Vol.008 — 95% of AI Pilots Fail. Here’s What the 5% Do Differently.
Date: 2026-04-18 | Newsletter
Key Summary
An MIT study found that 95% of enterprise AI pilots delivered zero measurable ROI — not low ROI, zero. A CIO.com survey adds texture: 40% of US businesses are still getting bulk AI value from ChatGPT-style tools, only 13% from agents, and 44% use AI as standalone tools operated by individuals rather than integrated workflows. Zenta introduces the three-era framework — Prompt Engineering, Context Engineering, Harness Engineering — but quickly pushes back on it: the era label doesn’t matter. What matters is whether the business owner can answer one question before writing a single prompt: what specific return am I expecting and how will I measure it? From Functional AI Partners’ work across ASEAN, the 5% that succeed share three behaviors: they start with the process that has the most system handoffs (where AI agents create real leverage), they treat demos as R&D and deployments as business commitments, and they build systems rather than tools. A GPT that answers a question about an invoice is a tool; an agent that receives the invoice, checks it, flags discrepancies, routes it, and logs the exception — without being asked — is a system. Do it simple. Focus on the return.