Most people are asking the wrong software question.
They ask: “Will AI replace software engineers?”
That is too blunt.
The better question is: which layer of software work gets compressed first?
I mapped 125 roles across frontend, backend, DevOps, data, testing, security, AI/ML, product, platform engineering, support, and research using my Replace / Amplify / Emerge framework.
The result is not mass extinction.
It is stratification.
The software economy is still expanding fast. Precedence Research estimates the global software market at $823.92 billion in 2025, growing to $2.47 trillion by 2035. Fortune Business Insights puts SaaS alone at $315.68 billion in 2025.
At the same time, AI has already become part of the default development workflow. GitHub says Copilot now serves 20M+ users across 77,000 enterprises, and its security stack protects more than 90% of Fortune 100 companies. GitHub’s 2025 Octoverse says more than 1.1 million public repositories now use an LLM SDK, and 80% of new developers on GitHub use Copilot in their first week.
But adoption is not the same as trust.
Stack Overflow’s 2025 Developer Survey says 84% of developers are using or planning to use AI tools in development. Yet 46% say they do not trust the accuracy of AI outputs, and 66% say their biggest frustration is dealing with AI solutions that are almost right, but not quite.
That trust gap explains almost everything.
In my role model, not a single software or internet role exceeded 90% automation. Only one role even touched it: L1 technical support.
So what is actually happening?
The career ladder is collapsing from the bottom up.
SignalFire’s 2025 State of Tech Talent report says Big Tech new grads now account for just 7% of hires, with new-grad hiring down more than 50% from pre-pandemic 2019 levels. Startups are even tighter: new grads make up under 6% of hires. The industry’s apprenticeship layer is shrinking first because AI is eating the exact work junior roles used to do.
That is the software story now.
Not “AI replaces all engineers.”
But: AI absorbs the routine layers, raises the bar for entry, and pushes the remaining human work upward toward judgment, ownership, architecture, and accountability.
The Numbers
| Category | # of roles | % of total | Avg replacement rate |
|---|---|---|---|
| Fully automated (>90%) | 0 | 0% | — |
| Heavy AI assistance (60-90%) | 32 | 25.6% | 67% |
| Limited AI assistance (30-60%) | 76 | 60.8% | 41% |
| Irreplaceable (<30%) | 17 | 13.6% | 19% |
Industry-wide AI replacement rate across the 125 roles I analyzed: 44.4% (unweighted average in my role model).
That is high enough to change org charts, flatten junior ladders, and compress execution teams.
It is not high enough to automate away the people who carry deployment risk, system responsibility, product judgment, or technical vision.
REPLACE Tier: The Implementation Layer Is Going First
The first software jobs AI crushes are the ones built on repeatable execution with low accountability.
L1 Technical Support Engineers — 90% automation
This is the clearest AI casualty in the entire industry.
Zendesk says its AI can automate up to 80% of support requests, while reducing resolution times and making agents more productive. Once the work is mostly “understand the request, match it to the right answer, execute a known workflow, and escalate exceptions,” the economics become brutal for human-first support teams.
That does not eliminate support entirely.
It eliminates the old staffing model.
The remaining humans handle escalations, edge cases, and customer situations where the script breaks.
QA and Test Engineers — 75% automation
Testing is one of the most exposed professional layers in software.
That makes sense. Test generation, regression coverage, visual diffs, flaky-test repair, and maintenance of standard test suites all benefit from pattern recognition and automation.
The routine part of QA is becoming software work.
What survives is strategy: deciding what should be tested, what risk matters, what failure is acceptable, and which edge cases still deserve human exploration.
API Developers — 75% automation
If your main value was building standard REST or GraphQL scaffolding, the moat has shrunk dramatically.
AI code assistants now generate endpoints, docs, schemas, request validation, and boilerplate faster than most teams can write tickets for them.
This is why API work feels simultaneously abundant and fragile. There is still demand. But less of it requires a full human workflow from scratch.
Report Developers and BI Builders — 75% automation
Software organizations spent years building layers of internal reporting work that were always more mechanical than strategic.
That layer is now being compressed hard.
Natural-language BI tools, Copilot-style analytics interfaces, and auto-generated dashboards are turning what used to be specialized reporting labor into a commodity interface problem.
The analyst or engineer does not disappear entirely.
But the role shifts from “build the report” to “decide what the report should mean.”
Frontend Engineers — 70% automation
Frontend is still real engineering. But the low- to mid-complexity layer is under direct attack.
