I’m going to be direct: software is not going away, but the shape of software work is changing faster than most people are willing to admit.
I believe we are entering an agentic era where AI doesn’t just autocomplete code, it co-implements systems. That changes who gets hired, what skills are considered “core,” and where human judgment still matters.
This post is intentionally forward-looking. I’ll separate my opinion/projection from what is currently supported by evidence.
My Thesis (Opinion)
1) Software architecture becomes rarer, higher-stakes, and more formal
I expect fewer people to hold true architecture roles, and those roles to become more selective and possibly more credentialed over time. In an agentic world, architecture is no longer “draw boxes and arrows.” It becomes:
- defining system boundaries AI can safely operate within,
- setting policy and compliance constraints,
- owning failure modes and rollback design,
- deciding what must stay deterministic vs probabilistic.
In short: fewer architects, but more responsibility per architect.
2) Data engineering fluency becomes the new baseline for “software engineer”
I expect a big shift where what we currently call “data engineering” becomes normal engineering literacy. If your product has AI in it, then data quality, lineage, retrieval, embedding strategy, and observability are not specialist concerns - they’re table stakes.
My stronger take: the engineer who can’t reason about data pipelines and model interfaces will feel like a frontend engineer in 2008 who refused to learn JavaScript.
3) DevOps does not disappear - it evolves into AI governance in production
DevOps/SRE becomes even more critical. The work shifts toward validating AI-proposed changes, enforcing guardrails, and making sure “it worked in a prompt” doesn’t become “we took down prod.”
Infra will be increasingly generated, but trust will still be earned through verification, policy, and incident response.
4) The entry-level ladder is getting steeper, right now
The painful truth: a lot of junior-level code tasks are exactly what agents absorb first. That doesn’t mean juniors are useless; it means old apprenticeship pathways are breaking before new ones are built.
Large companies with structured graduate programs may keep hiring at scale. Everyone else may expect “AI-augmented mid-level output” from day one.
What Current Evidence Supports
Strong support: AI/data skills are rising fast
- The WEF Future of Jobs 2025 reports strong growth in AI, big data, and software-related roles and skills. Source: World Economic Forum
- BLS still projects strong software developer growth while data-centric occupations remain among the fastest-rising categories. Source: U.S. Bureau of Labor Statistics
Strong support: DevOps/platform quality becomes more important with AI
- DORA findings suggest AI can improve parts of the development process, but delivery outcomes depend heavily on platform quality and operational fundamentals. Source: DORA 2024 Report
Moderate-to-strong support: junior role pressure is real, but uneven
- There is growing evidence and credible discussion that entry-level pathways are under pressure as AI handles routine implementation work.
- At the same time, hiring patterns are uneven by company type, geography, and maturity of internal training pipelines.
Strong support: AI coding tools increase productivity in many contexts - but not automatically
- Studies around AI coding assistants show productivity and confidence gains in many settings.
- Results vary by team process, review culture, and test rigor; quality regressions can occur without guardrails.
Where I’m Projecting Beyond the Data (And I’m Owning That)
These are my bets, not settled facts:
- “Few software architects”: evidence shows role polarization, but not definitive proof of formal accreditation.
- “Data engineering becomes the default SWE identity”: evidence supports convergence of skills, but not full replacement of traditional engineering tracks.
- “Junior ladder pulled up”: directionally supported, but likely to be cyclical and industry-dependent rather than absolute.
A Practical Career Map for the Agentic Era
If you’re a student, junior, or mid-level engineer, here’s the adaptation path I believe matters most:
Learn systems + data together
Build things where retrieval, metrics, and model behavior are first-class concerns.Treat AI output as untrusted code
Verification, testing, and failure analysis are career accelerators now.Develop “prompt-to-production” judgment
Anyone can generate code; fewer people can make safe, maintainable, compliant systems.Build in public with measurable outcomes
Show latency reductions, lower error rates, improved reliability - not just demos.Get good at platform constraints
CI/CD, policy-as-code, secrets, observability, rollback plans: this is where human leverage compounds.
Final Take
I don’t think software engineering is dying. I think it’s splitting.
One path becomes high-trust engineering: architecture, data systems, platform reliability, and governance.
The other path becomes commoditized implementation mediated by agents.
My opinion is simple: the winners won’t be the people who “use AI.”
They’ll be the people who can direct, constrain, verify, and operationalize AI at system level.
That’s the new craft.
Top comments (0)