We Paid Millions to Learn This So Amazon Agencies Don’t Have To
April 9, 2026·10 min
Steven Pope opens by positioning My Amazon Guy as a 420-person agency at approximately $20 million annual revenue that is actively pivoting toward agentic AI. He explains that he was skeptical of AI for two years but changed his view in the past 60 days, citing rapid capability improvements as the reason. He describes a $2.6 million failed software development investment that his COO rebuilt in a single weekend using Claude, now being productized under the name Konfidential — a CRM and billing tool targeted at legal professionals. He frames this as evidence that the economics of software development have been fundamentally disrupted. Steven characterizes AI adoption using a blunt metaphor: 'AI is like teenage sex — everybody thinks everybody else is doing it, nobody really is, and if they are, they're doing it wrong.' He identifies two early MAG AI moves: eliminating the entire copywriting division shortly after ChatGPT launched, and reducing the design team from 55 to roughly 30 by cutting 11 roles for employees who refused to adopt AI tools. He predicts that design roles will shift from requiring design skills to requiring proficiency with AI tools, citing a non-designer colleague outperforming trained designers using AI software. Steven describes offering free AI consulting calls to other agency owners in exchange for use cases, which surfaced the idea for an AI churn early-warning system. He explains that first-generation tools like Fireflies can flag keywords in individual calls but cannot detect sentiment drift across a client's entire call history — which is what MAG's system now does. The system generates probability scores for client churn risk and flags which accounts need proactive outreach from senior staff. He extends the data model to employee retention: MAG can now calculate the exact revenue at risk if a specific account manager leaves, enabling fact-based decisions about whether to meet salary demands or replace the role. He describes this alongside an 'intern model' designed to make any seat replaceable three-deep. Steven also mentions a MAG internal server with role-isolated access to SOPs, employee culture index surveys, and evaluation records — built using Claude rather than enterprise SaaS tools. He closes with a Jeff Bezos quote about too many good ideas being more dangerous than bad ones, framing AI as the force that has eliminated development cost as a filter for ideas — making prioritization the new bottleneck for operators.
