Hugging Face's Thomas Wolf and other experts say startups like DeepSeek and the rise of AI agents may erode the value of LLMs from OpenAI and Big Tech companies
Ryan Browne / CNBC : Bluesky: @aprilroach . X: @levie Bluesky: @aprilroach : Some great coverage on DeepSeek from CNBC's @ryanbrownecnbc.bsky.social looking at how the tech behind the Chinese startup and advances toward commoditization is paving the way for a shift to the era of next-generation AI agents. 👇🏾👇🏾 X: Aaron Levie / @levie : AI Agents will expand the TAM of software to be multiples larger. When software solves the customer's problem instead of just enabling a solution, traditional IT budget limits no longer apply. Now the business just pays for whatever productivity level they want to achieve.
Context & Ripple Effects
DeepSeek had already challenged the AI arms race’s assumption that larger models are inherently better, as reflected in coverage questioning the bigger-is-better narrative. This article extends that challenge from model-building economics to where value may accrue in AI products.
Earlier coverage also cast agents as a potential route to monetize models through task-oriented software. Thomas Wolf and other experts argue that cheaper, more capable models could make the underlying LLM less central than the agent layer built on top of it.
First-order effects
- The reported shift puts pressure on OpenAI and large technology companies to defend the value of proprietary LLMs as startups such as DeepSeek make capable alternatives more available.
- AI product builders gain a stronger rationale to compete on agents, workflows, and customer outcomes rather than treating access to a frontier model as their primary differentiation.
Second-order effects
- Model providers may face greater pressure to differentiate through distribution, developer tooling, reliability, and integration as baseline model capability becomes less exclusive.
- Software companies can redirect AI investment toward agent deployment and measurable productivity outcomes, reinforcing the agent-led model monetization thesis.
Third-order effects
- If capable models continue to proliferate at lower cost, industry value could shift from a concentrated model layer toward application, integration, and distribution layers—though frontier-model performance and operating costs remain important constraints.
- The AI market may increasingly be organized around agent economics: buyers paying for completed work or productivity rather than simply purchasing access to a model.
The trend: AI is moving from a race to own the most capable standalone model toward competition to deliver reliable, outcome-oriented agents on top of increasingly interchangeable model supply.