The Software Engineering Outlook: 6, 12, 18 Months

Every projection in this article is a scenario, not a forecast. The corpus behind it holds three clean months of data — April, May, and June 2026 — and three points do not make a trend line, let alone an eighteen-month one. What three points can do is fix a starting position precisely, and let us write down, ahead of time, which future observations would confirm or refute each of the stories the field is currently telling about software work. That is the whole design here: not a number for where software hiring lands in 2028, but a set of instruments and thresholds so that by this autumn we will be able to say which way it is going, and why.

Where the coding job stands, precisely

The corpus reads one applicant-tracking network — DirectEmployers, which supplies 93–99% of these postings each month and is used by large U.S. employers. Every claim below is scoped to direct-employer postings from large U.S. employers, not "the tech labor market." Inside it, software engineering is large and clearly resolved: over the primary window (April–June 2026), about 227,000 U.S. software postings resolve to roughly 220,000 distinct vacancy-locations, coded to the federal SOC software-developer family (15-12xx). (Both figures are single-scope — United States, primary window — the same universe every number below is drawn from; they are not the corpus's all-time, all-country totals.) The unit counted is a vacancy-location — one opening at one place, deduplicated against reposting and re-scraping — and every trend below is a share, never a raw count, because raw monthly totals in this corpus are shaped by ingest cadence as much as by hiring.

Three facts fix the starting position, all on the primary window (April–June; February and March are greyed ingest-burn-in months that carry no claim).

Software's share of postings is rising — but read the rise carefully. As a fraction of all SOC-coded postings, software engineering climbs across the primary window — 9.31% in April, 10.36% in May, 11.27% in June. The direction survives the mandatory batch-day jackknife (STANDARDS §13): dropping each of the fifteen highest-volume ingest days one at a time moves the monthly figure only within a narrow band — April 9.31–9.73%, May 10.19–10.41%, June 11.00–11.38% — and June's floor (11.0%) still clears April's ceiling (9.7%), so the climb is not an artifact of any single ingest day (r6_jackknife_swe_share.csv). What the rising share is not, however, is rising software demand in absolute terms: software vacancy-locations actually fell 27.6% from May to June (93,725 → 67,869), and software's slice grew only because all other covered postings fell faster (−34.2%). Decompose the April→June rise of +1.96 points into a within-employer component (each employer's own software intensity moving) and a composition component (the employer mix shifting toward software-heavy posters), and the majority — +1.26 points — is composition; only +0.70 is within-employer (r6_share_decomposition.csv); excluding the five largest posters, the rise shrinks to +0.28 points (7.79% → 8.07%, r6_share_ex_top5.csv). Software is taking a larger slice of a shrinking pie, and much of that is a shift in who is posting, not each employer independently leaning into software. The direction is still not a within-corpus quirk: Indeed's independently constructed software-development postings index has risen about 15% since agentic coding tools launched in early 2025 — while its all-postings index fell 7% over that span — a different universe reaching the same sign (Indeed Hiring Lab, 8 July 2026; see External anchors).

The entry rung is thin. As a share of all software postings, entry-level ones are a small minority: 3.96% in April, 3.44% in May (the figure is a share of all software vacancy-locations, and the seniority tag is itself an inference, so an under-tagging of "entry" cannot be fully distinguished from a genuinely thin rung). This is the number that speaks to the live dispute over whether AI is hollowing out junior hiring (the Stanford "Canaries" finding, 13 November 2025, of a 16% relative employment decline for 22–25-year-olds in the most AI-exposed occupations — and, for software developers specifically, a fall of nearly 20% from the late-2022 peak while older developers in the same firms grew 6–12%) or whether that softness is a rate-driven, cohort-mechanical artifact (the Economic Innovation Group rebuttal). Both anchors are dated in External anchors. Our data does not adjudicate that fight — but it confirms the object of it: in this network, the junior software posting is already a rare thing, well before any further AI pressure.

