The 10,000 Engineering Jobs Are a Barbell, Not a Ladder

The software engineer job market in 2026 isn't shrinking. It's shearing. Demand for engineers who build AI and infrastructure is climbing fast, while demand for routine CRUD and platform-specific work is being repriced toward zero. The old ladder, junior to mid to senior in one continuous band, has snapped into a barbell. Your job is to find out which end you're standing on.

Most career advice right now is built on a comforting average. The U.S. Bureau of Labor Statistics projects software developer employment to grow 15% from 2024 to 2034, with roughly 129,200 openings a year. Sounds healthy. It's also the most misleading number in the conversation, because an average smooths over a violent redistribution happening underneath it.

Why does "15% job growth" feel like a lie if you're job hunting?

Because the growth is real and the pain is also real, and they're happening to different people. Aggregate demand can rise while your specific corner collapses. That's not a contradiction. It's the whole story.

Look at what's actually moving. Software engineer postings on Indeed are down 49% versus early-2020 levels as of July 2025. Android, Java, .NET, iOS, and web developer postings are each down more than 60%. But machine learning engineer postings are up 59% versus pre-pandemic, even after falling 47% from their 2022 peak. Only 19% of tech job titles, 28 out of 149, exceeded their pre-pandemic posting level by 2025. Nearly as many had dropped by over 40%.

The headline says "growth." The data says "two crowds, walking in opposite directions, who happen to share a job title."

What sits at each end of the barbell?

One end is appreciating. The other is depreciating. The bar in the middle, competent generalists writing standard application code, is thinning.

In Praxy's job-postings data, Software Engineer is still the single largest role with roughly 10,700 live postings, and skill demand clusters around cloud, Kubernetes, AWS, and microservices, with AI and infrastructure skills emerging fast. So the title isn't dying. What's dying is the assumption that the title means one thing.

Here's the split, with real evidence on both sides:

	Appreciating end	Depreciating end
Roles	ML/AI engineers, infra and platform, systems programmers, forward-deployed engineers	Routine web dev, platform-specific dev, boilerplate CRUD, manual testing
Demand signal	GenAI postings up 170% Jan 2024 to Jan 2025; 639,000 AI postings added 2023 to 2025	Platform-specific dev postings down 60%+
What it asks for	AI fluency, systems judgment, domain depth, client-facing communication	Code volume, framework familiarity, ticket throughput
Comp signal	Forward-deployed engineer roles post a $173,816 median salary, with equity worth 0.1% to 1.5%	Falling, as the work gets automated

Weak position: A mid-level developer at an IT services firm maintaining a Java banking backend for a European client. Title says software engineer. The work is well-specified, repetitive, and exactly what code-generation tools do well now.

Strong position: A systems engineer building the data-pipeline infrastructure that trains foundation models. Same title. Far higher comp, equity upside, and almost no automation exposure, because the hard part is judgment at the boundary of hardware and software, where correctness isn't negotiable.

Is the ladder really broken, or is hiring just slow?

The ladder is broken at the bottom rung, and that's the part that matters most. The path that turned juniors into seniors is the part that's gone.

The numbers are blunt. Tech postings requiring 5+ years of experience grew from 37% to 42% of all tech postings between Q2 2022 and Q2 2025, while postings targeting 2 to 4 years fell from 46% to 40%. Senior and managerial titles are down 19% from five years prior; standard and junior titles fell 34%. Seniors held up. Juniors got hit nearly twice as hard. And a lot of that squeeze isn't AI eating the work, it's requirement inflation gating entry-level roles behind years of experience.

And it splits by age inside the same job. For IT and software roles with high AI exposure, employment fell 6% for workers aged 22 to 25 while rising 9% for those aged 35 to 49. Same occupation. Opposite direction. The variable isn't the work. It's where you are on the curve.

The leading indicator is the internship pipeline. Tech-specific internship postings are down 30% since 2023. If the apprenticeship layer keeps shrinking, the obvious question follows: in five years, who trains the seniors when there were no junior cohorts to grow them?

Is AI actually doing this, or is it the economy?

AI isn't replacing your senior architect. It's repricing the entry point. That's a different and more specific claim than "AI took the jobs."

