Four hundred and eleven applications since January. Nine screens, two take-homes, one final round that went to somebody with three years of experience applying for a role advertised as “0 to 2”. The top comment on the thread you just read says the jobs are all in Bengaluru and Kraków now and you should have studied nursing.
So: is global outsourcing destroying junior developer jobs? The junior software market really has deteriorated, and anyone telling you otherwise is not looking. But offshoring cannot carry the explanation alone, because it has run at scale for twenty-five years while the junior market fell apart in about thirty months. Something changed on a different clock. Several things did, at once, and offshoring is one of them rather than the one.
Which cause dominates decides whether the right move is to leave the field, change what you build, or change where you stand. I write this from an outsourcing company, which you should factor in.
What the labour numbers count, and what they don’t
Most figures thrown around in these threads are job-postings indexes. Postings are advertisements: not jobs, not hires, not employment. One requisition gets posted on four boards, reposted monthly, left open after the hire, or opened with no budget behind it. A postings index can halve while employment is flat. Employment data is a different instrument, and three series are worth your time.
Industry payrolls. The BLS Current Employment Statistics survey counts payroll jobs by industry. Computer systems design and related services, where much contract and consulting development sits, employed 2,362,700 people in August 2026 against 2,394,000 a year earlier: down about 31,300, or 1.3%. The information sector, home to most software and internet publishers, fell from 2,860,000 to 2,745,000 over the same twelve months, a drop of 4.0%. Professional and business services overall grew by 152,000. Real contraction in tech-heavy industries, then, but not a collapse and not economy-wide. What the series cannot do is follow occupations: a developer at a hospital is counted under health care. It is a proxy for the software job market, not a measurement of it.
Occupational projections. The BLS Occupational Outlook Handbook puts software developers, QA analysts and testers at 1,905,400 jobs in 2025, projects 10% growth to 2035, and estimates about 106,100 openings a year. Median developer pay was $135,980 in May 2025. Read that with the caveat printed on the tin: projections are models, forecasts rather than counts. They also say nothing about the experience distribution, which is the entire thing you care about. An occupation can grow 10% while hiring nobody without commercial experience.
Flows, not levels. This is the series that explains your inbox. In the BLS Job Openings and Labor Turnover Survey for July 2026, total openings were 7.3 million (a 4.4% rate), the hires rate 3.2%, the quits rate 1.9%, layoffs and discharges 1.0%.
Layoffs low. Quits low. Hiring low. This is not a market shedding people, it is a market that has stopped moving. Nobody leaves, so nobody backfills, so the door never opens. A frozen labour market is worse for entrants than a shrinking one, because an incumbent’s job depends on employment levels while a newcomer’s depends entirely on hiring flow. You are not losing a job to an offshore developer. You are waiting for a job to exist. The UK shows the same shape from a different agency: ONS put vacancies at 707,000 for May to July 2026, a level last seen in late 2014.
Five candidate causes, weighed honestly
1. The end of free money. The federal funds target sat at 0 to 0.25% from March 2020 until March 2022, reached 5.25 to 5.50% by July 2023, and stayed there about fourteen months before the first cut in September 2024. Software companies are long-duration bets: you pay engineers now for revenue years away. Move the discount rate from nothing to five and a quarter and those bets reprice, with headcount the largest line in the model. The hiring boom ends within a few quarters of the first hikes. The timing fits.
2. Pandemic over-hiring and the correction. Firms staffed for a demand curve that turned out to be a spike. Correcting an over-hire rarely means mass firing. It means stopping hiring and letting attrition do the work, which is exactly the JOLTS pattern above, and exactly the pattern that erases an entry-level cohort while senior headcount stays put.
3. AI-assisted productivity. Easy to state, hard to measure. In Stack Overflow’s 2025 Developer Survey (49,000+ self-selected respondents, 177 countries), 84% use or plan to use AI tools and 51.6% say the tools improve productivity. The revealing numbers are the negative ones: 46% actively distrust the accuracy of AI output, only 3.1% highly trust it, and the top frustration, cited by 66%, is “AI solutions that are almost right, but not quite.” That describes a tool which raises the output of someone who can spot the almost-right answer and creates review work for someone who cannot. Seniority-biased by construction. Whether it has measurably cut junior hiring is another matter: the strongest claims sit in working papers not yet peer reviewed, and I will not hand you a percentage I could not verify at source. Well-evidenced mechanism, unsettled magnitude.
