Harry Stebbings: 90% of AI Podcasts are False
Founder of 20VC, Harry Stebbings, stated that currently 90% of AI podcast content is inaccurate, as guests are afraid to speak candidly due to concerns about offending core model suppliers (their main source of income).
He pointed out that Chinese open-source models like Kimi K3 outperform top Western closed-source models, breaking the narrative that "foreign labs only distill American technology," and proving that the pace of ecological evolution far exceeds the capacity of government centralized regulation.
Stebbings emphasized that software will no longer serve as a corporate moat (due to its instant generation), and sustainable value will only belong to network effects and proprietary data moats; at the same time, 70% of Neolabs' research projects will fail, and survival will require a focus on hyper-growth P&L metrics.
Source: Public Information
ABAB AI Insight
Harry Stebbings' critique is not an isolated viewpoint but a reflection of the trend towards "de-bubbling" in the AI investment circle by 2026: 20VC has invested in 10 unicorns, and its podcast has shifted from "industry promotion" to "reverse due diligence," reflecting LPs' upgraded information demand after fatigue with AI narratives.
Similar cases can be compared to the rise of "due diligence podcasts" after the SPAC bubble burst in 2021, and the cessation of "The Industry Standard" at the end of the 2000 internet bubble—reflecting the "speechlessness dilemma" of "industry media" at the bubble's peak: advertisers are the subjects of reporting, leading to content distortion.
On the capital path, Stebbings proposed the "Neolab differentiation" framework: among 75 Neo Labs, 2/3 will head towards low-value acqui-hire, with only 1/3 able to achieve the "research → product → revenue" closed loop, similar to the filtering process of biotech companies in the 2010s shifting from "paper-driven" to "clinical data-driven."
Essentially, this belongs to the "moat migration" in technological substitution: software has devolved from a "differentiated asset" to a "basic production factor," with the mechanism being that AI drives the marginal cost of code generation close to zero, shifting corporate value from "owning software" to "owning data + network effects + distribution channels."
ABAB News · Cognitive Laws
- 90% of AI podcasts are advertisements, guests dare not offend their sponsors.
- Software is no longer a moat; data and network effects are.
- 70% of Neolabs will fail; the remaining 30% rely on P&L, not PPT.