Programmer and developer employment 25% below May 2024

Where things stand

Baseline and threshold. These come from BLS national_M2024_dl.xlsx (in oesm24nat.zip, file dated 2025-03-11):

• Computer Programmers (15-1251): 109,870
• Software Developers (15-1252): 1,654,440
• B = 1,764,310, so the YES threshold is ≤ 1,323,232.5.

Latest release. The May 2025 OEWS came out on May 15, 2026 (about six weeks late because of the 2025 shutdown). It shows Programmers 92,230 and Developers 1,687,890, a combined 1,780,120. That is +0.9% against B. Programmers fell 16%, while Developers rose 2%. To reach the threshold, the combined count must fall 25.7% from May 2025.

Which releases count. The question opened on 2026-09-16, and the reference period must end after that date. So the May 2026 estimates (due around spring 2027) do not count. The only qualifying releases inside the window are:

• May 2027 data, published around Mar–May 2028
• May 2028 data, around spring 2029
• May 2029 data, around spring 2030
• May 2030 data, around spring 2031

The required average yearly declines from May 2025 are −13.8% (to 2027), −9.4% (2028), −7.1% (2029) and −5.8% (2030).

Base rate (computed from BLS national files, 1999–2025, oes*nat.zip)

• This combined category. Programmers plus software engineers/developers has never fallen more than 8.6% peak to trough. That was 2000→2002, the dot-com bust (1.17M → 1.07M). In 2008–10 the fall was 4.6%.
• All large occupations. I looked at detailed occupations with at least 500K jobs, within periods using the same classification. A drop of 25% or more happened in 1.4% of 5-year windows (6 of 429) and 2.8% of 6-year windows (9 of 317). Most cases were small clerical jobs or classification artifacts.
• Recent fast declines. No occupation of about 1.8M has dropped this fast. Customer service representatives, a large AI-exposed occupation, fell only 9% from 2023 to 2025.

What the current evidence shows

• Job postings. Indeed's software-development postings index (FRED IHLIDXUSTPSOFTDEVE) bottomed at about 63 in May 2025 and was about 78 in October 2026, a rise of roughly 15%.
• Industry employment (BLS CES). Software publishers went from 653.1K (May 2025) to 659.6K (May 2026). Computer systems design went from 2,410K to 2,370K, and was 2,355K in September 2026. That is erosion of about 1–2% a year, not a collapse.
• BLS projections. Software developers are projected to grow 10% from 2025 to 2035.
• AI-linked layoffs are real: Meta cut 10% in May 2026, plus PayPal and Visa. But Meta dropped its plan for a second, deeper wave (Reuters, Aug 2026).
• How the survey lags. Under BLS's estimation method, the two most recent survey panels supply about 52% of the estimate and the oldest two about 15% (Monthly Labor Review, 2019). The estimate therefore picks up only about 70% of a fresh trend. For the May 2030 release to qualify, actual employment would need to fall roughly 30–36% from 2025.
• Crowd forecasts. The Metaculus Labor Hub median for software developers in 2030 is about −6.6% (Oct 15, 2026), up from −12.4% in April. That question appears to cover a broader, more AI-exposed group. A thin, long-dated Manifold market gives 40% to a fall of more than 15% between 2023 and 2028, which I discount heavily.

Paths, risks, and the forecast

• Path to YES. AI agents become reliable enough in 2027–29 to cause large-scale cuts to developer teams. The broken junior hiring pipeline compounds this, and a possible AI-capex bust or recession could add to it. This mostly affects the May 2029 and May 2030 releases.
• Paths to NO.
  • Rising demand for software and a shift toward senior roles keep employment stable.
  • The survey lags.
  • The 2028 occupation-code revision (final codes due 2027, in use "beginning 2028") could change these codes, or BLS could publish temporary merged codes as it did in 2019–20. Either would make the 2029–30 releases unusable for this question.
  • There is a small risk to BLS funding or to the survey continuing.