4. Offshoring and global capability centres. Real, large, continuing. Multinationals have shifted from buying vendor capacity to building their own engineering centres in India, Poland, Mexico and Vietnam, staffed with permanent employees doing product work rather than ticket work. A company that can hire a competent engineer for a third of a US package will keep testing how much of the org chart that covers.
5. Juniors are the first cut in any downturn. Not malice. Arithmetic, which gets its own section below.
The timing problem with the offshoring story
Offshoring is not new. It ran through the 2008 crash, through the 2010s, and through 2021, when American juniors were hired out of six-week bootcamps at six figures. The offshore industry was enormous in 2021. Junior hiring was also enormous in 2021. Both true, same year. A cause roughly constant for two decades cannot explain a change concentrated in thirty months. It can be a background condition that made the fall steeper once something else knocked the market over. It cannot be the trigger.
There is a second problem, and it is a measurement one. The United States has no official statistic counting jobs moved offshore. The nearest thing that ever existed, the BLS Mass Layoff Statistics programme, which surveyed employers on the reasons behind large layoffs, was eliminated in the 2013 sequestration cuts and never restored. So when someone gives you a confident figure for jobs offshored last year, ask where it came from. It will be an estimate from a consultancy or an industry association, both with a commercial interest in the number, in opposite directions. Absence of a statistic is not absence of the phenomenon. It does mean nobody here, me included, is arguing from a count.
Where offshoring genuinely does take junior work
Now the part an outsourcing company usually skips. Offshoring does not remove software jobs at random. It removes the ones specifiable precisely enough to hand to somebody who is not in the room: CRUD endpoints against a documented schema, a screen built to a finished Figma file, test automation for a defined suite, a framework upgrade, bug tickets with clean reproduction steps. Work with a written acceptance criterion and no requirement to hold an opinion.
That is not a coincidence. It is, almost exactly, the list of things a first-year developer used to be given.
I made the same argument on this blog about accounting, where the tasks that travel best offshore are the reconciliations and invoice coding first-year accountants learned on, and the firms that removed them ended up needing experienced reviewers they had stopped manufacturing. Software has the identical structure and a worse version of it, because the tacit knowledge is larger. Nobody designed “fix forty small bugs in a codebase you did not write” as a curriculum. It worked as one anyway. It is how you learn what a suspicious function looks like, why that retry has a three-second delay, and which parts of the system people are quietly afraid of.
Remove it and the ten-year consequence is not fewer juniors, it is fewer seniors in 2036. Firms bridge the gap by hiring experienced engineers from a pool everyone draws on and nobody refills, which works until everyone does it. So yes: offshoring takes junior work, specifically the training-ground kind, and the bill compounds over a decade rather than landing in this year’s payrolls. That is fair to be angry about. It is still not why four hundred applications went unanswered in 2026.
The uncomfortable arithmetic of a first-year developer
A junior developer has never been profitable in year one. Not in 2015, not in 2021. Salary is a real cost from day one, output for the first six to nine months is small and needs review, and that review comes out of your most expensive engineers’ capacity. The return arrives in years two through five, when they can work unsupervised and still cost less than the market rate for what they can now do.
That makes it an investment with a payback measured in years, needing two beliefs: that the company will still want engineers in three years, and that this person will still be there. At zero rates both were cheap and companies made them anyway. With a twelve-month planning horizon the same investment gets declined, and declined first, because it is the only headcount whose entire value sits in the future. That is why juniors go first in every downturn, in every industry, before geography or AI enters the conversation.
What is different this cycle is the other half of the deal. The apprenticeship worked because the trainee’s low-value work was still worth something to the business while they learned. If those tasks can be specified clearly enough to send to a capability centre, or generated in thirty seconds and reviewed, nobody needs them done by a trainee at all. Training used to be a by-product of work that had to happen. Now it must be paid for deliberately, out of a budget line somebody has to defend, and very few companies have created that line. Offshoring and AI are two routes to the same place here, not competing explanations.
What actually helps you now
All of that points one way: stop competing on attributes that can be specified and bought, and get close to the work that cannot be.
Shipped, used software beats credentials, and it is not close. Not a tutorial project. Something with real users, a deploy, an incident you fixed, a decision you can defend. Twelve users is enough. A hiring manager who cannot tell your degree from four hundred others can tell who has run something in production. The signal is not “I can write code”, which is cheap to buy and cheap to fake. It is “give me an ambiguous problem and I come back with something that works and did not break.”
Go where context is the product. The domains holding up best for entrants are those where the hard part is not the code: regulated healthcare and clinical systems, payments infrastructure, industrial and embedded systems, defence and anything cleared, logistics, energy, government. Requirements there cannot be fully written down, so remote specification fails, and there is usually a legal or safety reason a named accountable person must be reachable. Less glamorous than consumer product work, and where the hiring is.
