How AI will change software engineering
Kari McMahon · article · 2025-04-14 source ↗ Reader 🖥️
DORA & Engineering MetricsEnterprise AI Adoption & ReadinessJunior Hiring EconomicsPull Requests & Async Code ReviewSystemic Apprenticeship & Skill TransmissionVibe Coding
Generative AI is changing software engineering fast enough that confident predictions about the field's future are themselves unreliable, so engineers and organizations should focus less on forecasting the end-state and more on adapting skills, protecting junior pipelines, and treating the engineering team (not the individual) as the unit that has to stay productive.
practical · expository — A trade-press survey piece (LeadDev) that assembles quotes from a dozen named practitioners and researchers (Charity Majors, Matt Beane, Joe Kim, Joel Greensite, Sabrina Farmer, etc.) into a single narrative about how AI will reshape software-engineering roles, hiring, and skill development. It argues for a stance (resist confident predictions, protect junior pathways, adapt skills) more than it builds a single tight argument.
Structure
| Part | Role |
|---|---|
| Opening framing: the only consensus is that there is no consensus | context |
| Code generation's current impact (vibe coding, startup headcount claims, the 70% problem, DORA 2024 throughput/stability data) | evidence |
| GenAI as 'new junior developer': senior/junior asymmetry, hiring pullback, apprenticeship-pipeline risk | evidence |
| Roles reimagined: skill shifts toward systems thinking, data engineering, prompt engineering | evidence |
| Rate-of-change argument (SWE-bench, multi-agent demos) and closing advice to stand out via creativity | conclusion |
Key terms
- the 70% problem
Named after Addy Osmani's framing (cited via Alex Patow): AI-generated code gets a prototype most of the way there quickly, but the remaining gap to production-grade, scale-ready software is disproportionately hard to close with AI alone. - live wire
Beane's metaphor for how a company's core production function (software delivery) becomes dangerously exposed/volatile once AI-driven change outpaces organizations' ability to adapt their structures and skills.
Propositions
- explicitNo one — including the most confident C-suite predictors (Zuckerberg on AI agents matching mid-level engineers in 2025; Microsoft's CTO on 95% AI-generated code within five years) — currently has the data to make reliable predictions about where AI will take software engineering; the more confident the claim, the less it should be believed.conf: high
- explicitAI-driven productivity gains at the level of individual code generation are being offset by a 1.5% reduction in software delivery throughput and a 7.2% reduction in delivery stability for every 25% increase in AI adoption (Google's 2024 DORA report), reversing the improvement pattern seen in prior DORA reports.conf: highsupports2024_final_dora_report · p2 — Restates the same 2024 DORA finding (1.5% throughput / 7.2% stability decline per 25% AI-adoption increase) as independent corroborating evidence in a different outlet.
- explicitcorroboratedFast, AI-assisted code generation produces 'flab' — unnecessary code that reduces the cost of generation but increases the cost the rest of the team pays reading and maintaining the codebase, because teams spend far more time reading code than writing it.conf: highsupports ←The productivity paradox of AI coding assistants · p8 — Both describe the same underlying mechanism — fast AI-generated output that looks done but isn't — as the hidden cost that offsets apparent speed gains, framed by McMahon's piece as 'flab' that increases the rest of the team's downstream cost.
- explicitSoftware engineering is an apprenticeship industry where the team, not the individual, is the unit that has to be productive — optimizing for the cheapest individually-capable hires while ignoring the team as a system is short-sighted.conf: highsupportsGenerative AI Is Not Going To Build Your Engineering Team For You · p5 — Same claim (team, not individual, is the unit of software ownership/productivity) from the same source (Charity Majors), in near-identical language, in a different article.extendsGenerative AI Is Not Going To Build Your Engineering Team For You · p6 — This article's 'cheapest senior hires' framing is the specific failure mode the fuller Majors essay generalizes into 'an all-senior team is dysfunctional'.
- explicitAI tooling is closing off the pathways junior engineers use to enter the workforce and learn on the job, because senior engineers capture most of the productivity benefit (AI removes mundane work they can already do without, freeing them for complex problems) while companies increasingly favor hiring only senior developers.conf: highsupportsGenerative AI Is Not Going To Build Your Engineering Team For You · p8 — Both identify the real bottleneck as firms declining to hire/train juniors at all (an externalized cost), not a training-capacity problem once hired.
- explicitFaster AI-assisted code production is creating a review bottleneck: code is now produced faster than it can be reviewed and integrated, and senior developers — who must do the reviewing — are not seeing the same productivity gains as the juniors and mid-level engineers producing the code.conf: highsupportsFrom Async Code Reviews to Co-Creation Patterns · p1 — Independently corroborates the code-production-outpaces-review-capacity bottleneck with a different named source (Frank Licea vs. that document's sources).
