How Ramp solved the Fatal Flaw in AI Agent Strategy ft. Rahul Sengottuvelu

Ramp’s Rahul Sengottuvelu argues that agents fail because of feature parity gaps, and the fix is computer-use over your own frontend instead of rebuilding tools from scratch.

• Every major AI agent (Google Slides, Siri, etc.) fails the same way: feature-incomplete tools that frustrate users immediately after one success.
• Root cause: agent teams are strapped onto mature products and forced to rebuild tool-by-tool, always far behind the main frontend team.
• Ramp’s fix: run a computer-use agent in a headless browser with the user’s real credentials to drive the existing frontend, not a parallel tool layer.
• This gives instant full feature coverage — Ramp’s card-branding change (a niche feature) became agent-accessible without any bespoke tool work.
• Building over your own frontend also solves auth: the agent inherits user roles and permissions automatically rather than re-implementing them.
• Custom scaffolding (DOM heuristics, component-library-to-CLI rendering, nav trees) makes computer use reliable today, unlike raw general-purpose computer use.
• Recommendation for distribution-heavy companies: expose a complete tool-calling interface over all features, not a subset, to capitalize on the emerging agent economy.

2025-05-12 · Watch on YouTube