Take the adjacent seat. Support engineering, QA, solutions engineering, implementation, internal tools, data work. Be unromantic about it: BLS projects computer support specialists to decline 3% through 2035 while still averaging about 48,700 openings a year, and QA analysts and testers to grow 6%. Neither is booming. Both are doors that open when the front door will not, and both put you inside the codebase near the people making decisions. Conversion into development from there is common and usually faster than an outside application, and the entry bar is lower: BLS notes support candidates “may qualify with a high school diploma plus relevant information technology (IT) certifications.”
Be in the room. The least fashionable advice here and, I think, the most important. The work that never gets specified and sent elsewhere is the work that emerges from a conversation: the thing the head of ops mentions in a corridor, the customer call where you hear what they actually meant, the incident review where the real requirement surfaces. A junior working fully remotely for a company whose onboarding is a Confluence page is structurally equivalent to an offshore contractor and more expensive. That is uncomfortable to write and I am not going to soften it. Early on, proximity compounds. Later it matters far less.
And when you end up hiring
You will, sooner than you think. Budget training explicitly, as a named line item, because it is no longer a free by-product. Keep a defined share of the ambiguous, context-heavy work onshore and give it to people in their first two years, even when that is slower. And pair juniors with the offshore or AI-assisted output instead of replacing them with it, because reviewing work teaches judgement faster than producing it. The accounting firms that handled this well worked that out first.
The honest answer
The junior software market in 2026 is genuinely bad, and people telling you it is a skill issue are wrong. The anger in that thread is warranted. Global outsourcing is still not destroying junior developer jobs by itself. The best reading of the evidence: a rate shock and an over-hiring correction froze hiring; a frozen market falls hardest on entrants for reasons unrelated to geography; AI is plausibly seniority-biased, well understood in mechanism and unproven in magnitude; and offshoring is a slow, continuous force that has been removing the training-ground tasks for years and presents its real bill in a decade, when the seniors are missing.
The strategies differ, which is why this matters. If it were purely offshoring, the answer would be trade policy and there would be nothing useful for you to do this week. Because it is mostly a frozen market with a structural change underneath, there is: build something people use, go where context cannot be exported, take the adjacent seat, get into the room. None of that is consolation. All of it is actionable, which is more than the thread is offering.
Questions that come up next
Is it offshoring or H-1B? Different mechanisms, constantly conflated. H-1B is a visa programme affecting who fills a US-based role, with published caps and wage rules. Offshoring moves the role out of the country and is subject to none of that. Those are two arguments, and blending them weakens both.
Is a CS degree still worth it, or should I do a bootcamp? Treating this as the central question is the error, because neither is a differentiator now. A degree keeps doors open with regulated and government-adjacent employers that have formal requirements, and it is a visa prerequisite in many countries. Both leave you holding a credential four hundred other applicants hold.
Should I move for work? If you can, yes, specifically to where employers in your chosen domain physically are. It is the most underrated lever an early-career developer has in a frozen market. It is also the one piece of advice here that plenty of people cannot act on for good reasons, which is why it is last rather than first.
Now the disclosure, to weigh as you like. AB7 Solutions is an outsourcing and staffing company, so we are part of the system this article is about. Practically: we recruit and place people into contract, C2C and staff-augmentation roles, which puts us on the hiring side of the door you are trying to get through. We also run data annotation, labelling, HITL, RAG evaluation and AI red teaming work, which hires early-career people and builds the judgement this article argues has become scarce: catching the confidently wrong answer and saying in one sentence why it is wrong. For a conversation about what is actually open rather than a pitch, call +1 321 341 7733, or email ab@ab7solutions.com or director@ab7solutions.com. More at www.ab7solutions.com. If nothing we have fits you, we will say so rather than keep your CV warm.
Sources: US Bureau of Labor Statistics, Employment Situation, Table B-1 (August 2026 preliminary; computer systems design and related services, information, professional and business services); BLS Job Openings and Labor Turnover Survey, July 2026; BLS Occupational Outlook Handbook, Software Developers, Quality Assurance Analysts, and Testers and Computer Support Specialists (May 2025 wage data, 2025 to 2035 projections); BLS Mass Layoff Statistics programme (eliminated under sequestration, 2013); Board of Governors of the Federal Reserve System, Open Market Operations (federal funds target rate history); Office for National Statistics, UK labour market overview (vacancies, May to July 2026); Stack Overflow Developer Survey 2025 (49,000+ self-selected respondents, 177 countries).