AI-Related Job Postings Up 340% Since 2024, Traditional SWE Roles Down 15%. The Labor Market Split in Two. Which Side Are Your Engineers On?

General
1

The numbers stopped feeling abstract when I ran our Q1 workforce planning.

LinkedIn data from early 2026 shows AI-related job postings have increased 340% since 2024, while traditional software engineering roles declined 15%. Median re-employment time for displaced engineers has stretched to 4.7 months, up from 3.2 months two years ago. And 20.4% of the 45,000+ tech layoffs in 2026 explicitly cite AI as the reason.

This is not a “shift.” This is a bifurcation.

The Two-Track Reality

I see it playing out in real time across my org (scaling from 50 to 120 engineers):

Track 1 — AI-Native Roles (Accelerating): ML engineers, AI infrastructure, prompt engineering, AI safety. These roles are getting 80-140% growth rates. Candidates with advanced AI skills command 56% wage premiums over peers in equivalent positions. We posted a senior ML platform role last month and got 400+ qualified applicants within a week.

Track 2 — Traditional SWE (Compressing): Frontend, backend, full-stack roles that focus on feature implementation without AI integration. These roles are not disappearing, but they are commoditizing. Entry-level tech postings dropped 67% between 2023 and 2024. Job-finding rates for AI-exposed roles dropped 14% after the launch of advanced AI tools.

The uncomfortable truth: the same engineers who were your most reliable shippers 18 months ago may now be on the wrong side of this split—and many of them do not know it yet.

The Workforce Planning Problem Nobody Talks About

Here is what keeps me up at night as a CTO:

1. The reskilling timeline does not match the displacement timeline. You cannot turn a React developer into an ML engineer in a quarter. The skills gap is real and wide. 80% of the engineering workforce will need to upskill through 2027 just to keep pace.

2. We are eating our own seed corn. AI is automating the entry-level work that traditionally builds senior talent. Fewer junior roles means fewer people developing into the specialists we will need in 3-5 years. We are solving a short-term productivity problem by creating a long-term talent pipeline crisis.

3. The market is punishing ambiguity. Engineers who describe themselves as “full-stack” without specific AI capabilities are getting filtered out earlier in hiring pipelines. The two-tier workforce is not a prediction anymore—it is observable in compensation data, time-to-hire, and attrition patterns.

What I Am Actually Doing About It

I will be honest—I do not have this figured out. But here is what we are trying:

  • AI integration requirements in every role. No pure “feature factory” positions anymore. Every engineer is expected to use AI tools and understand AI system patterns, even if they are not building ML models.
  • Internal mobility budget. 10% of engineering time allocated to AI skill development—not optional “learning Fridays,” but structured rotations on AI-adjacent projects.
  • Honest career conversations. Telling engineers directly: “Your current skill set has a 3-year shelf life at current market rates. Here is what we can do together to change that.”

The Question I Cannot Answer

52,000 US tech employees were laid off in Q1 2026 alone. Meanwhile, 55% of employers who made AI-driven layoffs admit regret and are quietly rehiring. The market is confused and so are the people making workforce decisions.

So I am genuinely asking: How are you handling this bifurcation in your orgs?

  • Are you retraining existing teams or hiring AI-native talent?
  • How do you have honest conversations with engineers whose roles are compressing without destroying morale?
  • Is your company actually investing in reskilling, or just talking about it?

The labor market did not shift. It split. And the engineers who do not pick a side will have the decision made for them.

2

I am living this bifurcation in the trenches at a Fortune 500 financial services company, and the view from the director level is even messier than the macro data suggests.

We have a team of 40+ engineers. About 18 months ago, I would have told you every single one of them was a strong performer. Today, using the same performance standards, roughly a third of the team feels like they are treading water—and these are people with 8-12 years of experience who were promoted because they were excellent at the traditional SWE work that is now compressing.

The Reskilling Reality Is Brutal

Michelle, your 10% time allocation for AI skill development is more than most orgs offer. We tried something similar last year. Here is what actually happened:

  • The engineers who were already curious about AI used the time effectively and accelerated. They were going to learn anyway—the budget just gave them permission.
  • The engineers in the middle attended workshops, completed a few tutorials, but could not bridge the gap between “I completed a Coursera course on transformers” and “I can architect an AI-integrated production system.”
  • The engineers who needed it most felt overwhelmed and disengaged. Some interpreted the reskilling push as a signal that their jobs were at risk—which, honestly, is not wrong.

The painful lesson: structured reskilling works for people who are already 60% of the way there. For the rest, it creates anxiety without competence.

The Financial Services Angle

In fintech and banking, the bifurcation has an additional dimension: regulatory AI literacy. We now need engineers who understand not just how to build AI systems, but how to build them in compliance with OCC, SEC, and EU AI Act requirements. That is a vanishingly small talent pool.