The descriptions are absorbing AI — mostly by naming it. This is the report's central signal, and it is a signal of transformation rather than disappearance; but its two tiers move very differently, and the honest reading turns on the difference. The share of software postings that mention AI at all climbs fast, from 50.9% in April to 58.0% in June — and the fast-moving number is precisely the boilerplate-exposed one: our own methodology flags the mention tier as "the posting text mentions AI," not "the role requires an AI skill." The share requiring a concrete AI skill as a competency is both lower and much flatter — about 29–31%, rising only ~1.7 points across the whole window. Narrow to Software Developers specifically (SOC 15-1252, the largest sub-code, computed at six-digit grain from the underlying panel), and the competency figure is non-monotonic: AI-skill penetration ran 35.1% → 39.9% → 36.5%, peaking in May and falling in June rather than climbing steadily, while AI mention rose 57.1% → 60.0% → 66.5%. So by June two of every three software-developer ads reference AI and better than one in three demand it as a competency — but the rewriting of the job around the tool is, on this window, far more a rise in AI being named than in AI being required. The job is not vanishing from the corpus; its description is being rewritten, and for now mostly at the surface.

Who is posting, and where

The signal is concentrated, and the concentration matters for reading it. Five employers account for 30.8% of employer-attributable software postings — about 210,000 of the ~220,000 window vacancy-locations carry a named large employer — and twenty account for 55.4%. The rank order is Oracle and Deloitte, then Cognizant, EY, Maximus, GovCIO, Amazon, Accenture, CVS Health and Capital One (Google sits lower, sixteenth). That roster is a fact about who posts software work, not a curated tech leaderboard: two of the ten largest software posters are a federal IT integrator (GovCIO) and a health insurer (CVS Health). Two structural slices carry their own dynamics and are broken out so a shift in either is not misread as the whole market moving:

• Staffing and IT-services firms post about 10.3% of software vacancy-locations (a sensitivity check on the size-band cube puts the figure at 9.7%). Staffing volume flexes ahead of full-time hiring, so this slice is an early-warning gauge, not core demand.
• Federal contractors — Deloitte, Maximus, GovCIO, Leidos, SAIC, General Dynamics IT, CACI and peers — account for roughly 21.6% of software postings on a name-matched estimate; with staffing firms included, the two together are about 31.9%. That 21.6% is a coarse name match, not a clean federal share and not a lower bound: bare-name matching sweeps in the entirety of predominantly-commercial consultancies — Deloitte alone is 40% of this slice, most of its software work non-federal — while other firms are caught only through their federal subsidiaries, so the figure leans high, not low. What it robustly captures is that this slice is governed by the federal hiring freeze — Executive Order 14356, "Ensuring Continued Accountability in Federal Hiring," signed 15 October 2025, which continues the freeze as standing policy on a four-to-one departures-to-hires ratio — a force entirely independent of AI or interest rates, and it must be tracked separately or it will contaminate every other reading.

Geographically the AI signal tracks the map you would expect. AI-skill penetration in computer-and-math postings runs highest in the AI-dense metros — about 57.2% in San Jose–Sunnyvale–Santa Clara and 50.6% in San Francisco–Oakland–Fremont, 43.3% across California, 41.7% in Washington State, 39.1% in New York State — and lower, near 24–28%, in the federal-contractor belt of Virginia and Maryland, where the work is more agency-facing than model-facing.

Three scenarios, and the numbers that separate them

Here is the honest core. Three clean months cannot tell us which of the following is unfolding. But each scenario makes different predictions about specific, monthly-observable numbers in this same corpus — and the monthly snapshot-append pipeline is built to extend every series below. The promise is not a forecast; it is that by roughly this November, three-to-four further clean months in hand, we will be able to say which column the data is tracking, and point to the indicator that settled it.

