Whitepaper · June 2026
The Talent Cliff.
COBOL, RPG, and AS/400 workforce projections 2026–2030 — and what enterprises must do before their last expert retires.
The retirement curve is not the risk. The combined retirement-and-modernization curve is.
Every legacy-systems shop has known for a decade that its COBOL, RPG, and AS/400 engineers were approaching retirement. The standard institutional response — recruit, train, retain — has not been enough. The pipeline has not refilled at the rate the retirement curve requires. Projecting an aging cohort against conventional retirement-age distributions, the available legacy-skills workforce at the most-exposed enterprises is likely to roughly halve over the second half of this decade.
This whitepaper does not argue that modernization is urgent because the systems are old. It argues that modernization is urgent because the workforce that operates those systems is finite and shrinking. Every quarter that the modernization is deferred is a quarter that the operating workforce gets smaller, scarcer, and more expensive. For some enterprises, within this decade, the realistic available staffing will no longer be sufficient to safely run the legacy estate alongside the modernization itself.
"The risk is not running out of COBOL engineers. The risk is running out of COBOL engineers before the modernization is complete. Those are different deadlines."
What the demographic looks like.
The demographic data on the legacy-systems workforce is not classified — it is available to anyone willing to triangulate from public sources. The picture is consistent.
The aging cohort
The median age of a senior production-COBOL engineer in a tier-one bank is over 55. The cohort is heavily weighted toward the 55-65 range. The fraction of the working population under 35 with production-grade COBOL exposure is, by every survey we have seen, in the low single digits.
The retirement curve
Projecting conventional retirement-age distributions across published cohort estimates, the available production-grade COBOL workforce is likely to roughly halve between 2026 and 2032. The rate is steeper for AS/400 RPG and Adabas/Natural specialists, both narrower-base populations with older medians. The directional pattern is consistent across the sources we have reviewed; the specific timing varies by sub-discipline and region.
The active codebase
Industry analyst estimates place active production COBOL in the tens to hundreds of billions of lines globally, with figures depending heavily on methodology and what counts as "active." The figure has been broadly stable across recent years — enterprises run as much COBOL today as a decade ago because the workloads have grown faster than retirements have removed them.
The replacement gap
The annual production of new COBOL engineers globally — through formal training programs, vendor academies, and on-the-job conversion — is in the low thousands. The retirement loss is materially larger. The gap is structural; no amount of recruiting closes it.
Sources and methodology. Figures in this section reflect a synthesis of public industry analyst estimates (IBM Z workforce reports, Open Mainframe Project surveys, Gartner and Forrester legacy-modernization market analyses), national labor-statistics projections in major mainframe-heavy economies, and Ionate's own field engagement data across 50+ enterprise customers since 2016. Specific figures vary across sources; the structural direction does not. Where ranges are given they reflect cross-source spread rather than precision targets. Available on request: a full source bibliography and Ionate's underlying engagement-level data summary.
You will not recruit your way out of this one.
The standard explanation for the workforce gap — "young engineers prefer modern languages" — is partial but inadequate. The fuller story is that every structural incentive in the engineering career market is pointed away from legacy systems work.
The compensation gap
Senior production-COBOL roles pay well in absolute terms but poorly relative to peer software roles. A 30-year-old engineer choosing between a $250K cloud-native role and a $220K mainframe role chooses the cloud-native role. And the gap widens with seniority.
The career-optionality gap
Skills in modern languages are portable across companies and industries. Skills in mainframe COBOL are portable across a shrinking set of employers. The optionality value of a modern stack is materially higher.
The educational gap
Most CS curricula do not teach mainframe systems. Most engineering bootcamps do not. Students who go into legacy systems work do so by accident, not by design.
The cultural gap
Modern engineering culture rewards visible, fast-cycle delivery. Mainframe systems work is invisible (it sits behind the systems customers see) and slow-cycle (releases on monthly or quarterly cadence). The cultural mismatch with what young engineers seek is significant.
Try this thought experiment: how many of your strongest engineers under 35 would willingly take a five-year posting on the COBOL team? If the honest answer is zero or one, you have answered the question this paper is asking.
Where the cliff arrives first.
The cliff does not arrive everywhere at the same time. Some sectors will feel it before 2028; others have a longer runway. The differentiator is the depth of legacy in the operational core.
| Sector | Legacy Footprint | Exposure | Mitigation Window |
|---|---|---|---|
| Banking (core) | Heavy COBOL/DB2 | Very High | Already starting |
| Insurance | Heavy COBOL/Natural | Very High | By 2028 |
| Government (federal) | Heavy mainframe | High | By 2028 |
| Healthcare (payer) | Heavy COBOL claims engines | High | By 2029 |
| Retail (back-office) | AS/400 RPG concentration | Medium-High | By 2029 |
| Telecom (BSS) | Mixed legacy | Medium | By 2030 |
| Manufacturing (ERP) | AS/400 + COBOL pockets | Medium | By 2030 |
Why banking is first
Banks have the largest active mainframe codebases, the oldest engineer cohorts, and the highest cost of operational error. The combination concentrates the workforce risk earlier than in other sectors.
