Software Engineering’s Uneven Rebound

U.S. software-development postings have risen from their 2025 low, but the recovery’s composition—and the limits of the evidence—matter more than the headline.

A job market can rebound while still feeling broken. You can see more software-development openings advertised, send out another round of applications, and encounter the same narrow hiring criteria and crowded interview pipelines. That is not necessarily a contradiction: a rising market can remain depressed, and the openings returning may not be the ones you are qualified for. The U.S. data available through September 2026 point toward precisely that distinction—a recovery in one important measure of advertised demand, not a universal revival of traditional software-engineering careers.

For backend, frontend and full-stack engineers, the useful question is therefore not simply whether jobs are “up” or “down.” It is: which measure is moving, from what starting point, and for whom?

The bounce is real—but look at the floor

The clearest evidence of a rebound comes from Indeed’s software-development job-posting index. It sets February 1, 2020 to 100, allowing subsequent posting activity to be compared with a pre-pandemic baseline.

The reported trajectory looks like this:

Sources: Indeed Hiring Lab’s July analysis and August labor-market snapshot.

The arithmetic explains why two apparently competing descriptions can both be true. Moving from 61.1 to 74.4 represents an increase of approximately 22%. But reaching 74.4 still leaves the index roughly a quarter below its February 2020 level.

Indeed also reported an increase of almost 15% between late February 2025 and June 2026. That is a different comparison, not a conflicting result: it uses a different starting point and endpoint.

This is why you should be wary of a percentage without dates attached. “Up 22%” describes the improvement from a trough. It does not describe year-over-year growth, and it certainly does not mean the market has returned to its earlier level.

A rebound tells you which way the market is moving. It does not tell you how healthy the market has become.

There is a second boundary to keep in view. These are relative levels of postings on Indeed, not an absolute count of software vacancies across every employer and job board. The evidence supports “Indeed’s software-development posting index rose from its 2025 low.” It does not, by itself, support “the number of employed software engineers rose 22%.”

Nor does a September snapshot describe the whole of 2026. The year is incomplete.

The returning jobs are not evenly distributed

The rebound becomes more revealing when you examine its composition.

According to Indeed’s analysis, senior roles accounted for 71% of the increase in software-development postings between May 2025 and May 2026. Jobs mentioning AI in the title accounted for 37% of the increase.

Those percentages cannot be added. A senior AI-engineering role can belong to both groups. They also describe contributions to the increase, not shares of every available job.

A separate figure supplies that second perspective: senior-level positions represented 69.3% of software-development postings in the first quarter of 2026, according to the same analysis. Together, the findings suggest that the recovery was heavily oriented toward experience.

If you are an established engineer, that is more encouraging than a continued contraction. But it is not evidence that every conventional specialty recovered at the same pace. The broad software-development category cannot tell you, without further breakdowns, whether junior frontend openings, mid-level backend roles or non-AI full-stack positions all increased.

The distinction is especially consequential at the entry level. LinkedIn’s February 2026 software-engineering talent report found that entry-level software-engineering hiring had not rebounded at the end of 2025.

That result measures hires, not postings. It also covers a different time window. Nevertheless, it provides an important warning against interpreting a recovering aggregate as a recovering first-job market.

It does not establish that AI caused the weakness. LinkedIn explicitly cautioned that the evidence was insufficient for that conclusion.

What an official source can—and cannot—confirm

If you want government corroboration, the first challenge is choosing the right question.

“Are more people employed as software developers?” is not the same question as “Are more software-developer vacancies being advertised?” A labor market can produce different answers to those questions at the same time.

The Bureau of Labor Statistics’ occupational employment estimates put employed software developers at 1,654,440 in May 2024 and 1,687,890 in May 2025. The difference is 33,450, or approximately 2%.

Those estimates support a limited but meaningful conclusion: measured software-developer employment increased over that period. They do not tell you whether advertised openings increased in 2026. They are also survey estimates, not a real-time census or a precise count of individual hires.