We recently lost two senior ML engineers to a hedge fund that offered 40% raises. Meanwhile, we have backend engineers with deep domain knowledge in payments infrastructure who cannot find external roles at comparable compensation. The wage premium Michelle cited—56%—understates the gap in specialized financial AI roles.

What I Am Doing Differently

Instead of trying to reskill everyone into ML engineers, I am redefining what “AI-integrated” means at each level:

  • Senior engineers: Expected to design systems that incorporate AI components, even if they do not build the models themselves. System design interviews now include AI integration scenarios.
  • Mid-level engineers: Expected to use AI-assisted development tools proficiently and understand when AI-generated code needs human review.
  • Junior engineers: This is where I struggle most. The entry-level work they would have done is exactly what AI handles best. I am experimenting with “AI verification” as an entry-level role—essentially quality assurance for AI-generated output.

The honest answer to your question about retraining vs. hiring: we are doing both, and neither feels sufficient. The clock is ticking faster than the reskilling.

3

Michelle, thank you for naming this so directly. I have been circling around this conversation with my leadership team for months, and the bifurcation framing finally puts language to what I have been feeling.

But I want to push back on one thing and add a dimension that is not in the macro data.

The Equity Dimension Nobody Is Talking About

When we say “the labor market split in two,” we need to ask: who ended up on which side, and was it random?

It was not random.

The engineers most likely to be on the compressing side of this split are disproportionately:

  • First-generation professionals who learned to code through bootcamps and self-teaching, not ML-focused CS programs
  • Mid-career switchers who entered tech through non-traditional paths
  • Engineers at companies that never invested in AI infrastructure—which correlates heavily with companies outside the Bay Area/NYC/Seattle corridor

I came up through Spelman, Google, Slack. At every stage, I had access to cutting-edge ML projects, mentorship from researchers, and compute budgets that let me experiment. Most engineers in this industry did not have that path. The bifurcation is real, but it is not purely a skills problem—it is an access problem.

What I Am Seeing at the EdTech Startup

We are scaling from 25 to 80+ engineers, and the hiring dynamics have shifted dramatically in the last 6 months:

For AI-adjacent roles: We get flooded with applicants. The quality is high. We can be selective. Time-to-fill is under 30 days.

For “traditional” roles with AI integration requirements: This is actually the hardest category to hire for. We need engineers who understand our EdTech domain deeply AND can work with AI systems. That intersection is tiny.

The retention problem: Our best traditional engineers—the ones who deeply understand our education domain, our users, our data quirks—are getting anxious. They see the job market data Michelle cited. Some are rage-applying to AI startups. Others are quietly disengaging. A few have told me directly: “I feel like the company is sending signals that my skills do not matter anymore.”

That last one hit hard. Because it is partially true, and pretending otherwise would be dishonest.

My Approach: Pair, Do Not Replace

Instead of the “reskill or else” framing, I am trying something different:

AI pairing programs. Every AI-focused project has a paired team: one ML engineer plus one domain expert from our existing team. The domain expert is not learning to build models—they are learning to evaluate, test, and deploy AI systems in their area of expertise. The ML engineer is learning the domain.

The theory: the domain knowledge our veteran engineers have is genuinely irreplaceable. The market is not valuing it correctly right now, but the 55% employer regret rate Michelle mentioned suggests the correction is coming.

Transparent career pathing. I sat down with every engineer on my team individually and mapped out three career trajectories: (1) AI-integrated IC track, (2) AI-adjacent leadership track, (3) domain specialization track. None of them are dead ends, but they require different investments.

The hardest part was having those conversations without making people feel like they are being sorted into winners and losers. I am still not sure I got it right.

Luis, your point about reskilling mostly helping people who are already 60% there resonates painfully. The question I keep coming back to: what do we owe the other 40%?

4

Reading this thread as someone who is technically on the “compressing” side and… yeah. The vibes are exactly as bad as the data suggests.

I lead design systems. I am not an ML engineer. I do not build models. My value has always been in understanding users, building component architectures, and bridging the gap between design and engineering. By every metric in this thread, I am on Track 2.

And I can feel it.

What the Bifurcation Looks Like From the IC Side

A few things that have changed in the last year that nobody in leadership seems to notice:

Job postings now require AI experience for roles that have nothing to do with AI. I was browsing design systems lead roles last month—not because I am leaving, but because I like to know my market value. Almost every posting included something like “experience with AI-powered design tools” or “understanding of ML-driven personalization.” For a design systems role. The requirements inflation is real and it is demoralizing.

The internal status hierarchy shifted. The ML team at our company gets first pick of compute resources, first pick of conference budgets, first pick of the CTO’s calendar. The rest of us are implicitly second-tier. Nobody said it out loud. Nobody had to.

