A blind test comparing RAG to Kimi K3's 1M-token context window shows long context won on answer quality but costs 16x more and runs 3x slower. The real answer isn't one or the other—it's picking the right tool based on corpus size, query volume, and task shape.

• Long context beats RAG on quality when your corpus fits under 20% of the window (tested with 127K tokens), but the cost advantage flips hard at scale: $3.82 vs $0.23 for 12 queries becomes $3,800 vs $230 for 12,000 queries on the same data.
• Query volume, not corpus size, decides the economics; corpus size decides accuracy. A small, frequently-queried corpus still favors RAG despite worse completeness.
• Three distinct tools exist for three task shapes: RAG for frequent narrow lookups on large stable corpora, long context for occasional deep reads of one corpus, and Agent Swarm/context graphs for wide research across many unconnected sources where connections matter.

The author breaks down how large language models lower software development costs but don’t eliminate human-driven feedback loops and ongoing maintenance expenses. By comparing real-world SaaS prices (Jira at $400/month vs. Salesforce at $500/seat) to engineer-hour costs, he defines a “zone of viability” where buying remains cheaper than LLM-powered rebuilding. He frames his own project River against this threshold to gauge its business potential today.

• Rebuilding cheap SaaS with LLMs doesn't pay off: replacing $400/month Jira takes over three years to break even at $96/hour engineer costs, even with minimal maintenance.
• Expensive per-seat SaaS like Salesforce ($25,000/month for 50 seats) crosses into "build" territory since that budget covers 1.5 full-time engineers.
• The "zone of viability" for buy-vs-build depends on both price and novelty/difficulty of re-implementation, not price alone.
• River (Go/Postgres job queue, $125/month Pro tier for up to 20 devs) is positioned to stay on the "buy" side because its design and performance edge make LLM replication costly despite feature copyability.