The missing government measure is more specific than it might appear. BLS’s monthly Job Openings and Labor Turnover Survey reports openings by industry, but does not collect occupational detail. Its occupation-specific employment surveys, in turn, do not measure openings.

An industry is not an occupation. Software engineers work in banks, retailers, manufacturers and hospitals as well as technology companies. Conversely, an opening at a software company need not be an engineering role. You cannot substitute a technology-industry vacancy series for a software-engineering vacancy series without changing the question.

There is another subtle provenance trap: the St. Louis Federal Reserve publishes an Indeed software-development posting series. That makes the data accessible through an authoritative institution, but it does not create independent confirmation. The underlying source is still Indeed.

An official host is not necessarily an independent source.

For your purposes, the government evidence is useful corroboration against a sweeping claim that software-developer employment was collapsing through May 2025. It is not verification of the precise posting rebound that Indeed reports into 2026.

The independent check complicates the story

A separate view comes from CompTIA, the nonprofit industry association whose Tech Jobs Report analyzes Lightcast job-posting data rather than Indeed’s.

Its September 2026 report counted 43,607 active software-developer/engineer postings in August—2,602 fewer than in July. That implies a monthly decline of about 5.6%.

The report also counted nearly 600,000 active postings across all technology occupations. That larger number should not be presented as software-engineering demand: it covers a much broader occupational set.

The CompTIA result neither independently confirms Indeed’s exact rebound nor necessarily contradicts it. A decline over one month can occur within an upward movement over fifteen months. More fundamentally, a Lightcast-based count and an Indeed index should not be treated as interchangeable without reconciling their coverage and definitions.

What this independent source establishes is substantial advertised demand in August, alongside a latest-month decline. One monthly observation cannot settle the full 2025–2026 trend.

Long-range forecasts require similar discipline. The BLS outlook cited for software developers, quality-assurance analysts and testers projects 10% employment growth from 2025 to 2035. That is a forecast for a combined occupational group over a decade—not an observed increase in current software-engineering postings.

Put the measures side by side, and the picture is coherent but incomplete: Indeed shows a rebound from its trough; BLS estimates show employment growth through May 2025; CompTIA shows many active listings but a monthly decrease; LinkedIn shows persistent entry-level hiring weakness at the end of 2025.

That is triangulation, not unanimous confirmation of one statistic.

What you should make visible now

The practical implication is not that conventional software engineering has disappeared. It is that you should avoid positioning yourself solely around implementation speed.

AI coding tools can accelerate routine work, but generated output still needs to fit requirements, interfaces, security constraints and operational realities. Someone must decide whether a plausible patch is correct, whether a test covers the actual failure mode, and whether a migration can be reversed when production behaves differently from staging.

That mechanism helps explain why experienced judgment can matter more as code becomes easier to produce. It does not prove that AI caused the observed hiring mix.

For an experienced engineer, the strongest evidence you can present is end-to-end ownership: how you took an ambiguous requirement, chose a design, shipped it safely and supported it afterward. Make the trade-offs legible—performance against cost, delivery speed against maintainability, architectural purity against migration risk.

For an early-career engineer, the same principle applies at a smaller scale. A finished feature with meaningful tests, clear failure handling and an explanation of its design decisions says more than a large volume of generated code. If you use AI, make your independent evaluation visible: what you accepted, what you rejected and how you verified the result.

Domain knowledge strengthens that case. Understanding a company’s users, data, regulatory obligations or operational constraints gives you context that a generic implementation demonstration cannot supply. Location, industry and employer type still matter; a bank modernizing a legacy platform is not hiring against the same constraints as an AI-infrastructure startup.

The defensible conclusion is narrower than either a collapse narrative or a comeback headline. U.S. software-development postings on Indeed rose meaningfully from their 2025 low, but remained substantially below their pre-pandemic baseline, with much of the improvement concentrated in senior and AI-titled roles. Other sources illuminate parts of that picture without independently verifying the exact same trend.

For you, the signal is cautiously encouraging. More advertised demand creates possibilities; it does not remove selectivity. The work is still there—but the strongest claim on it is increasingly not merely that you can write the code, but that you can be trusted with what happens after it runs.