About

The problem with AI in 2026 isn't buying tools and adding them to the experimental layers of a company — it's keeping them from adding clutter and slop, and breaking what already works. With years of experience building reliable AI and data systems, Hrishi talks about two things that make a difference: building first-mile data connectors to expose the right information to agents, and putting in place AI-executable SOPs that allow for fixing mistakes and preventing regression. AI should meet companies where they are instead of needing a bottom-up rebuild — and here's how.