Price Cuts Don't Mean Agents Are More Profitable
krishnan · x · 2026-07-15
DeepSeek's 75% price drop for V4-Pro seems like a boon for AI developers, but the author argues it does not automatically solve agent profitability.
The core argument: a chatbot typically handles a single user query with one model call, whereas an agent might break a request into multiple loops like planning, retrieval, tool usage, verification, summarization, and follow-ups. Thus, actual token consumption might scale faster than price reductions. Behind a single user-facing answer lies a potentially expensive chain of executions.
Consequently, many AI business models become fragile: if a product is priced as SaaS but operates like high-consumption infrastructure, the most active users might actually be the least profitable. The author believes the next crucial phase isn't about "cost per token," but rather:
- Which steps warrant a frontier model
- Which steps can be routed to smaller models
- Which contexts should be cached, compressed, or ignored
- Which loops should be terminated early
Ultimately, the metric to watch is cost per completed task, not token price or seat price.
More from Venture
- Anthropic files S-1 with SEC, formally kicking off its IPO process — ns123abc · 2026-10-26
- Anthropic's top 2 customers fell from 54% of revenue in 2023 to about 21% by 2026 — BenBajarin · 2026-10-26
- Nearly half of Claude revenue now flows through cloud marketplace commitments, up from 11% — BenBajarin · 2026-10-26
- Ferryman cross-posts to 15+ social platforms from one post, priced $30-$100/mo — KevinNaughtonJr · 2026-10-26
- Anthropic is FCF positive even after SBC, countering bearish macro narratives — GavinSBaker · 2026-10-26
- Anthropic ARR goes vertical after price cuts: tokens up 4x, $100B+ ARR back in play — GavinSBaker · 2026-10-26