- implicitSenior developers are becoming more reluctant to mentor and oversee juniors on complex tasks now that AI tools are available as an alternative source of help, which further erodes the informal apprenticeship relationship juniors depend on.conf: medium
- explicitAs AI absorbs the mechanics of coding, the skills that remain differentiating shift toward systems engineering, high-level architecture, data engineering and governance, and precise, persistent prompt engineering (including teaching junior engineers to interrogate rather than accept AI outputs at face value).conf: high
- explicitIf the trajectory implied by rising SWE-bench accuracy and multi-agent collaboration demos continues, the disruption to engineering roles and firm organization over the next year is likely to exceed the disruption of the prior year, making 'keep experimenting' an inadequate plan for individuals or firms.conf: medium
Arguments
Organizations should not plan around confident, specific predictions of AI's endpoint for software engineering; instead they should plan for sustained high uncertainty and fast change. strength: medium
Explicit premises:
- Prominent, highly-confident predictions (Zuckerberg, Microsoft's CTO) have already diverged from measured outcomes (DORA's throughput/stability decline).
- Domain experts (Majors, Beane, Greensite) explicitly refuse to make firm predictions and say those who do are 'selling something'.
Implicit premises:
- Past prediction failures in this domain are evidence that future predictions carry similarly low reliability, not just that those particular predictors were wrong.
Inferences:
- Hiring and skill-development strategy should be built for adaptability (optionality) rather than optimized for a specific predicted future state of the job.
The junior-engineer pipeline is being damaged by the combination of AI-driven senior productivity gains and AI-as-alternative-mentor, with consequences that will constrain the supply of future senior engineers. strength: medium
Explicit premises:
- Senior engineers get more benefit from AI than juniors because it removes mundane work they can already do without.
- Companies are shifting hiring toward senior-only.
- Senior engineers are becoming less willing to mentor juniors now that AI is available.
Implicit premises:
- Today's juniors are the only realistic future source of tomorrow's seniors — there is no other pipeline (e.g., lateral hires from other fields, or AI itself eventually substituting for seniority) that could fill the gap.
- On-the-job mentorship from current seniors is necessary for developing senior-level judgment, and AI cannot substitute for that relationship.
Inferences:
- If this pattern holds industry-wide, there will be a shortage of senior engineers in roughly a decade, even as AI makes senior-level judgment more valuable, not less.
Narratives
- Holly Cummins builds a Quarkus demo app using AI prompts and finds 80% of the generated code unnecessary.
→ Illustrates the 'flab' proposition — AI code generation trades a lower authoring cost for a higher team-wide reading/maintenance cost. - Startup founders tell a venture investor they're shipping code with up to 65% less headcount; Y Combinator's CEO says a quarter of a recent cohort launched with 95% AI-written code.
→ Supports the vibe-coding/headcount-reduction narrative at the small-scale/greenfield end of the spectrum, which the article then immediately qualifies with the '70% problem' at production scale.
Questions addressed
- ✅ Is AI actually making software delivery faster and more stable right now, net of individual productivity gains?
No — DORA's 2024 data shows a net throughput and stability decline as AI adoption rises, even though individual code-generation productivity is up. - ⭕ What happens to the junior-engineer pipeline if AI primarily benefits senior engineers and reduces senior willingness to mentor?
The article names the dynamic (closing pathways, senior-only hiring trends) but does not propose or evaluate a fix. - ⭕ If today's hiring-only-seniors strategy is individually rational for each firm, who (if anyone) will train the next generation of seniors?
Gaps
- incompleteThe article names the senior-hiring-only trend and the junior-pipeline risk as live concerns but never engages the obvious collective-action problem: if the pipeline is a shared industry resource that no single firm has an incentive to fund alone, what — beyond individual advocacy — would actually fix it? This is a question other documents in the corpus engage head-on (e.g. Majors's essay frames it as a professional-advocacy fight engineers must wage inside their own orgs).
Read next: Charity Majors's fuller essay, 'Generative AI Is Not Going To Build Your Engineering Team For You', develops exactly this argument with specific org-design prescriptions (team composition, ramp-cost accounting, who should and shouldn't hire juniors). - incompleteThe hiring-trend data cited (CompTIA) and every named practitioner are US/global-tech-hub voices; the article gives no sense of whether the senior-only hiring shift and apprenticeship squeeze play out the same way in smaller markets or regulated industries with different labor dynamics (e.g. Norway's telco/public sector, with stronger job protection and different new-grad pipelines).
Read next: Look for Nordic or European engineering-leadership sources on junior hiring under AI pressure.
Related documents
- The Bet On Juniors Just Got Better
- State Of AI-Assisted Software Development
- Leading your engineers towards an AI-assisted future
- Engineering Leadership Report 2026
- AI coding creates two kinds of debt. You're only measuring one
- AI isn't making developers more productive – it's making them busier
- Podcast: The AI Joy Gap: Why Some Developers Thrive While Others Struggle
- From Async Code Reviews to Co-Creation Patterns
- Redefining the Software Engineering Profession for AI
- Writing Code vs. Shipping Code: Productivity Effects Across Generations of AI Coding Tools