Beyond The Pilot: Enterprise AI in Action Most enterprise AI agents are Slop - here’s why they fail
Jan 7, 2026
Amjad Masad, CEO of Replit and a visionary in programming accessibility, unpacks the $10 trillion potential of enterprise AI in a fascinating discussion. He critiques the majority of AI agents as 'Slop,' explaining their generic failures and the need for quality inputs. Masad introduces 'Vibe Coding,' where non-tech experts shape software, and reveals Replit's 'Computer Use' hack, making agent development cheaper and faster. He warns leaders to ditch rigid roadmaps in favor of rapid iteration to thrive in the evolving AI landscape.
AI Snips
Chapters
Transcript
Episode notes
Agents Must Act, Not Just Chat
- Most enterprise agents are "toys" because they retrieve information rather than autonomously act and handle messy infrastructure.
- True agents must be autonomous and take actions, not just serve as chatbots.
Invest Platform Effort To Reduce Slop
- Imbue agents with "taste" by investing platform effort: use curated prompts, design systems, and spend tokens for quality.
- Add a testing loop so a verifier agent checks outputs and feeds back improvements.
Database Isolation Fixed A Production Loss
- After a July incident that deleted a developer's production data, Replit isolated dev and prod databases by default to prevent future mistakes.
- They now migrate schemas but never share development data with production, making errors non-destructive.