Scenario 1 — Baseline: role transformation, not disappearance

Assumption: AI is absorbed into the software job rather than eliminating it. Postings hold or grow their share; AI mention keeps climbing while the competency requirement holds (it need not rise every month — 15-1252 skill penetration already dipped in June); the entry rung stays compressed but not collapsing. This is the scenario most consistent with what three months already show, and with the external anchors: BLS projects the software-developer occupation group growing 15% over 2024–2034 (versus 3.1% for all occupations, on about 129,200 openings a year), and Indeed's 8 July 2026 read — pointedly titled "From Destruction to Creation?" — finds U.S. software-development postings up ~15% since early 2025 against a 7% fall in postings overall, with 71% of that rise in senior roles and 37% in roles that name AI in the title. Both are dated third-party anchors; see External anchors.

Scenario 2 — AI-drag: absorption tips into substitution

Assumption: the same absorption we measure is the leading edge of displacement. AI-skill penetration keeps climbing while software's posting share stalls and then falls; the entry rung erodes further; the senior-to-entry ratio widens. This is the Canaries story generalized from payroll to postings, and its early tell is a divergence — penetration up, volume-share down.

Scenario 3 — Rate-relief: a rate-driven rebound

Assumption: the softness was never mainly about AI — it was the fastest tightening cycle in forty years, exactly as EIG argues. If the Federal Reserve eases, postings revive broadly, entry-level recovers, and software's share rises alongside a general recovery rather than by itself. The confound to isolate here is federal: the EO 14356 freeze would keep the contractor slice depressed regardless of rates, so it must be read out separately.

The indicators are not exotic; they are the seven columns this corpus already produces every month, tabulated with their baselines in the accompanying exhibit (r6_scenario_indicators.csv). Three carry a load-bearing caveat. The entry-share and senior-ratio readings depend on seniority tagging that filled reliably (96–98%) in April and May but dropped to 83.9% in June, so the apparent June entry jump to 8.94% is a tagging-composition shift, not a hiring signal — which gates two of the seven indicators, the entry share [IND-2] and the senior-to-entry ratio [IND-5], off entirely until the fill holds above 95% for three consecutive months, so Scenario 2's "juniors squeezed first" test cannot fire on current data. And the central AI indicator carries its own caution: 15-1252 AI-skill penetration has already touched Scenario 1's 40% confirm region (39.9% in May) and fallen back from it (36.5% in June), so — the series being soft-thresholded and non-monotone — no single month's reading of it will settle a scenario either. We will say so on every future reading rather than let a data-quality wobble or a one-month wiggle masquerade as a labor-market turn.

What would tell us we are wrong

A scenario report earns its keep by being falsifiable, so: the baseline is refuted if software's posting share falls below 10% for a full quarter while AI penetration keeps climbing — that pattern is Scenario 2, and we would say so. All three scenarios are refuted, and the instrument itself thrown into doubt, if the AI penetration series reverses without any external event to explain it, which would point at our own regex definitions rather than at hiring. And none of this resolves the Canaries-versus-EIG dispute; it was never built to. It supplies one modality — postings from one large-employer network — to a question that will be answered, if it is, by triangulating payroll, survey, and postings evidence together. Our contribution is to make this modality's monthly verdict legible in advance.

Methodology

Every number above is drawn from a single corpus snapshot dated July 12, 2026, and reproduces from it; the primary builder is analysis/r6-swe-outlook/r6_build.py and the exhibits are the r6_*.csv files it emits, with the mandatory batch-day jackknife (STANDARDS §13) and the April→June share decomposition in the companion script analysis/r6-swe-outlook/r6_jackknife_shiftshare.py (outputs r6_jackknife_swe_share.csv, r6_share_decomposition.csv, r6_share_ex_top5.csv). The full method — flow rather than stock, share of total rather than raw counts, the April–June primary window with February and March greyed, suppression below thirty postings and above a single-employer majority, the single-source scope limit, and the three-tier AI regex definitions reported by tier — is set out on the methodology page, which this article is written to be checked against.