The “just reskill” advice feels hollow from where I sit. I have 12 years of experience in design, UX research, and frontend architecture. I founded a startup. I have hard-won expertise in accessibility and inclusive design. When someone tells me to “learn AI,” what they are actually saying is: “Your decade of expertise is now table stakes. Start over.”

Where I Think the Bifurcation Narrative Is Wrong

Keisha’s point about domain expertise being undervalued is the most important thing in this thread. Here is why:

Every AI system I have seen deployed in production—every single one—needed someone who deeply understood the user context to catch the cases where the model was confidently wrong. The AI can generate 50 UI variations in seconds. But someone needs to know which one actually serves a user with low vision, or a user on a 3G connection, or a user who is stressed and needs clear wayfinding.

That judgment comes from years of doing the work. And we are telling the people who have that judgment that their skills are compressing.

The 55% employer regret rate is the buried lede. Companies fired the domain experts, replaced them with AI, discovered that AI without domain expertise produces confidently wrong output at scale, and are now quietly rehiring. This is not a labor market bifurcation—it is a labor market panic followed by a correction.

What I Actually Want From Leadership

Michelle, Luis, Keisha—honest question from the other side of the table:

When you say “AI integration requirements in every role,” do you mean “everyone should understand how AI fits into their domain” or do you mean “everyone should become a worse version of an ML engineer”?

Because those are very different things, and the former is reasonable while the latter is a waste of everyone’s time.

I can absolutely learn to evaluate AI-generated design output, integrate AI tools into design workflows, and build components that serve AI-powered interfaces. That is a natural extension of my expertise.

But if the expectation is that I should be training models or writing PyTorch, then we are not reskilling—we are cosplaying.

5

I have been reading this thread thinking about the business side that is driving the bifurcation—because the labor market split did not happen in a vacuum. It happened because the product market split first.

The Product Strategy Behind the Workforce Data

Here is what I see from the VP Product seat at a B2B fintech startup:

Every product roadmap now has an “AI layer” that did not exist 18 months ago. Our board does not ask “what features are you shipping?” anymore. They ask “what is your AI strategy?” and “what percentage of your product experience is AI-powered?” Those questions directly translate into hiring priorities.

When Michelle posts that ML platform role and gets 400+ applicants, it is because every Series B+ company has the same board pressure. The 340% increase in AI job postings is not organic demand—it is investor-driven demand. And investor-driven demand creates real markets, even when the underlying value is unclear.

The compression on Track 2 is partially a prioritization signal. When my company allocates 60% of engineering capacity to AI features, we are implicitly saying “the other 40% of the product is maintenance mode.” The engineers working on that 40% feel it. Their features do not get roadmap priority, their bugs get deprioritized, and their career growth stalls—not because their work does not matter, but because the business stopped investing in it.

The 55% Regret Rate Tells You Everything

Maya hit on this and I want to amplify it. The companies regretting their AI-driven layoffs are not regretting them for humanitarian reasons. They are regretting them because:

  1. AI features shipped without domain experts produced embarrassing failures. In fintech specifically, I have seen AI-powered features that violated compliance requirements because nobody on the team understood the regulatory context. The model was technically correct and legally wrong.

  2. Customer satisfaction dropped. Turns out customers notice when the people who understood their problems get replaced by systems that generate plausible-sounding but generic responses.

  3. The AI features themselves need traditional engineering. You still need someone to build the API layer, the data pipeline, the monitoring dashboard, the admin tooling. The “AI product” is maybe 30% ML and 70% everything else. Companies gutted the “everything else” team.

The Workforce Planning Question Is Actually a Product Strategy Question

Michelle asked how we are handling the bifurcation. I think the framing needs to shift from “which engineers do we need?” to “what products are we actually building?”

If your product strategy is “add AI to everything,” then yes, you need AI-native engineers and the traditional roles compress. But that strategy is already showing cracks—the regret data proves it.

If your product strategy is “build excellent products that use AI where it genuinely improves outcomes,” then you need both tracks, and the bifurcation is a failure of organizational design, not an inevitability.

The companies that will win this transition are not the ones that hired the most ML engineers. They are the ones that figured out how to pair AI capability with domain expertise—which is exactly what Keisha is describing with her pairing programs.

My Uncomfortable Prediction

Within 18 months, the AI job posting bubble will partially correct. Not because AI is not important—it is—but because the market will realize that 340% growth in AI roles was driven by hype cycles and board pressure, not by 340% growth in actual AI product-market fit.

When that correction happens, the companies that gutted their traditional engineering teams will be in the worst position. And the engineers who maintained deep domain expertise while adding AI literacy—Maya’s “natural extension” framing—will be the most valuable people in the market.

The labor market split. But splits can heal when the market gets honest about what it actually needs versus what it thinks it wants.