Key insight from 2026 Anthropic AI job impact report: AI is already capable of doing 94% of jobs in the "computer and maths" category, but current actual coverage is only 33%. There's a massive gap between what AI could do and what is actually being done.
The Thought Experiment
If you built a $10Bn+ e-commerce shop from scratch today, how would you structure the workforce and agents across functions? This requires fundamentally rethinking how work gets done.
Instead of a traditional pyramid structure heavily reliant on human capital for execution, this new org resembles an hourglass:
- Small group of strategic directors at the top
- AI agents and automated systems forming the massive middle layer of execution
- Highly specialized human layer managing exceptions, physical operations, and oversight
The Numbers
| Revenue | ~$12 Billion |
| SG&A / Overhead | $2B+ |
| Headcount | 12,000 employees (60% reduction vs. traditional org) |
Function-by-Function Breakdown
Engineering
AI agents write/test/deploy code. Elite "Orchestrators" replace junior devs, shifting the human role from creator to editor.
Customer Support
AI resolves 95% of Tier 1/2 inquiries. Remaining humans act as "Empathy Escalation Specialists" for high-value VIP issues.
Catalog & Merchandising
Multimodal AI ingests catalogs and generates 3D/SEO assets instantly. Human roles become purely strategic trend forecasters.
Corporate SG&A (HR, Legal, Finance, Commercial)
A tiny "BizOps" team runs the company using AI simulations.
Logistics & Warehouse (The Physical Moat)
AI optimizes routing, but physical labor remains; robotics is evolving fast though.
The Economic Impact
Revenue per employee increases 2-2.5X, decoupling growth from headcount. An AI-first architecture allows you to reinvest into the core pillars of e-commerce, creating a flywheel effect. It allows you to slash prices for customers and give them a "magical" experience.
As product discovery shifts from search engines (Google) and social feeds (Meta) to conversational commerce (AI assistants like Gemini, ChatGPT), the traditional performance marketing playbook breaks too.
Open Questions
- Can incumbents pivot? Against the massive friction of "how we've always done it"? How to re-structure incentives?
- Creativity unleashed? Today's structures in large orgs kill creativity — as AI agents take over the mundane and repetitive, does it get unleashed?
- OPEX shift: You're trading human OPEX for compute OPEX. How to forecast this massive compute costs?
- Security: New security concerns with AI-driven execution?