Three points deserve naming here. First, the software universe is the SOC 15-12xx family, whose crosswalk from free-text titles is right about 72% of the time strict (≈84% resolved, ambiguous cases excluded) at minor-group grain, and about 68% strict (≈79% resolved) at six digits (STANDARDS §12) — stated at matched scoring rules so the reader is not comparing a resolved number to a strict one. The one six-digit claim in this article (15-1252 AI penetration) is computed on the largest, best-populated sub-code and cross-read against the two-digit aggregate, but it inherits that six-digit crosswalk error and is stated as an estimate, not a precise rate. Second, the AI penetration figures are regex definitions, disclosed in three tiers (core role, skill required, any mention) precisely because a single number would hide how much of "AI in the posting" is boilerplate versus competency; we report all three and lean on the middle tier. Third, this is one applicant-tracking network, and its composition can shift for pipeline reasons that look like labor-market signal; the source-mix monitor and the seniority-fill gate above exist to catch exactly that.

Limitations

• Three clean months are not a trend; the projections are scenarios, not forecasts. No number in this article is a point estimate of any future value. April–May–June 2026 is the entire primary window. The scenarios exist to be adjudicated by future monthly data, and the article's claim is about which indicators would adjudicate them, not about where software hiring lands.
• Scope is one applicant-tracking network. 93–99% of these postings are DirectEmployers. Every claim is scoped to direct-employer postings from large U.S. employers, not "the software labor market." Independent corpora (Indeed, Lightcast) agree on the direction of the rising-share signal, but this article certifies only its own.
• The entry-level and seniority readings are the least stable series. Seniority tagging filled 96–98% in April–May but only 83.9% in June, so month-over-month seniority trends — including the apparent June entry-share jump — are not reportable until the fill holds ≥95% for three consecutive months. The stable April–May entry figure (~3.4–4.0%) is reported; the June seniority mix is not.
• The six-digit 15-1252 figure inherits crosswalk error, and is non-monotonic. Six-digit SOC precision is about 68% strict (≈79% resolved); the 36.5% June penetration for Software Developers is an estimate on the best-populated sub-code, cross-checked against the two-digit aggregate (~31%, the June SWE-family rate), not a precise rate. The series peaked in May (35.1% → 39.9% → 36.5%) and fell in June, so it is reported as a path, not as a number still climbing.
• AI penetration is a regex definition, and the fast-rising tier is the boilerplate-exposed one. The 29–37% skill-penetration and 58–66% mention figures depend on the definitions in enrichment/ai_flags.py, reported in three tiers so the reader can see the sensitivity; the rapidly rising figure is AI mention (Tier 3, which the methodology flags as "mentions AI," not "requires AI"), while the competency tier is roughly flat at ~30%. About 13% of the 15-1252 skill-tier flags rest on an "experience with AI"-style bare-token match rather than a named tool or technique, placing the 36.5% at the permissive edge of "skill required." A methodology change to those definitions would move the numbers; per the vintage rule, it would ship as a new article, not a silent edit.
• Employer concentration and the federal slice are structural confounds. Five employers are 30.8% of employer-attributable software postings (~210,000 of ~220,000); a single large employer's posting-calendar change can move a monthly share. The federal-contractor slice (~21.6%) is governed by EO 14356, independent of AI and rates, and is tracked separately for exactly that reason. The fed-contractor identification is a coarse name-match heuristic that over-attributes rather than under-counts — bare-name matching sweeps entire commercial consultancies in (Deloitte alone is ~40% of the slice, most of its software work non-federal), while other firms are caught only via their federal subsidiaries; it is an inconsistent, upper-leaning heuristic, not a lower bound.
• External anchors are third-party figures, not corpus measurements. The Indeed +15% / −7% index contrast, the BLS 2024–2034 +15% projection, the Stanford "Canaries" 16% figure, the EIG rebuttal, and the Indeed "From Destruction to Creation?" title are outside-corpus citations; this vintage certifies only its own numbers. Each was re-verified against its primary source on 13 July 2026 and is dated and linked in External anchors below; they corroborate direction only and carry no load in any corpus measurement.
• Staffing share is a sensitivity pair, not a fact about demand. Staffing/IT-services firms post ~10% of these vacancy-locations; that slice flexes ahead of full-time hiring and is reported as an early-warning gauge with its included/excluded variants, not folded into a single demand number.
• No causal claim, and no resolution of the Canaries–EIG–NY Fed dispute. This article describes within-corpus shares and their direction. It does not claim AI caused any movement, and it does not adjudicate whether entry-level software softness is AI-driven or rate-driven; it supplies one postings-based modality to that open question.