Watch where the pressure lands. 70% of hiring managers believe AI can do the work of interns; 57% trust AI's output more than the work of interns or recent grads. Meanwhile computer science graduate unemployment is 6.1% and computer engineering is 7.5%, nearly double the 3.6% national rate. The degree that used to be a guaranteed on-ramp is now a coin flip for a chunk of grads.

But the substitution is messier than the hype. 80% of developers now use AI tools, yet trust in AI accuracy fell from 40% to 29% year over year, and 66% report spending more time debugging flawed AI-generated code. So AI is a high-error productivity tool that still needs a human reviewing it. That's exactly why the value moved from the person who writes the code to the person who knows whether the code is right. The forward-deployed engineer role captures this perfectly: FDE postings exploded 1,165% year over year, comparing January to October 2025 against the same window in 2024. Three years ago, that wasn't a category.

Why is this sharper for Indian engineers?

Because India is exposed at both ends at once. The campus-hiring layer feeds the commodity end that's being automated, and the IT services model that absorbed millions of mid-career developers is unwinding on top of it.

The clearest tell is companies cutting headcount while revenue grows. TCS trimmed its workforce by over 12,000 employees, roughly 2%, despite positive revenue, and Infosys paused fresher onboarding and tightened trainee intake, as AI reshaped service delivery. Read that carefully. The output went up. The bodies and the campus pipeline went down. The engineers who got retained weren't the ones who wrote the most code. They were the ones who understood the client's business well enough to specify what the AI should build.

There's a timing nuance worth holding honestly. About 32% of all currently-listed AI engineering jobs sit in the Bay Area, so the appreciating end is most concentrated in the U.S. hub, and our read is that for engineers in Hyderabad, Pune, or Bangalore the same shift lags by roughly a year or two. That concentration is also why the same engineering role can pay several times more depending on where it's anchored. That lag is not safety. It's a window. The work that compounds, deep systems and domain knowledge, is the same work that protects you on either side of the ocean.

What if the barbell is just a bad year?

Take this seriously, because part of it is true. Some of the decline is a trough, not a structural shift, and pretending otherwise would be its own kind of dishonesty.

Three honest counterpoints. First, the BLS still projects 129,200 annual openings through 2034, so the floor isn't collapse. Second, a chunk of the Indeed posting decline began before late 2022, which means post-pandemic over-hiring explains some of the shape, and a partial recovery in junior hiring is plausible as AI tooling stabilizes. Third, the trust crisis above, AI tools you still have to babysit, means full substitution of junior developers is ahead of the data, not behind it.

So the responsible read isn't "the field is dying." It's that the safe middle thinned and the ends pulled apart. Even if the middle partly refills, the relative reward for moving toward the appreciating end doesn't reverse. The WEF Future of Jobs 2025 puts AI and ML specialists among the top five fastest-growing roles globally and expects 39% of workers' core skills to change by 2030. The direction is steady even where the timing is fuzzy.

So what do I actually do now?

You read your own position honestly, then take one step toward the appreciating end. Not a leap. A step you can repeat.

Name the trade-off plainly first. Moving toward the scarce end costs real time, often six to twelve months of deliberate skill-building on top of a day job, and it usually means a stretch where you're paid for what you can do today while building what pays later. That's the price. If your current role is comfortable and the comfort is what you actually want, that's a legitimate choice. Just make it with your eyes open, knowing the middle is where the automation pressure concentrates.

If you want to move, here's the sequence:

1. Locate yourself. Is your daily work mostly specified-and-repeatable, or mostly judgment-and-ambiguity? The first is the depreciating end. Be honest, not flattering.
2. Pick one adjacent appreciating skill, not a total reinvention. A Java backend engineer moves toward data infrastructure or ML systems far more credibly than toward, say, design. Compounding beats starting over.
3. Use AI to augment, not to coast. The 66% debugging AI's output are building the exact review judgment that's appreciating. Make that your visible skill, and treat AI fluency as baseline literacy now showing up in every job description, not a differentiator.
4. Build one real story, not ten certificates. One production system you owned end to end, with the trade-offs you made, beats a wall of course completions. Your career goes big on one story, not on tenure.

The barbell isn't a verdict. It's a map. It tells you, with unusual clarity for once, which direction to walk.

Want to figure out exactly where you sit on this barbell and the most credible next move from your actual experience? Talk to Praxy on WhatsApp. Tell me your current role and what you've shipped, and we'll map it together.