Why Are Software Developers Still In Demand?

I asked Google a question I already knew the answer to. Are software engineers in demand in 2026?

The AI Overview chirped back the kind of thing you’d expect from a college career counselor working off a quota. Yes. Sixty-seven thousand openings. Seventeen percent employment growth projected through 2033. AI engineer postings up 83 percent. ML roles up 63 percent. Top industries: finance, industrial automation, healthcare tech, e-commerce. Mastery of Kubernetes, Docker, CI/CD, and GenAI is now “increasingly necessary.”

Then it asked me, helpfully, whether I was “a student or looking to break into the field.”

If you are, this article is for you. And it is going to ruin your week.

The headlines are right. The pull quote is wrong. Software engineers are absolutely still in demand in 2026 — but not the way the Google AI Overview implies, and not for the people who think the summary is good news. The market is two-tiered, the top tier is bolted shut behind ten years of production scar tissue, and the bottom tier no longer exists. That is the entire story. Everything else is decoration.

The Numbers Everyone Reads

Job listings for software engineers are up roughly 30 percent year over year, with more than 67,000 openings tracked across major employers — the highest in three years. This is happening at the same time as 52,050 tech workers got cut in Q1 2026 alone, the worst quarter since 2023, and roughly half of those cuts were attributed directly to AI. Both things are true.

Senior engineers with current cloud or security experience are closing offers in two to four weeks once they decide to look. KORE1 and Indeed’s Hiring Lab confirm what every working architect already knows: the people who can be trusted with production are getting their pick. Senior AI engineers are clearing $180k–$220k base, with total comp over $350k at the top of the market. The Bureau of Labor Statistics still projects the broader software developer category to grow 15–17 percent through the early 2030s. The sky has not fallen for senior people. The sky is, in fact, doing pretty well for senior people.

This is the part Google quoted. It is not wrong. It is just incomplete in a way that costs careers.

The Numbers Nobody Quotes

New computer-science graduates have a 5.8 percent unemployment rate. That is higher than the general U.S. unemployment rate. Read that sentence twice.

Workers aged 22 to 25 in occupations the research community classifies as “AI-exposed” saw a 13 percent decline in employment in 2025. Big Tech junior hiring is down 25 percent. New graduates make up roughly 7 percent of Big Tech hires — down from 32 percent in 2019. Entry-level software postings have been falling consistently since 2024 and have not recovered. The fresher pipeline at the largest employers in our industry has, functionally, been turned off.

The single most absurd data point I’ll throw at you in this article: per Indeed, hiring an electrician now takes 56 days. Hiring a programmer takes 54. Trades are statistically harder to fill than software jobs. For the first time in fifty years, trade unemployment dropped below college-graduate unemployment. The U.S. Department of Labor responded to that flip by making AI training mandatory in every Registered Apprenticeship program in the country. The federal government has officially conceded which side the wind is blowing on.

If you’re a junior, the demand the Overview is talking about isn’t yours. The demand is for the people who used to be juniors twenty years ago.

Why the Junior On-Ramp Is Gone, and Why It Isn’t Coming Back

Companies hired junior developers for a reason. Somebody had to write the CRUD endpoints, the unit tests, the React components, the README, the CI scripts, the boilerplate Lambda handlers, the validation logic, the migration scripts. That work was the price of training the next senior engineer. It was a paid apprenticeship the industry pretended was a career.

That work didn’t disappear. It just stopped paying $90,000 a year. Cursor does it. Copilot does it. Claude Code does it. SWE-Bench Verified — the benchmark that runs models against real GitHub issues from real open-source projects — went from roughly 30 percent in 2024 to nearly 77 percent in early 2026. For the typical small-bug-or-feature ticket against a clean codebase, the model now wins on time-to-PR and frequently on quality. That is the junior developer’s job description. Convert Jira ticket into syntax. The entire entry-level value proposition.

And it gets worse, because there’s a second front nobody in our industry wants to look at squarely. The small businesses, agencies, and SaaS startups that used to hire a junior dev for $60k–$80k to build a customer portal, an internal tool, or a quick CRUD app? They are not hiring that person anymore. They are using Bubble, Replit Agent, Lovable, Wegic, Make.com, n8n, Airtable, or one of the twenty other no-code AI builders that crossed the “good enough” line in the last eighteen months. A non-technical ops manager can describe an app in plain English in the morning and have a working prototype with a database, auth, and a deployable URL by lunch. The agency model that used to keep a thousand junior devs in WordPress and Wix work is being eaten alive.

Both ends of the junior market — enterprise apprenticeship and SMB freelance — are collapsing simultaneously. Not slowing. Collapsing. The exit ramp is closed at both interchanges.