External anchors — dated sources

Every figure here is a third-party measurement, not a corpus number; each was verified against its primary source on 13 July 2026, and each corroborates direction only — none carries load in a corpus reading.

• BLS — the +15% projection and its scope. The software developers, quality-assurance analysts, and testers occupation group is projected to grow 15% from 2024 to 2034 — versus 3.1% for all occupations — with about 129,200 openings projected each year, on average, over the decade; software developers alone held about 1.7 million jobs in 2024. U.S. Bureau of Labor Statistics, Occupational Outlook Handbook: Software Developers, Quality Assurance Analysts, and Testers (2024–2034 projections) and Employment Projections — 2024–34 (released 2025–26). https://www.bls.gov/ooh/computer-and-information-technology/software-developers.htm
• Indeed Hiring Lab — "From Destruction to Creation?" U.S. software-development postings have risen almost 15% since Claude Code launched in late February 2025 while overall postings fell 7% over the same span, leaving software-development postings about 27.5% below their pre-pandemic level (Feb 2020 = 100, i.e. an index near 72.5; the exact two-decimal value is a data-portal series figure, not a narrative number). Of the May 2025–May 2026 increase, 71% is senior roles and 37% is roles that name AI in the title. Guillermo Gallacher, "AI and Job Postings: From Destruction to Creation?", Indeed Hiring Lab, 8 July 2026. https://www.hiringlab.org/2026/07/08/ai-and-job-postings-from-destruction-to-creation/
• Stanford Digital Economy Lab — "Canaries," 16% (and ~20% for software). Early-career workers (ages 22–25) in the most AI-exposed occupations show a 16% relative decline in employment after controlling for firm-level shocks; for software developers specifically, 22–25-year-old employment is down nearly 20% from its late-2022 peak while older developers in the same firms grew 6–12%. Erik Brynjolfsson, Bharat Chandar & Ruyu Chen, "Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence," version dated 13 November 2025. https://digitaleconomy.stanford.edu/publication/canaries-in-the-coal-mine-six-facts-about-the-recent-employment-effects-of-artificial-intelligence/
• Economic Innovation Group — the rebuttal. EIG argues the entry-level softness is confounded by interest-rate sensitivity — the most AI-exposed occupations concentrate in the sectors most sensitive to capital costs — and that AI-exposed vacancies began declining in mid-2022, before generative AI's public release, undercutting the AI-timing story. Iscenko & Millet, Economic Innovation Group (January 2026); see also Stanford's reply, "Canaries, Interest Rates, and Timing." https://eig.org/wp-content/uploads/2026/01/TAWP-Iscenko-Millet.pdf · https://digitaleconomy.stanford.edu/news/canaries-interest-rates-and-timinga-more-on-recent-drivers-of-employment-changes-for-young-workers/
• Executive Order 14356. "Ensuring Continued Accountability in Federal Hiring," signed 15 October 2025, continues the federal hiring freeze as standing policy — a four-to-one departures-to-hires replacement ratio and agency Strategic Hiring Committees, with national-security, immigration-enforcement, and public-safety roles exempt. The White House. https://www.whitehouse.gov/presidential-actions/2025/10/ensuring-continued-accountability-in-federal-hiring/