Scale AI lost its focus on product, says Mercor’s CEO
Summary
Mercor cofounder and CEO Brendan Foody told the 20VC podcast that Scale AI, while strong on distribution and sales under former CEO Alexandr Wang, ‘lost the focus on product’ — specifically on scaling and maintaining data quality. Foody contrasted Mercor’s high pay and elite annotator recruitment with rivals, saying Mercor pays an average of about $95 an hour versus roughly $30 an hour at companies like Scale. The comments come after a tumultuous period for Scale that included a $14.3bn investment from Meta, reported security lapses involving public Google Docs, and layoffs of full-time staff and contractors.
Key Points
- Brendan Foody criticised Scale AI for sidelining product quality and the ability to scale quality annotation work.
- Mercor says it pays contractors significantly more (around $95/hr average) and recruits elite annotators; Scale’s typical pay is cited at roughly $30/hr.
- Scale saw scrutiny after a large Meta investment and reports of security lapses (public Google Docs exposing training materials).
- Scale responded by defending its data quality metrics and emphasising it remains market-leading.
- Earlier this year Scale cut about 200 full-time staff and 500 contractors, citing overhiring and organisational inefficiencies.
Context and relevance
The story matters for anyone tracking the AI-training supply chain and the economics of human-in-the-loop systems. Quality of annotation directly affects model performance and safety; pay and recruitment practices determine who does the work and how well. The piece highlights tensions between rapid scaling, investor-driven growth (eg Meta’s investment), and operational discipline — issues central to the broader AI industry as companies race to supply training data and talent.
Why should I read this?
Because if you care who actually trains the models and whether shortcuts are being taken, this is the kind of backstage drama that explains why some AI firms stumble even with huge funding. Short version: payouts, people and data security matter — and it changes the product. We read it so you don’t have to.
Author style
Punchy — the piece is framed as a direct critique from a rival CEO and flags practical, industry-facing concerns (pay, quality, security) rather than abstract debate. Worth a quick read if you follow AI ops, procurement or trust in model development.
