Anthropic’s Boris Cherny, creator of Claude Code, says there are days he manages tens of thousands of AI agents at once

Anthropic says Claude Code is now generating all of its code internally, contributing to an 8X increase in code written at the company since the start of the year. Boris Cherny described the product as managing hundreds to tens of thousands of AI subagents and said it may be the first product that “just takes off.” The article is broadly positive for AI software adoption, though it also flags recursive self-improvement as a key risk.

Analysis

This is less a single-company product story than an inflection point for software labor economics: if agentic coding systems can coordinate subagents, the marginal cost of shipping code falls faster than the market is pricing. That should compress the value of low-complexity implementation work first, then move up the stack into QA, code review, and even parts of product management, because the bottleneck shifts from writing code to specifying good objectives. The second-order winner is whoever owns the workflow layer and distribution into enterprise development teams; the loser is point-solution tooling that depends on high human-touch services.

The immediate equity implication is not "all software is bullish," but a widening dispersion inside software. Names with strong developer workflow lock-in and an AI-native seat expansion story should outperform, while legacy horizontal SaaS with weak moat and heavy services attach will face pricing pressure as customers ask why they need as many engineers, QA vendors, or outsourced dev shops. Over 6-18 months, the market will likely reward evidence of AI-generated throughput converting into measurable gross margin expansion, not just usage metrics.

The biggest underappreciated risk is governance, not model capability: once systems begin proposing roadmap items and acting autonomously, security, provenance, and liability become board-level constraints. A single high-profile security incident or model-driven bad deploy could create a multi-quarter pause in enterprise adoption even if productivity gains are real. On the upside, if no major incident occurs, the pace of adoption could accelerate nonlinearly because the ROI compounds: fewer engineers needed per feature, faster release cycles, and higher software capital efficiency.

The contrarian view is that the market may be overestimating near-term monetization and underestimating integration friction. Enterprises rarely translate productivity gains directly into headcount cuts; they usually reinvest in more projects, which delays margin capture but increases long-run revenue opportunity for the platform winners. That makes the best expression a relative trade, not a blanket short on software.

Key Decisions for Investors

• Long MSFT / short a basket of legacy low-moat SaaS names over 3-6 months: Microsoft has the clearest path to monetize AI coding across the full developer stack, while commoditized workflow vendors are most exposed to budget scrutiny and seat consolidation.
• Long NET or DDOG on any 5-8% post-runback weakness over the next 1-2 months; both benefit from more machine-generated traffic, more deploys, and more security/observability demand as code volume rises. Risk/reward is attractive if AI increases release frequency faster than it reduces spend.
• Short EPAM or other IT services/pro services proxies on 3-6 month horizon; if agentic coding cuts implementation labor even modestly, utilization and pricing pressure should show up before topline growth fully rolls over.
• Buy 6-12 month call spreads on CRM or NOW only on pullbacks: these platforms can monetize AI-driven workflow expansion, but upside depends on proving net revenue retention and attach rates, so use defined risk rather than outright longs.
• Pair trade: long AI infrastructure beneficiaries (NVDA or AVGO) against short mature enterprise software with weak AI differentiation. The infrastructure leg captures compute demand from recursive agent loops; the short leg faces margin compression and slower seat growth.