AI in Software Development: Real Data vs. Layoffs Myths
A July 2025 METR study found that 16 experienced developers using AI tools worked 19% slower on real open-source tasks. Confidence interval: +2% to +39%. Subjectively, developers estimated a 20% speedup.
February 2026 update: newer models, more participants. For returning developers: 18% speedup (interval: −38% to +9%). For newcomers: 4% speedup (−15% to +9%). The authors note only weak evidence of improvement—but the trend is shifting toward acceleration. Wide confidence intervals underscore the effect’s instability.
The first study went viral; the second went largely unnoticed. This illustrates media’s selective narrative framing.
Block and Klarna Case Studies: Layoffs—and Rehiring
In February 2026, Block Inc. laid off 4,000 employees (40% of its workforce). CEO Jack Dorsey attributed this to 'intelligence tools.' Shares surged 24%; market cap rose $8 billion. Yet 2025 revenue stood at $24.2 billion—flat YoY—while net profit halved. Gross profit per employee rose from $500K (2019) to $1M (2025), but that trend began years earlier.
Piper Sandler analysts flagged rising transaction losses (from 11% to 18%). Block quietly launched rehiring efforts. Dorsey acknowledged flexibility to course-correct.
Klarna cut 700 support staff after deploying an AI chatbot. It saved $40 million across 2.3 million conversations—shrinking headcount from 5,000 to 3,000. Customer satisfaction dropped 22%. The bot failed on disputes and fraud. Engineers were flooded with tickets. The CEO admitted overreach.
The shift toward 'Uber-style' gig work: lower pay, no benefits.
Key lessons from these cases:
- Layoffs often deliver short-term stock boosts fueled by narrative—not productivity gains.
- Declining quality metrics trigger rehiring—often under worse terms.
- Forrester: 55% of employers regret AI-driven layoffs; one-third spent more on rehiring than they saved.
The Economics of Fear: Stakeholder Incentives
CEOs get immediate market reactions to the 'we’re AI-forward' story. Block +24%, Salesforce — down. Vendors (e.g., Anthropic) stoke fears of 'eliminating 50% of entry-level roles' to sell tools. OpenAI doubled its headcount to 8,000; 1.3 million AI-related jobs have been created (+13× since 2022).
Media: clickbait headlines like 'AI Will Replace 80% of Developers.' Education: $2K–$25K courses promising 'AI-proof careers.' Consulting: Accenture tracks AI login activity for performance reviews.
Facts vs. Interpretations:
Verified data:
- METR: trend toward acceleration—but wide confidence intervals.
- 26.9% of production code now AI-assisted (up from 22%).
- Lab studies show 55% speedup on routine tasks; real-world project gains are more modest.
- U.S. IT salaries up 16.7% since 2022 (Dallas Fed).
- 67,000 open engineering roles (+78%).
- U.S. labor productivity up 2.7% (2025).
Interpretations:
- Many layoffs reflect Wall Street storytelling—not automation-driven efficiency.
- Panic serves stakeholders’ interests.
- AI-literate developers remain highly secure.
Practice: AI’s Strengths and Weaknesses
AI accelerates:
- Boilerplate, routing logic, database migrations, test scaffolding, and early drafts.
Example in Laravel: prototype built in 1.5 hours instead of months.
AI undermines:
- 45% of developers spend more time debugging AI-generated code.
- 45% of AI-written code contains OWASP Top 10 vulnerabilities.
- CodeRabbit reports 1.7× more issues vs. human-authored code.
Recommendations:
- Never delegate authentication or payment logic to AI without rigorous human review.
- Use AI for routine tasks—but design architecture yourself.
- Organized teams + AI = 50% fewer incidents; chaotic teams + AI = double the incidents.
What Matters Most
- METR confirms a trend toward acceleration—but the effect remains unstable.
- 55% of AI-driven layoffs are regretted, per Forrester.
- IT salaries and job openings continue rising—despite alarmist narratives.
- AI amplifies teams when used thoughtfully—not replaces them.
- Manual review is non-negotiable for security-sensitive code.
— Editorial Team
No comments yet.