McKinsey's Automation Curve: A 6-Level Framework for the Future of AI Shopping

McKinsey Publishes the Full Automation Curve

Agentic commerce: How AI shopping agents can change retail

Discover how agentic AI and AI shopping agents are transforming agentic commerce through automation, machine-readable retail strategies.

McKinsey & Company has published a report titled "The automation curve in agentic commerce," systematically organizing how automation progresses in agentic commerce — where AI agents search, compare, and purchase products on behalf of consumers — through a 6-level framework.

In their previous report published in October 2025, the firm characterized this change as a "seismic shift." This latest report serves as a sequel, analyzing where automation accelerates and where it stalls, based on how much purchasing authority consumers delegate to AI agents.

Background and Industry Trends

The foundation for agentic commerce is forming rapidly. McKinsey identifies three driving forces in this report.

First, AI agents have reached "decision-grade" utility. Consumers can now delegate not just inspiration but also shortlisting, basket assembly, and even execution to agents.

Second, the ecosystem has begun building infrastructure to support autonomous transactions. Open-source protocols such as MCP, A2A, AP2, ACP, and UCP are being developed, and the Linux Foundation established the Agentic AI Foundation. Anthropic, Google, Microsoft, OpenAI, and others are participating to build interoperability, identity, and payment infrastructure.

Third, consumer "intent" is increasingly being generated upstream. Birthday party conversations, travel calendar reminders, out-of-stock signals from devices — a world where agents spring into action the moment purchase intent arises is approaching.

Regarding market size, McKinsey predicted in their previous report that even in a conservative scenario, AI agents will mediate $3–5 trillion globally and up to $1 trillion in the U.S. alone in consumer commerce by 2030. According to Adobe's data, traffic to U.S. retail sites from GenAI services surged 4,700% year-over-year as of July 2025.

The 6-Level Automation Curve

McKinsey's framework organizes the progression of automation into six levels. Crucially, this is a "curve," not a "ladder." Higher levels are not always better — the goal is "optimal delegation."

Level 0: Programmatic Convenience (Set & Forget)

The pre-agent foundation stage. This includes subscription and auto-replenishment for consumables like coffee pods, detergent, and diapers. Automation is rule-based and useful, but cannot adapt when needs change. Amazon's "Subscribe & Save" is the prime example — as of 2024, approximately 23% of U.S. Amazon users had at least one active subscription.

Level 1: Assist (Cognitive Sidekick)

The agent supports the "thinking" of shopping but does not execute. It handles analytical tasks like "find four gifts under $75 that arrive by Friday" or "compare sound quality, battery life, and comfort across three noise-cancelling headphones." No cart, no basket, no transaction preparation. It replaces search and comparison, but assembly and execution remain with the buyer.

Level 2: Assemble (Personal Shopper)

A qualitative turning point. The agent transitions from analysis to "orchestration." Say "put together a warm winter outfit for under $150," and the agent resolves trade-offs and constraints, handles taxes, shipping, loyalty perks, and substitutions, returning a checkout-ready basket. The buyer's role shifts from "comparing options" to "approving or adjusting a proposed solution."

Level 3: Authorize (Supervised Executor)

Consumers delegate not just actions but "rules." Under instructions like "order groceries if the total stays under $120 and delivery is Friday between 6–8 PM" or "buy my favorite sneakers from a trusted seller if they drop below $80," the agent executes the entire end-to-end workflow. It only escalates to humans when conditions fall outside the rules.

Level 4: Autonomize (Intent Steward)

Rather than individual transactions, the agent operates toward ongoing goals. Set long-term objectives like "keep monthly household spending under $300" or "maintain my airline loyalty status at minimum cost," and the agent anticipates needs, compares across merchants, and continuously optimizes.

Level 5: Network Autonomy (Multi-Agent Commerce)

Still nascent, this is a world where commerce defaults to agent-to-agent interactions. Personal agents negotiate directly not just with merchant sites but with networks of specialized agents for price optimization, logistics, payments, and loyalty. Intent is brokered, trust is transmitted via reputation signals, and settlement occurs through shared protocols — a "procurement-as-a-service" world.

How Delegation Varies by Category

McKinsey emphasizes that automation does not progress uniformly across all categories.

For "task-oriented" categories like household goods and consumables, delegation climbs the curve quickly. Once consumers verify the agent's accurate basket assembly and appropriate substitution handling, they fully delegate execution. Operational reliability matters more than brand narrative or front-end experience.

For "identity-oriented" categories like luxury goods and milestone purchases, delegation stalls at Levels 1–2. Consumers eagerly ask agents to research and analyze, but make the final purchase decision themselves. Low autonomy here does not mean low value — human involvement is part of the product itself.

For complex purchases like travel and electronics, delegation is "selective." Agents autonomously handle research, comparison, and monitoring, while escalating trade-offs requiring judgment to humans.

Impact and Actions for E-Commerce Businesses

Based on McKinsey's report and CIO's analysis, the actions e-commerce businesses should take are clear.

First, "data quality" becomes the new storefront. AI agents don't browse web pages — they query structured data. Clean catalogs, consistent metadata, and real-time inventory feeds determine whether AI can "find" you.

Second, investment in "machine readability" is essential. McKinsey calls for exposing catalogs, pricing, inventory, shipping, promotions, and return logic as APIs, enabling agents to assemble baskets with human-level accuracy.

Third, addressing "agent authentication" is urgent. This includes supporting authenticated APIs for agents being developed by Visa and Mastercard, purchase authorizations constrained by budget, time, and category, and auditable transaction logs.

Conclusion

The essence of McKinsey's "automation curve" is that agentic commerce is not a linear evolution toward a single goal — the optimal level of delegation varies by category and context. While full automation advances for everyday goods, human involvement remains part of the value for high-ticket and identity-related purchases.

The competitive axis for e-commerce businesses is shifting from "visually appealing UI for human eyes" to "data structures that AI agents can accurately understand." Identifying where your business sits on this curve and which level to target is the starting point for competitive strategy from 2026 onward.