Strongest case against my estimate. AI coding ability is advancing extremely fast: governments have delayed frontier model releases, and GitHub reports that one in three pull requests involves an AI agent. AI-focused crowds put fatter tails on outcomes than I do. That pushes me above the roughly 2% base rate, but not to their roughly 8–10%.

Hazards by release. I put the chance of first reaching the threshold at about 0.2% for the May 2027 release, 0.7% for May 2028, 1.7% for May 2029 and 2.5% for May 2030. I add a 0.1% floor for unknowns. For timing, I assume about 28% of each year's release lands by March 31 and about 97% by June 30. That gives a cumulative probability of about 5% by the end of 2031.

Sources: BLS OEWS tables and news release (https://www.bls.gov/oes/tables.htm; https://www.bls.gov/news.release/ocwage.nr0.htm, 2026-05-15); technical note (ocwage.tn.htm); MB3 method (https://www.bls.gov/opub/mlr/2019/article/model-based-estimates-for-the-occupational-employment-statistics-program.htm); BLS projections (https://www.bls.gov/news.release/ecopro.nr0.htm, 2026-08-27); 2028 SOC page (https://www.bls.gov/soc/2028/2028_soc_revision.htm, modified 2026-09-30); Indeed Hiring Lab (2026-07-08); CNBC (2026-10-11); BLS API CES series; Metaculus Labor Hub update (July 2026) and question 42613; TechSpot/Reuters on Meta (Aug 2026).

Question. By the last day of each quarter, has the latest BLS Occupational Employment and Wage Statistics (OEWS) release shown combined computer programmer and software developer employment at least 25% below the May 2024 level?

Definitions.

• Employment: the national, cross-industry, all-ownerships employment estimate (TOT_EMP) for Computer Programmers (2018 SOC 15-1251) plus that for Software Developers (2018 SOC 15-1252). Use the published job counts, not employment shares, projections, vacancies, or counts of unique workers. Software Quality Assurance Analysts and Testers (15-1253) and other computer occupations are excluded. OEWS coverage applies, including its exclusion of self-employed workers.
• May 2024 level (B): the sum of those two TOT_EMP entries in BLS's May 2024 national estimates file, national_M2024_dl.xlsx, distributed in oesm24nat.zip, using the version available at the question's opening date. This baseline is fixed; subsequent baseline revisions do not change it.
• Latest release: the regular annual OEWS national estimates for the most recent May reference period available when BLS publishes a new release or a dated correction to that release. A correction to an older reference period does not qualify once a newer annual release exists. Research estimates do not count.
• At least 25% below: the combined employment estimate is less than or equal to 0.75 × B. Calculate directly from published employment counts without rounding the percentage change or requiring statistical significance.
• Successor classifications: renamed or renumbered occupations count only where BLS's official crosswalk establishes that their combined scope is identical to the two specified occupations. Splits may be summed if exhaustive and nonoverlapping. A merger including other occupations does not qualify unless BLS separately publishes employment for the original combined scope. Missing or suppressed counts are not zero.

Resolves YES if a qualifying release or dated correction published within the Resolution window shows the latest-reference-period combined estimate at or below the threshold. Apply the Observation period convention: the May reference period must end after the opening date; the use of older survey panels in constructing that estimate does not disqualify it. No attribution to AI is required. Once a qualifying publication establishes YES, a later rebound or revision does not reverse it, under the Resolution window convention.

Resolves NO for a quarter if no qualifying publication has met the threshold by that quarter-end. Publications that only repeat or revise pre-opening reference periods, declines in only one occupation without the required combined decline, and non-OEWS estimates do not count. A discontinued series or lack of comparable published counts does not itself establish YES.

Resolving source. BLS only: the annual OEWS National Occupational Employment and Wage Estimates national download files linked from OEWS Tables, the corresponding BLS national online tables, and dated BLS correction notices. Use the national download file if presentations disagree, unless a dated BLS correction explicitly supersedes it. These are a narrower resolving-source requirement than Public confirmation (A2). Apply Resolution date (A3), the Resolution window, and the Observation period conventions; the resolving date is the publication date, not the May reference date.