Vercel says v0 converts text and image prompts into React UIs and streams component-based interfaces. In its WPP partnership, Vercel says AI tools like v0 and Cursor are reducing ideation and iteration cycles from weeks to hours.
That matters because a huge percentage of frontend work historically lived in the zone between design interpretation and implementation cleanup.
That zone is exactly where generative tools are strongest.
Complex state, accessibility, performance, and system design still need people.
But “turn the mock into a working UI” no longer guarantees a durable moat.
AMPLIFY Tier: Fewer Engineers, More Ownership
This is where the lazy narrative falls apart.
AI does not just delete engineering roles.
It increases the span of control for the engineers who remain.
Backend Engineers — 65% automation
The backend layer is not disappearing.
It is being reweighted.
AI is exceptionally good at CRUD scaffolding, routine endpoint work, database wiring, framework migration, and standard integrations. That is why the middle of backend engineering feels squeezed.
But as soon as the work involves tricky domain logic, distributed consistency, reliability tradeoffs, or messy real-world constraints, human value returns fast.
The likely result is fewer junior backend engineers per team, more senior ownership, and a much harsher bar for what counts as valuable human effort.
DevOps and Cloud Operations — 60-65% automation
Infrastructure as code, CI/CD configuration, cost signals, and routine remediation are increasingly machine-assisted.
But the infrastructure profession is not dying.
It is moving from command execution toward policy, orchestration, and exception handling.
The DevOps engineer who only writes YAML is exposed.
The DevOps engineer who designs safe release systems, rollback strategy, environment policy, and production governance becomes more valuable.
Data Engineers and Analysts — 60-70% automation
The data layer is becoming more self-serve, more AI-mediated, and more dangerous.
That means basic pipeline work, standard ETL transforms, simple exploratory analysis, and dashboard building are all being compressed.
But the closer the work gets to data governance, semantic modeling, causal interpretation, and high-stakes decision support, the more human judgment matters again.
Product Managers — 35% automation
AI is helping PMs write faster, summarize faster, synthesize faster, and prototype faster.
It is not replacing the judgment layer.
Product work still lives in prioritization under uncertainty: deciding what to build, for whom, in what order, under what constraints, and with what tradeoffs across engineering, design, go-to-market, and executive pressure.
That is exactly the kind of work AI supports well but owns poorly.
IRREPLACEABLE Tier: The Accountability Ceiling
Here is the single most important concept in the whole piece:
Software has an accountability ceiling.
Stack Overflow’s 2025 survey shows developers are most resistant to using AI for deployment and monitoring, where 76% do not plan to use it heavily, and for project planning, where 69% do not plan to rely on it.
That is not because developers hate productivity.
It is because responsibility changes the economics.
When failure is cheap, AI moves fast.
When failure is expensive, human oversight comes roaring back.
CTOs and Chief Architects — 10-25% automation
These roles are not protected because they write better code.
They are protected because they own consequences.
Someone has to choose architectures, accept risks, balance speed against reliability, decide where AI is trusted, and absorb the fallout when those decisions fail.
That is not autocomplete work.
That is judgment under consequence.
AI Safety and Alignment Researchers — 15-20% automation
This is one of the safest technical roles in the entire software economy.
The stronger AI systems become, the more valuable the humans who evaluate, constrain, red-team, and align them become.
AI can assist that work.
It cannot replace the people defining what “safe enough” even means.
Security Architects and Senior Reliability Roles — 30-40% automation
AI is very good at surfacing vulnerabilities, summarizing incidents, and accelerating triage.
It is much worse at owning the blast radius of a bad deployment, a security failure, or a distributed systems breakdown.
The tooling gets more automated.
The accountable humans get more valuable.
EMERGE Tier: The New Software Jobs AI Is Creating
The software industry is unusual because AI is not only deleting layers.
It is also spawning entirely new ones.
AI Agent Architects — 25-35% automation
This is one of the clearest new jobs in the stack.
The problem is no longer just “build the feature.”
It is “design a system where multiple models, tools, prompts, permissions, failure states, and human checkpoints all work together safely.”
That is a very different discipline from classic software engineering.
LLM Evaluation, Inference, and Reliability Engineers — 30-40% automation
As more products depend on models, the bottleneck shifts from writing software to making model behavior dependable.
That means evaluation harnesses, inference performance, observability, rollback logic, retrieval quality, prompt governance, and error containment.
These are not disappearing jobs.
They are the new infrastructure layer.
AI Governance and Ethics Roles — 15-50% automation depending on scope
The more AI a software company ships, the more governance work it creates.
