AI writes 100% of the code at Anthropic, OpenAI top engineers say—with big implications for the future of software development jobs

Anthropic engineers report that AI tools now generate the vast majority of their code: Claude Code and Opus 4.5 are credited with “100%” of code for some engineers (Boris Cherny says he hasn’t written code in over two months and shipped 22 and 27 PRs on consecutive days, all AI-written). CEO Dario Amodei predicts AI may handle most or all software engineering within 6–12 months; Anthropic leveraged Claude Code to build a non-coder product, Cowork, in roughly 1.5 weeks and is shifting hiring toward generalists. The development boosts productivity and could democratize software creation, but the article notes quality caveats (subtle conceptual errors, dead code) and potential impacts on entry-level engineering roles.

Analysis

Market structure: AI-native code generation is a demand shock for GPUs, cloud compute, and model-serving infrastructure (winner set: NVDA, AMD, AMZN, MSFT, GOOGL) and a supply shock for human coding labor (losers: entry-level hiring, offshore IT services). Expect developer-productivity to compress labor demand by 20–40% for routine tasks within 12–24 months while increasing cloud/GPU spend 30–60% for large enterprises that adopt LLM pipelines. Cross-asset: tighter credit metrics for cloud/AI leaders (positive for corporate bonds), higher implied vols for semiconductor equities; modest upward pressure on power/energy demand in data-centers over 1–3 years.

Risk assessment: Tail risks include rapid regulatory action (EU/US AI safety/IP rules within 6–12 months), large-scale model-caused outages or IP suits, and supply-chain limits for high-end GPUs leading to price spikes. Immediate market reactions (days) will be headline-driven; adoption-driven revenue shifts play out over quarters (2–8) and labour-market reallocation over years (2–5). Hidden dependency: concentration on NVIDIA-style accelerators and a small number of cloud providers creates single-point systemic risk.

Trade implications: Direct long bias to NVDA (core infra), MSFT/GOOGL/AMZN (cloud + tooling) and selective longs in verification/cybersecurity (SNPS, CRWD) to monetize increased testing/security demand; selective shorts in legacy offshore IT services (INFY, CTSH) and smaller consultancies exposed to low-complexity coding. Use 3–9 month call spreads on NVDA/MSFT to capture upside while selling premium; implement pair trade long NVDA / short INFY with size 1–2% each and horizon 3–12 months.

Contrarian angles: Consensus underestimates integration, security and maintenance costs — generated code will drive demand for MLOps, QA, and IP insurance, not pure headcount elimination. Reaction may be underdone for cloud/GPU spend (upside) but overdone in valuations of pure-play AI app names; historical parallel: CAD automation boosted semiconductor and tooling vendors even as manual labor shifted. Watch for 10–20% drawdowns if major regulation or a chip shortage occurs; that makes hedges and verification longs asymmetric beneficiaries.

Key Decisions for Investors

• Establish a 2–3% portfolio long in NVDA (NVIDIA) for a 6–12 month horizon via a buy-and-hold or buy 6–9 month 1:1 call spread (buy ATM, sell 20–25% OTM) to cap cost; target +40–60% upside, stop-loss at -15%.
• Allocate 1.5% each long to MSFT and GOOGL and 1% to AMZN (total ~4%) to capture cloud services monetization over 3–12 months; implement 3–6 month covered-call overlays if IV spikes above 40% to harvest premium.
• Implement a 1–2% pair trade: long NVDA (or MSFT) vs short INFY and/or CTSH (split exposure) to profit from automation reducing low-end outsourcing demand; close if short leg outperforms the long by >10% or after 12 months.
• Add 1–2% long in verification/security software (SNPS, CRWD) as a hedge to increased QA/security spend; target 6–18 month horizon and scale up if adoption metrics (enterprise LLM POCs >50% in surveys) rise in next 3 quarters.
• Buy disaster hedge: purchase 6–12 month ATM puts equal to ~30% of NVDA/MSFT position notional if EU/US AI regulatory bills advance (monitor Congressional/EC announcements over next 60–180 days) to cap regulatory tail risk.