court denied review. The Eleventh Circuit had already read it the same way. So the neutral-policy theory that would most naturally fit an algorithmic screen is unavailable to applicants in those circuits.

Gap three: the odds. Across federal district courts from 1998 to 2006, ADEA plaintiffs won 11.67% of 7,105 adjudicated cases, and when employment-discrimination plaintiffs did win at trial, defendants got 41.10% of those verdicts reversed on appeal versus 8.72% for plaintiff appeals. That dataset predates Gross, so if anything it flatters the current odds. (Babb v. Wilkie softened the standard in 2020 — but only for federal-sector employees.)

None of this is a reason not to file if you have been genuinely wronged; talk to an employment attorney in your state, and note that the economy-wide cost of this is estimated at $850 billion in foregone GDP in 2018 alone. It is a reason not to build your job-search strategy on the assumption that the law will clear your path. It will not, in time to matter.

The Résumé: Cut the Signals, Not the Substance

First, a myth to retire. The ubiquitous advice to show "only the last 10 to 15 years" is convention, not a research finding — there is no published study measuring callback rates for truncated versus full work histories in professional roles. And if you see a blog citing "29% fewer callbacks for résumés implying age 50+," it traces to content marketing, not data. Do not make decisions on invented numbers.

What we do know is which signals get read. In a survey of 1,000 hiring managers, 42% said they consider a candidate's age when reviewing a résumé; among those, 82% inferred it from years of experience and 79% from graduation year. That is a vendor survey on an opt-in panel — weaker evidence than anything above — but it matches what the Colorado and Connecticut statutes were written to stop.

Given a screen that averages 7.4 seconds, make the first third of the page contradict the three inferences driving 41% of the penalty:

• Drop graduation years. Keep the degree and institution. The year is the single most efficient age signal on the page and adds nothing a hiring manager needs.
• Keep the full history, compress the tail. Roles older than ~12 years become a two-line "Earlier experience" block, company and title only. No credibility lost, two-thirds of the page reclaimed.
• Put a dated currency signal in the top third. Not "20 years of experience" but "Led the 2025 migration of 140 services to OpenTelemetry; owns the internal agent-evaluation harness." Recency of tools is the exact inference you are contradicting.
• Cut deprecated stack detail. Your Struts and CoffeeScript line items are pure age signal with zero current relevance.
• Lead with scope, not duration. "Owned reliability for a 400-engineer org" is a level claim. "18 years of experience" is an age claim a machine can bucket.

None of this is deception — every fact stays true and verifiable. It is the same discipline covered in the senior engineer résumé guide: lead with what the reader needs to decide, and stop volunteering the fields they will use as a shortcut.

Disarming "Overqualified" and "Culture Fit" in the Loop

Once you are in the room, two objections do nearly all the damage, and both have a known mechanism you can address directly.

"Overqualified" means they think you will leave. The ASQ research is unambiguous that the penalty runs through inferred commitment, not doubted ability. So answer the real question — not "I'm not overqualified," which argues the wrong point. Instead: "I have done the bigger-scope version of this job and chose to come back to hands-on ownership. I want to be the person who fixes the thing, not the one who writes the doc about who should fix it. That is a durable preference, not a phase." Then give the concrete reason it is durable.

"Culture fit" usually means energy and adaptability. Given that 44%/15% split, expect the fit read whether or not anyone names it. Counter with evidence rather than enthusiasm: name a technology you adopted in the last twelve months and what you got wrong while learning it. Admitting a recent learning curve is the strongest available refutation of "rigid and untrainable," and far more credible than claiming you love change.

Expect the level question too — being routed to a lower band is common enough here to deserve its own playbook, covered in what to do when you get down-leveled.

Route Around the Funnel Entirely

Every effect described above is concentrated in one place: the anonymous top of the funnel, where a stranger or a model reads a document and infers things about you. The single highest-leverage move is to spend less of your time there.

The numbers on this are stark. Applicant-tracking data covering more than 14 million applications found that employee referrals delivered more than 30% of all hires and 45% of internal hires — and that candidates arriving through external sources needed roughly four times as many applications to reach an interview and twice as many interviews to land an offer.

Practically: list every engineer you have worked with who is now at a company you would join. Do not ask them for a job. Ask them what their team is actually struggling with, and whether the req you saw is real. That conversation both bypasses the screen and tells you whether the posting is worth your time. The recruiter screen playbook covers what to do once the conversation converts.

The 90-Day Operating Plan

Days 1–14 — Fix the artifacts. Strip graduation years. Compress pre-2014 roles into one line. Put a dated 2025–2026 technical achievement in the top third of page one. Delete deprecated stack references. Rewrite the summary to lead with scope, not tenure.

Days 15–30 — Build the currency evidence. Pick one genuinely current thing in your domain and go deep enough to have a real opinion, including what it is bad at. Ship something small and public if you can. This is not résumé padding; it is the direct counter to the 41% of the penalty running through assumed staleness.

Days 31–60 — Work the warm channel. Target 15–20 former colleagues at companies you would actually join. Ask about problems, not openings. Convert to referrals where the fit is real. Cold applications continue in the background, capped at roughly 30% of your effort.

Days 61–90 — Rehearse the two objections. Have evidence-backed answers to "why this level" and "how do you stay current" that you have said out loud twenty times. Rehearsed is not scripted, and the difference is audible — see using AI prep without sounding scripted.