Bias detection, model documentation, safety reviews, escalation policy, red-team protocols, and customer-facing disclosures all become part of the build process.
Software is not escaping bureaucracy.
It is building a new kind of it.
The Ladder Collapse
This is the part most people in tech are still underestimating.
The first real casualty is not “software engineering” as a category.
It is the apprenticeship model.
SignalFire’s 2025 report says Big Tech new grads now account for just 7% of hires, and new-grad hiring is down more than 50% from 2019. Startups are below 6%.
That is not a small cyclical dip.
That is a structural warning.
Junior engineers used to learn by doing the exact tasks AI now handles best:
- boilerplate implementation
- routine QA
- simple dashboards
- basic API work
- low-risk support and tooling
If those layers disappear, the industry does not just lose jobs.
It loses its training ground.
That is why this is not merely an automation story.
It is a pipeline story.
The Builder’s Paradox
Software is the first major industry building the tools that erase its own lower-level work.
That creates a strange split.
The best engineers are becoming dramatically more leveraged.
The path to becoming one is getting narrower.
GitHub’s own data shows AI is now standard for new developers. Stack Overflow shows the trust gap is widening. SignalFire shows new-grad hiring is collapsing. Vercel shows product designers and developer-adjacent roles can now build and ship ideas independently.
Put those together and the pattern is clear:
Software is not being automated evenly.
It is being re-tiered.
What This Means For You
If you work in software or internet companies, four things matter now:
-
If your work is mostly routine implementation, assume compression. QA execution, L1 support, internal tools, reporting, dashboarding, simple API scaffolding, and low-complexity frontend are the most exposed layers.
-
If you are early-career, optimize for ownership, not just output. The old path of “do enough tickets and slowly level up” is breaking. You need systems thinking, debugging skill, taste, and the ability to correct AI, not just use it.
-
If you are mid- to senior-level, your leverage is increasing. AI can expand your span of control, but it also raises the bar. More of the stack will flow through fewer humans who are expected to own harder decisions.
-
The strongest long-term roles sit where code meets consequence. Architecture, AI infrastructure, reliability, security, product judgment, model evaluation, and governance all become more important as AI gets stronger.
Software is not dying.
It is becoming harsher.
The implementation layer is being compressed. The judgment layer is being amplified. And the ladder in between is getting steeper.
That is the real AI story in tech.
This is part of my 119-industry AI replacement analysis series, based on the Replace / Amplify / Emerge framework. I’ve analyzed 125 software and internet roles across engineering, testing, DevOps, support, data, security, AI infrastructure, product, and research.
Previously: HR.
Follow for the next analysis: Finance.
Sources
- Precedence Research — Software Market Size 2025 to 2035: https://www.precedenceresearch.com/software-market
- Fortune Business Insights — SaaS Market 2025 (via Fortune Business Insights blog): https://www.fortunebusinessinsights.com/blog/top-software-as-a-service-companies-10883
- GitHub — Copilot: Faster, Smarter, and Built for How You Work Now: https://github.blog/ai-and-ml/github-copilot/copilot-faster-smarter-and-built-for-how-you-work-now/
- GitHub — Gartner Positions GitHub as a Leader in the 2025 Magic Quadrant for AI Code Assistants: https://github.blog/ai-and-ml/github-copilot/gartner-positions-github-as-a-leader-in-the-2025-magic-quadrant-for-ai-code-assistants-for-the-second-year-in-a-row/
- GitHub — Octoverse 2025: A New Developer Joins GitHub Every Second as AI Leads TypeScript to #1: https://github.blog/news-insights/octoverse/octoverse-a-new-developer-joins-github-every-second-as-ai-leads-typescript-to-1/
- Stack Overflow — 2025 Developer Survey: AI: https://survey.stackoverflow.co/2025/ai
- Stack Overflow — 2025 Developer Survey Press Release: https://stackoverflow.co/company/press/archive/stack-overflow-2025-developer-survey/
- SignalFire — State of Tech Talent Report 2025: https://www.signalfire.com/blog/signalfire-state-of-talent-report-2025
- Vercel — Introducing AI SDK 3.0 with Generative UI Support: https://vercel.com/blog/ai-sdk-3-generative-ui
- Vercel — WPP and Vercel: Bringing AI to the Creative Process: https://vercel.com/blog/wpp-and-vercel-bringing-ai-to-the-creative-process
- Zendesk — Zendesk Unveils the Industry’s Most Complete Service Solution for the AI Era: https://www.zendesk.com/newsroom/press-releases/zendesk-unveils-the-industrys-most-complete-service-solution-for-the-ai-era/