Anthropic posts a median software engineer compensation of $570,000. That’s not a signing bonus for a VP. That’s the median for engineers building the models that are already writing code at other companies.

Meanwhile, the engineers whose code those models are replacing? Their job market is contracting.

This isn’t a technology story. It’s a business strategy story. And if you’re running a company that depends on software, which in 2026 means every company, you need to understand what this split means for your team and your budget.

The Data Behind the Split

What Stanford and ADP payroll data show

The clearest signal comes from actual payroll data, not job posting counts. Researchers at the Stanford Digital Economy Lab, working with ADP Research, analyzed employment patterns across millions of payroll records. Their finding: employment for young workers (22-25) in software and IT services dropped roughly 20% from its 2022 peak, while employment for workers 35 and older held steady or grew.

Age group	Employment change (2022 peak to 2025)	Interpretation
22-25	~20% decline	Entry-level roles shrinking
26-34	Slight decline	Mid-level pressure building
35+	Stable to slight growth	Senior roles holding or growing

This isn’t a recession pattern. A recession would hit proportionally across age groups. This is something targeted. Entry-level engineering tasks, the ones companies traditionally gave to junior developers, are the same tasks that AI coding tools do well: writing boilerplate and implementing standard features from specs. Straightforward bug fixes fall into the same category.

The global signal

India provides the sharpest example. According to reporting from Rest of World, citing an EY analysis, Indian IT companies cut 20-25% of entry-level engineering roles. India hired only about 75,000 fresh engineers in FY 2024-25, the lowest in over 20 years. In a country that graduates over a million engineers annually, that number is a structural shift, not a hiring cycle.

Microsoft’s 30-to-40 sequence

In April 2025, Satya Nadella told CNBC that “maybe 20-30%” of Microsoft’s code was written by AI, qualifying it as applying to “some projects.” A month later, reporting from Bloomberg via TechCrunch showed that roughly 40% of Microsoft’s Washington state layoffs hit software engineers.

Nadella’s qualifying language matters, he said “maybe” and specified “some projects.” But the sequence reads like cause and effect: AI writes more of the code, company needs fewer of the people who used to write that code.

What the Bifurcation Actually Is

The headline “AI replaces engineers” is wrong. What’s actually happening is a polarization. Some engineering roles are worth more than ever, while others are disappearing. A new category is also forming, one that didn’t exist two years ago.

Role type	Direction	Why
AI systems architect	Rising	Designs what gets automated and how
Senior integration engineer	Stable to rising	Connects legacy systems to AI, requires institutional knowledge
Junior feature developer	Declining	AI tools write standard features from specs
AI output reviewer / validator	Emerging	Human judgment applied to AI-generated code and decisions
Domain expert with AI fluency	Rising	Rare combination, knows the business and the tools

The last row is the one most companies underestimate. Someone who understands your supply chain and compliance requirements, and especially your customer workflows, while also knowing how to direct AI tools, is becoming the most valuable person on the engineering team. Not because they write the best code, but because they know what the code should do.

What This Means for Your Engineering Budget

Here’s the counterintuitive part: the cost of a mediocre engineering team is going up, not down.

If you’re paying for ten engineers who manually write features that AI tools could produce, you’re overpaying. But if you try to cut headcount without investing in the people who can direct and validate AI output, you’ll end up with a codebase that nobody fully understands, generated by AI, reviewed by no one.

Three budget moves we consistently examine with clients:

1. Audit which tasks in your stack are AI-replaceable. Not “which people”, which tasks. A senior engineer might spend 30% of their time on work that AI tools handle well and 70% on architecture decisions that AI can’t touch. You don’t replace that person. You give them better tools and redirect the freed-up hours.

2. Figure out where domain knowledge plus AI fluency is the real value. Your best engineers aren’t the fastest coders. They’re the ones who prevent expensive mistakes because they understand the business context. When AI writes the code, that judgment becomes more valuable, not less.

3. Renegotiate outsourced development contracts. Specifically, contracts that don’t reflect AI productivity gains. If your outsourced team is billing the same hours they billed in 2023 but using GitHub Copilot and Claude for half their output, you’re paying a margin on AI-generated work. Most companies haven’t had this conversation yet.

How to Assess Your Team’s Position

At IQ Source, when we work with a company’s engineering leadership on this, we start with three questions. The answers tell us more than any audit.

What percentage of your team’s output is AI-assisted versus AI-reviewed versus written from scratch? Most teams don’t track this. They should. Not for surveillance, for strategy. If 60% of your codebase is AI-generated and none of it gets meaningful human review, you have a quality risk. If 5% is AI-assisted, you’re leaving productivity on the table.

Then ask: Can your senior engineers describe the failure modes of their AI tools? If the answer is “it sometimes makes mistakes,” that’s not good enough. A senior engineer who uses Claude or Copilot daily should be able to tell you what kinds of mistakes it makes and in which contexts. They should also explain how they catch them. If they can’t, they’re not reviewing, they’re accepting.

The third question is the one where most companies stall: Who owns AI tooling decisions (and do they have authority? In our experience at IQ Source, these decisions get made informally) one developer tries a tool, others follow, nobody evaluates whether it’s the right choice for the organization. The engineer who should own this decision often lacks the budget authority or the mandate to make it.

For more on why this fluency gap matters at the team level, see our piece on the AI fluency gap in engineering teams.

Build vs. Outsource in 2026

The old “build vs. buy” framework doesn’t capture what’s happening. It’s not about building your own software versus buying off-the-shelf. It’s about which engineering capabilities belong inside your company and which don’t.

Keep in-house	Access externally
AI tool selection and evaluation	Specialized model fine-tuning
Architecture decisions that touch business logic	Infrastructure scaling and DevOps
AI output validation and review	One-off implementation projects
Domain-specific workflow design	Security audits and penetration testing
Engineering culture and standards	Vendor integration work

For mid-market companies, the ones we work with most at IQ Source, the sweet spot is typically 2-3 senior engineers who understand both your domain and the AI tooling ecosystem, plus access to specialized implementation capacity when you need it. Not a 30-person development team. Not a fully outsourced model where nobody internal understands the code.

The Fractional CTO model works well here because you need strategic engineering leadership, not more hands on keyboards. The hands can be AI-assisted or externally sourced. The judgment has to be yours.

For a deeper look at the economics of building AI capabilities, see our analysis on enterprise AI economics in 2026.

One Concrete Next Step

If you’re reading this and thinking “I don’t actually know where my team stands on any of this,” you’re not alone. Most engineering leaders we talk to are making decisions based on gut feel about AI’s impact rather than data about their own organization.

We run a two-hour team capability assessment. We look at what your engineers are actually building and how much of that work is AI-assisted. We map skill gaps against where your business is headed, and flag which roles are creating value versus generating code that AI could write. You walk out with a clear picture, not a slide deck or a multi-month roadmap. A picture of where you are and what to do about it.

Schedule a team capability assessment →