Why Senior Engineers Have a Ten-Year Runway

The reason a senior engineer with cloud or security chops closes an offer in two weeks is not that they are great at coding. AI is great at coding. It is that senior engineers can do the things AI cannot:

Read 220 billion lines of legacy COBOL that processes 95 percent of U.S. ATM transactions and not break it. Carry the on-call pager when a $40M trading platform goes sideways at 3 a.m. Refactor a monolith without violating the SLA the company has bet quarterly earnings on. Negotiate with a product manager who doesn’t know what they want until you build the wrong thing twice. Sign their name on a HIPAA migration plan and accept legal liability if it goes wrong. Tell the AI no.

The legacy code economy is the part of this conversation almost everyone misses, and it is the entire ballgame for the next decade. Per the GAO and IT-CISQ, 70 percent of banks globally still rely on legacy systems. Forty-three percent of global banking systems run on COBOL. The ten federal legacy systems most in need of modernization cost $337 million annually to operate and consume roughly 80 percent of those agencies’ IT budgets. The average COBOL developer is 55 years old. Ten percent of them retire every year. Sixty percent of organizations using COBOL say their single biggest operational challenge is finding people who can read it.

This is the sleeper market for the rest of the 2020s. Not the AI startup. Not the YC-funded vibe-coded SaaS. The bank you wrote a check from this morning. The insurance company that pays for your kid’s appendix. The Social Security Administration. The state unemployment office that fell over in 2020 because nobody knew how to read its forty-year-old code.

Anthropic recently demonstrated that Claude Code can analyze a substantial COBOL codebase well enough to spook IBM’s stock by 11 percent in a single trading session. Note the verb. Analyze. Not migrate. Not own. Not sign the change order. A senior human engineer still has to drive the modernization, keep the AI honest, and put their name on the document when the trustees of a $40 billion pension fund ask who is responsible. That is your job for the next ten years if you are good. Maybe fifteen if the field of autonomous program reasoning runs into the wall I think it’s about to run into.

AI is a 2x–3x productivity multiplier for a strong engineer. For a weak one, it accelerates the rate at which mistakes reach production. Both halves of that sentence pay senior salaries: the first because the company gets a 3x engineer, the second because the company eventually realizes it needs the senior to clean up the mess the AI-augmented mid-level made. Either way, the line item on the budget is named the same thing. Senior.

The Snake Eats Its Tail

Here is the part that should keep every working senior up at night, and it is the reason I am writing this article.

If juniors do not get hired, juniors do not become mids. If mids do not accumulate scar tissue — the kind that comes from carrying the pager, owning the migration, and shipping the bug that costs the company a customer — they do not become seniors. In ten years, the seniors who currently close offers in two weeks retire. There is nobody behind them. The pipeline has been broken since 2024 and nobody is fixing it because the economics of fixing it do not work for any individual company.

There are two ways this resolves, and I do not love either:

One: the remaining seniors become absurdly valuable for one more cycle, the industry consolidates around an autonomous AI engineering loop in which “software engineer” stops meaning “writes software” and starts meaning “approves software,” and the headcount required to run a Fortune 500 engineering org drops by a factor of ten. Two: the work itself gets eaten by AI faster than the seniors retire, and the transition happens before anyone has time to figure out what to do about it.

In either resolution, the people reading this article — the 30-and-40-something engineers who have already built careers, the senior architects, the staff-plus crowd — are the last generation that will do this job for a living the way we currently understand it. The 22-year-old in their second year of a CS degree right now is training for a profession that will not exist in its current shape when they graduate, and almost certainly will not exist at all by the time they are 35.

This is not a hot take. It is the math.

So What Do You Do

If you are a senior with ten or more years of production experience and a working understanding of distributed systems, security, regulated industries, or legacy modernization: ride this wave. Specialize hard. Learn to direct AI tools well enough that you become the 5x person on a team of three rather than the 1x on a team of fifteen. Take the consulting roles. Take the contract roles. Charge what the market will bear, because the market is now bearing a lot, and bill for the next decade like the runway has a hard end — because it does.

If you are a CS graduate, a junior dev, a bootcamp finisher, an undergrad declaring a major, or a parent paying for any of the above:

Stop. Get out. Now.

I do not say this lightly. I have been on the Microsoft stack for thirty years. I run a software company. I write code every day. My career is software. I love this work. And I am telling you with the authority of someone whose paycheck depends on the opposite advice being correct: the door has closed behind the people already inside, and you should be looking at a different door.

The Door That Is Wide Open

Here is the part of the labor market that the Google AI Overview did not summarize for me, because nobody types “are plumbers in demand in 2026” into Google.

Demand for skilled trades is growing roughly three times faster than demand for white-collar professional roles. That is from Randstad’s analysis of 150 million job postings between 2022 and 2026. Robotics technician postings are up 113 percent. HVAC engineer postings are up 78 percent. Construction roles are up 30 percent. Welders, up 25. Electricians, up 18.