Why government is slower to act but no less exposed
Federal agencies have larger workforce inertia and slower procurement cycles. The cliff arrives at the same time but their reaction window is narrower.
What to do, in what order.
The exposure analysis maps to an operational playbook. The actions are simple individually; the discipline is in sequencing them while the workforce window is still open.
Document while the engineers are still there
The most-valuable artifact a retiring engineer can produce is a SOTERIA-grade behavioral specification of the systems they own. The window for this is now. Every quarter of delay loses irreplaceable institutional knowledge.
Inventory the workforce risk by system
Map each critical legacy system to the number of qualified engineers who can safely operate it. Identify the single-points-of-failure. Those are the systems to modernize first.
Negotiate continuity retention with key engineers
Retention agreements for critical engineers through modernization completion. Even at expensive terms — are usually cheaper than the production incidents their departure would cause.
Begin agentic modernization on Wave 1
AI-driven modernization is the only approach that can transform at the scale and speed the workforce window demands. Manual or consulting-led programs cannot keep pace.
Stand up a successor operating model
The modernized estate must be operable by the workforce that will exist in 2030, not the workforce that exists today. Plan the operating model for the people who will be there.
How to operate during the transition, and what it costs to delay it.
No multi-year modernization completes overnight. The legacy estate must continue to operate safely while the modernization runs. And every quarter that delay is chosen, the operational math of running the estate gets worse. The two questions — how do we bridge, and what does waiting cost — are the same question seen from two sides.
Bridges
The same agentic AI capabilities that drive transformation can be redirected toward legacy operations: incident triage, change-impact analysis, knowledge surfacing from the original code. The engineers who remain operate with materially more leverage. Knowledge-extraction sprints capture critical institutional knowledge from individual engineers before they retire, producing structured, machine-readable artifacts that survive the engineer's departure. Operations partners — global SIs, specialty mainframe providers — can supplement internal staffing during the bridge period; this is a bridging arrangement, not a perpetual one. Critical engineers can sometimes be retained in advisory or part-time capacity past nominal retirement age, extending the workforce window measurably.
What waiting costs
The case against delay is mathematical. The hourly rate for senior production-COBOL engineers is rising at a pace materially faster than general engineering compensation. The mean time to recovery for legacy-system incidents lengthens as the workforce thins; some incidents are no longer recoverable from in-house expertise and escalate to a shrinking pool of external specialists at premium rates. The modernization itself becomes harder to execute when the engineers who could explain the legacy system are no longer available — SOTERIA scans run on the code regardless, but the human conversation that contextualizes them gets thinner. And every quarter the estate is on legacy is a quarter the business cannot pursue capabilities the modernized estate would enable. The crossover — where modernization becomes unambiguously cheaper than continued legacy operation — has already happened for the most-exposed enterprises and is approaching for the rest.
The numbers do not get better.
One additional discipline matters at board level: the curves compound. Workforce attrition reduces operating capacity in the same year that vendor-rate inflation increases operating cost, in the same year that modernization complexity rises because fewer institutional translators remain available. None of these curves are independent. Each accelerates the others. The cost of acting in 2026 is significantly lower than the cost of acting in 2028; the cost of acting in 2028 is significantly lower than the cost of acting in 2030. The discount rate any board uses to model deferred modernization needs to reflect this compounding, not a linear approximation.
The hardest argument to make to a steering committee is the one that says "the macro numbers are getting worse and we should accept the cost of acting now to avoid the higher cost of acting later." The questionnaire in section 08 is the operational tool we use to make that argument concrete enough to act on.
What the next quarter should look like.
The 2026–2028 playbook is most useful when started immediately. Before scheduling a vendor briefing, run the questionnaire below across your top five legacy-system owners. The answers usually surprise the steering committee more than the macro statistics in this paper do.
| Question | Comfortable answer | Concerning answer |
|---|---|---|
| How many engineers can safely operate this system without supervision? | Three or more | One, or "it depends" |
| What is the median age of those engineers? | Below 50 | 55 or above |
| If the most senior of them retired tomorrow, would on-call coverage hold? | Yes, documented succession | No, or "we would scramble" |
| When was the last successful knowledge-transfer to a junior engineer? | Within the last 18 months | "Not formally" or "years ago" |
| What is the written specification for the system's critical business rules? | Current and complete | Partial, stale, or in the head of one person |
| If we had to hire externally, what's the realistic onboarding time? | Under six months | "Over a year" or "we would not" |
Three or more "concerning" answers on any single system make it a Wave 1 candidate. Two concerning answers across three systems makes the macro problem already operational. Most enterprises will discover both patterns somewhere in their estate.
Ready to face the cliff before it arrives?
We will help you map your workforce exposure system-by-system and prioritize the modernization waves that the demographic curve actually requires. The first conversation is no-cost.