Programmer and developer employment 75% below May 2024

Where things stand

Baseline and threshold. I took both numbers from national_M2024_dl.xlsx inside BLS's oesm24nat.zip (https://www.bls.gov/oes/special-requests/oesm24nat.zip, file dated Mar 11, 2025):

• Computer Programmers (15-1251): 109,870
• Software Developers (15-1252): 1,654,440
• B = 1,764,310, so the threshold is 0.25B = 441,078.

Latest release. The May 2025 estimates came out May 15, 2026 (oesm25nat.zip). They show programmers at 92,230 and developers at 1,687,890, for a total of 1,780,120. That is 0.9% above B. Reaching the threshold would take a 75.2% fall from the latest level.

Which releases can count. A release only qualifies if its reference period ends after 2026-09-16. That rules out the May 2026 estimates due in spring 2027. The eligible releases are:

• May 2027 estimates (about April 2028)
• May 2028 estimates (spring 2029)
• May 2029 estimates (spring 2030)
• May 2030 estimates (spring 2031)

Past national release dates were Mar 31 in 2020–2022, Apr 25 2023, Apr 3 2024, Apr 2 2025, and May 15 2026 (delayed by the shutdown). So I put about 25% of each year's hazard in Q1 and most of the rest in Q2.

Base rate

I computed 6-year changes using national OEWS files where the occupation codes are comparable (2012→2018 and 2019→2025):

• Occupations with ≥1M workers: worst decline −16.6% (office clerks, 2019–25).
• Occupations with ≥500k: worst −29% (executive secretaries, 2012–18).
• Occupations with ≥50k: worst about −57% (telemarketers). Computer programmers fell 54% over 2019–25.

No occupation of any meaningful size fell 75%. A 75% drop in a 1.76M-job group would be roughly four to five times the worst decline any large occupation has shown, so the historical reference class gives about 0.

Why the survey reacts slowly

OEWS blends six half-yearly survey panels covering three years (BLS technical notes; BLS's 2019 Monthly Labor Review article on the model-based method). Under that method the two newest panels supply 52% of estimated employment and the two oldest supply 15%. Older staffing patterns are reused when an employer's size has changed by less than 50%.

My own weighted-panel arithmetic gives these implications:

• May 2028 estimates: about 29% of the weight comes from panels collected Nov 2025–Nov 2026, when there was no collapse. A YES from that release is effectively impossible.
• May 2030 estimates (the last in the window): to print ≤25%, actual developer employment would need to fall roughly 85–90% by mid-2029, mostly during 2028.

Current evidence

• Software developer employment in OEWS rose 2% in May 2025.
• Indeed's software-development job postings are up about 15% since February 2025 (CNBC, Oct 11 2026, https://www.cnbc.com/2026/10/11/ai-software-developer-jobs.html).
• TrueUp shows tech job openings rising (Business Insider, Oct 2026).
• Meta's plan to cut teams by up to 60% using AI collapsed (Reuters, Aug 26 2026).
• BLS projects software developers growing +10% from 2025 to 2035.
• The pain so far is concentrated in junior hiring.

Paths to YES

The only real path is a very fast AI takeoff. The AI Futures Project's Aug 2026 update (https://blog.aifutures.org/p/q25-2026-timelines-update-uplift) puts Kokotajlo's "automated coder" median around 2028 and says reality is running at 70–90% of the pace in the AI 2027 scenario; other forecasters there are later.

YES needs three things together:

1. That capability arrives by about early 2028: ~15%.
2. Given that, employers across the whole economy cut nearly all developers within 12–18 months: ~5%.
3. BLS keeps publishing comparable codes: ~85%.

That gives about 0.6%. Small additions for slightly slower but still extreme paths and classification quirks bring it to about 0.8%.

Paths to NO beyond "no collapse"

• The 2028 occupation-code revision is behind schedule (BLS page, Sep 30 2026: the second Federal Register notice has no date yet). If OEWS adopts it within the window, it could bring merged hybrid codes, which would not count.
• BLS funding pressure could disrupt or cut the survey.

Strongest case against my estimate

Capabilities are moving fast (new frontier coding models from Anthropic and OpenAI). Metaculus's labor-automation question on the same OEWS data (broad group 15-1250, 2030 vs 2025; https://www.metaculus.com/questions/42613/) has a community median of −6.6% and puts 4.1% below −50% (40 forecasters, aggregate dated about Oct 10, 2026). Fitting a tail to that suggests roughly 0.4–1.5% below −75%. My 0.8% sits inside that range. I discount the open-bound tail somewhat because of the survey lag and the historical record.

Forecast

Cumulative probability is about 0.01% through 2028, 0.03% by mid-2029, 0.19% by mid-2030, and 0.8% by mid-2031 and end-2031.

Question. By the last day of each quarter, has the latest BLS OEWS release shown combined computer programmer and software developer employment 75% or more below the May 2024 level?

Definitions.

• Employment: the national, cross-industry, all-ownerships employment estimate (TOT_EMP) in the Bureau of Labor Statistics Occupational Employment and Wage Statistics (OEWS) annual national estimates, summed over Computer Programmers (SOC 15-1251) and Software Developers (SOC 15-1252). These are job counts, not vacancies, hours, employment shares, or full-time-equivalent counts. Use OEWS coverage, without adding self-employment or other employment outside its scope.
• May 2024 level (B): the sum of those two rows in BLS's May 2024 national estimates, file national_M2024_dl.xlsx in oesm24nat.zip, using the version available at the question's opening date. This baseline is fixed; later revisions to May 2024 do not change it.
• 75% or more below: a combined published estimate E satisfying E ≤ 0.25 × B. Calculate using the published employment counts without further rounding or a statistical-significance requirement. Neither occupation individually needs to decline by 75%.
• Latest release: the regular OEWS national estimates for the newest reference period available when the qualifying publication is issued. A dated BLS correction to that latest release also counts. A later revision to an older reference period does not count once a newer reference period is available.
• Occupational continuity: renamed or renumbered occupations count if BLS identifies them as equivalent. If either occupation is split, sum all its successor categories only where BLS's official classification or crosswalk establishes that they exhaust the same scope without adding other occupations. A broader merged category does not count unless BLS separately publishes the employment of the original combined scope.

Resolves YES if an eligible BLS publication reports both employment estimates, or their exact combined total, and E ≤ 0.25 × B. The Resolution window and Observation period conventions apply: the release must be published on or after the opening date and its reference period must end after the opening date. Once a qualifying release is published, YES persists for later horizons even if subsequent estimates rebound or are revised upward.

Resolves NO if no qualifying publication exists by the quarter-end. Suppressed, missing, or discontinued rows are not zeros. Declines in only one occupation, wage declines, projections, research estimates, state or industry subsets, and broader occupational aggregates do not independently qualify. Software Quality Assurance Analysts and Testers (SOC 15-1253), web developers, and other computing occupations are excluded. No attribution to AI is required.

Resolving source. BLS only: the annual OEWS National Occupational Employment and Wage Estimates, available through the OEWS Tables page, its national downloadable spreadsheets, and official dated BLS corrections. Prefer the national spreadsheet if displayed tables conflict, unless a BLS correction explicitly supersedes it. This is a narrower resolving-source rule than Public confirmation (A2). Use publication date under Resolution date (A3), not the May reference date; a correction qualifies on its own publication date.