Roughly 530,000 skilled-trade jobs sit unfilled in the United States right now. Ninety-two percent of construction firms cannot find workers. The U.S. needs about 300,000 new electricians over the next decade plus replacements for the 200,000 expected to retire. Forty-one percent of the current construction workforce will retire by 2031. For every 100 young people entering manufacturing, 102 leave. The pipeline is not just broken — it is leaking.

The pay caught up while everybody was looking the other way. Median electrician earnings sit around $62k, with top of the trade clearing $106k. Specialized electricians on AI data-center builds are pulling six figures as a starting position; senior data-center electricians with liquid cooling and fiber-cabling specialties are reportedly clearing $250k–$280k. Mike Rowe says he met three electricians under thirty making $240k–$280k. Plumbers, HVAC techs, and welders all sit in the $51k–$120k band depending on specialty and metro. Unionized journeymen in major cities clear $150k in good years. Industrial maintenance, elevator install, and high-voltage are higher still.

Microsoft’s president called the electrician shortage “the number one problem” slowing data-center expansion. Google committed $15 million to the Electrical Training Alliance. Oracle pushed project timelines back a full year because they couldn’t find enough tradespeople. BlackRock launched a $100 million initiative to train plumbers, electricians, and HVAC technicians. Lowe’s put up $250 million for the same. Meta and CBRE launched a recruiting program for technicians to build out Meta’s data centers. The smart money in the AI buildout is paying for plumbers.

“But What About the Robots”

You will hear the counterargument. Optimus. Figure. Boston Dynamics. The humanoids will replace plumbers too. The trades are no different than software, the argument goes — just on a longer fuse.

They will not. Not for a long time. Not on a fuse short enough to matter to a 22-year-old’s career arc. Here is the actual research.

McKinsey’s 2025 outlook on humanoids in the construction industry estimates large-scale humanoid deployment is “at least a decade away,” and that estimate is for the structured tasks — moving blocks, basic material handling. Nature published a 2025 paper on construction humanoids whose conclusion, paraphrased, is that dexterous foolproof manipulation for intricate tasks like wiring and plumbing remains unsolved at the research level, not just the engineering level. The International Federation of Robotics tracks unstructured commercial humanoid deployments at, effectively, zero. Tesla Optimus has missed every public timeline since 2022. Industry analysts — even bullish ones — put realistic real-world humanoid deployment at five-to-ten years for warehouse-class structured environments, with the giant disclaimer that humanoid timelines have been “five years away” for a literal decade.

A modern LLM can write you a production React component in twelve seconds. A humanoid robot cannot reliably crawl under a sink, identify a corroded P-trap by feel, decide which fitting to swap, discover that the previous owner used the wrong solder, improvise around a pipe that’s been buried in a slab, work around a shutoff valve that won’t close, and not flood your house. That is twenty years of robotics work, minimum, and that is the optimistic version. Real plumbing automation requires general-purpose physical intelligence operating in unstructured environments. We do not have it. We are not close. By the time we do, you will have had a twenty-year career.

A roofer in Vegas in July, on a 12/12 pitch, working around a chimney flashing the previous guy installed wrong? That is unsolvable for current robotics. Not “expensive to solve.” Unsolvable with the present approaches. A mechanic diagnosing a P0420 code on a 2014 Tundra that isn’t actually a catalytic converter problem, because the previous owner installed an aftermarket exhaust that’s reading wrong on the downstream O2 sensor? That isn’t a robotics problem. That is an experience-and-vibes problem. AI doesn’t have vibes. Mechanics do. Roofers do. Plumbers do.

And critically: the trade work that can be automated — prefab framing, factory robotic welding, repetitive pick-and-place — has already been getting automated for thirty years and the industry is still short half a million people. That tells you the elastic demand for skilled, judgment-driven trade work is essentially uncapped. AI eats the structured middle of every industry it touches. The trades’ middle was already eaten in the 1990s. What is left is the part AI is worst at.

Close

The Google AI Overview said yes, software engineers are still in demand. It told the truth and it lied at the same time.

If you are senior, you are in demand for ten more years. Use them. Specialize, charge real money, learn to direct the tools, and assume the runway is finite.

If you are anyone else, the demand the Overview is selling isn’t yours. The demand is for AI-fluent senior architects, regulated-industry specialists, legacy modernization leads, and a small ecosystem of researchers building the AI tooling that will, within your professional lifetime, automate this entire profession down to a single approval button.

Meanwhile the trades that built this country are short half a million workers, the pay is real, the federal government just bet the apprenticeship system on it, and the robots are not coming for them on any timeline that matters to a career.

Software is the last place a 22-year-old should be betting a career in 2026. The trade you cannot outsource to an LLM is the trade you should be learning.

I would tell my own kid the same thing. I am telling you now.

— Gal Ratner

CTO, WhiteStar Labs