Unconventional AI CEO Naveen Rao says agentic coding is overhyped as a market category and not worth investing in
Naveen Rao said that agentic coding is overhyped as a market category. He believes the value mainly comes from model providers, rather than harnesses or IDEs, and that users can switch fairly smoothly between these tools, leaving them with insufficient stickiness. Aside from model providers, he would not currently invest in this category.
Rao is the CEO of Unconventional AI and previously served as CEO of MosaicML and Nervana. The above views came up in a discussion with Harry Stebbings. He also pointed out that pure software investors face challenges, the robotics market will be far larger than coding agents, and the inference market may reach a scale of $10 trillion in the next decade.
Original sources:
Title: Harry Stebbings-related X post and discussion summary Publisher: Harry Stebbings (@HarryStebbings) Publication date: 2026-10-25 URL：https://x.com/HarryStebbings/status/2114523656546316565 Source nature: social media post Verification status: public content read in this session Supporting content: direct quotation of Rao's remarks on the agentic coding market judgment
Title: Background and views related to Naveen Rao Publisher: public personal profile and paraphrase Publication date: 2026 URL：https://x.com/NaveenGRao Source nature: public personal information Verification status: read in this session Supporting content: Rao's identity, past company experience, and related remarks on the robotics and inference markets
ABAB AI Insight
Rao's judgment separates the tool layer of agentic coding from underlying model capabilities. Model providers capture the main pricing power through continued training and inference scale, while independent harnesses or IDEs need to rely on specific workflows to lock in users; switching costs are currently low.
This view is consistent with recent market observations: some developers report that agents still require substantial human verification on complex tasks, and verification costs may offset the gains from automation. Rao's prior company experience involved the sale of hardware and training infrastructure, and his stance emphasizes value capture at the infrastructure and model layers rather than at the application layer.
If model capabilities continue to improve rapidly, the differentiation space for independent coding agent tools may narrow further, and capital allocation may flow more toward model training, inference infrastructure, and robotics hardware. This judgment is currently a personal investment view and does not reflect public data on large-scale capital exits or product failures.
ABAB News · Cognitive Laws
The stickiness of the agentic coding tool layer depends on workflow lock-in, not on model capabilities themselves.
When model providers capture the main value, return expectations for independent tool categories decline.
Rising verification costs may limit the net efficiency gains from agent automation.