The Agentic Commerce Framework
A structural taxonomy of autonomous buying. Six levels. One trajectory. From the human at the keyboard to the algorithm that needs no keyboard at all.
Six Levels of Autonomous Buying
This framework traces the transition of AI from passive research assistant to fully authorized transactional proxy capable of managing enterprise-grade risk or consumer budgets without human initiation.
Commerce
Pure Manual Commerce
The AI plays no role in the transaction. The human uses traditional search engines, navigates individual merchant websites, manually evaluates inventory, inputs shipping details, and executes payment. Every point of discovery, data entry, and authentication rests entirely on the human operator.
There is no hand-off. The human is the system from intent through execution.
& Discovery
Intent Matching & Discovery
The agent interprets complex, multi-variable human intent and surfaces relevant matching products. A query such as "find a durable commercial-grade espresso machine under $2,000 with a footprint smaller than 2 square feet" returns a structured shortlist rather than a raw search results page.
The agent provides a structured list of options with hyperlinks. The moment the human clicks a link, the agent's involvement terminates. The user must still navigate the merchant's external checkout funnel, create an account, and manually input payment details.
Data Routing
Assisted Data Routing
The agent bridges the gap between discovery and checkout by automating administrative friction. Utilizing browser automation (RPA) or deep API integrations, it navigates to the checkout page, pre-populates shipping addresses, inputs billing details, and applies relevant loyalty codes.
The agent halts at the final checkpoint. It presents the completed screen to the human operator, who must visually verify the order details and physically click the submit button. Synchronous human presence is still required to clear the payment step.
Execution
Delegated Execution
The human gives the agent explicit, time-bound or context-specific permission to execute a specific purchase within strict boundaries. Example: "Buy 5 additional licenses of this software if the price drops below $40 per user per month before Friday." The agent holds secure access to tokenized payment credentials via rails such as Google Pay, Stripe Agentic Suite, or secure virtual cards and executes the final API handshake autonomously.
Requires explicit, hard-coded constraints. If any variable falls outside the human's precise mandate (unexpected tax, shipping delay, undisclosed fee), the transaction aborts and escalates back to Level 2 for human review.
At Google I/O 2026, Google announced Agent Payments Protocol (AP2), its infrastructure for exactly this pattern: agents authorized to complete purchases within user-defined brand, product, and budget constraints. When conditions are met, the agent executes. When they are not, it stops.
Buy
Advanced Buy
The agent processes broad, multi-item outcome-based objectives rather than single-SKU commands. It orchestrates, splits, and consolidates multi-merchant transactions inside a single virtualized environment. Operating on unified data layers such as Google's Universal Commerce Protocol or the OpenAI-Amazon ACP, it can evaluate live inventory stability, calculate distinct regional tax liabilities across different vendors, pool multiple items into a Universal Cart, and execute separate downstream payments simultaneously under a single pre-authorized global budget.
The human provides a high-level operational objective. The agent dynamically manages the fragmented merchant landscape to deliver that outcome without returning for approval at each vendor boundary.
Google's Universal Cart, announced at Google I/O 2026 and rolling out summer 2026, is the first large-scale public implementation of this layer: a single cart spanning Search, Gemini, YouTube, and Gmail, operating across merchants via the Universal Commerce Protocol. Level 4 is no longer theoretical.
Autonomous
Fully Autonomous Agentic Commerce
The agent operates continuously in the background without requiring initiating human prompts. It monitors inventory depletion, tracks real-time enterprise supply chains, and assesses changing operational needs. It autonomously predicts demand, negotiates machine-to-machine pricing agreements with merchant agents, issues smart contract micro-payments, and handles automated returns or exceptions without human notification unless policy thresholds are exceeded.
Governed strictly by high-level systemic policies, cryptographic spend-control certificates, and continuous anomaly detection algorithms. Commerce becomes an invisible, self-sustaining utility layer optimized by algorithmic efficiency alone.
Moving up this ladder is less about improving the underlying model and more about the maturity of the infrastructure beneath it.
Reaching Level 4 requires three pillars to align simultaneously. Most organizations currently building "agentic" procurement capabilities are working on only one of the three. That gap is where transformation stalls.
What This Means for Your Organization
Where you sit on this ladder today determines what investments will actually generate ROI versus what will only add technical complexity without unlocking capability.
For Procurement Leaders
The transition from Level 2 to Level 3 is the most consequential decision in agentic commerce adoption. It is not a technology decision. It is a trust and governance decision. Before granting any agent Level 3 authorization, the organization must have:
- A clear spend-mandate policy that an agent can parse as hard constraints
- Tokenized payment rails that do not expose credentials to the model
- Escalation triggers that return control to humans when bounds are exceeded
- Audit logging that captures every transactional decision the agent makes
For Technology Strategists
The model is not the bottleneck. Every major frontier model today is capable of executing Level 3 transactions if the surrounding infrastructure exists. Investment should be directed at:
- Merchant API normalization (structured, reliable data contracts)
- Identity and payment vault architecture separate from the agent layer
- Cross-vendor cart protocol that does not depend on screen-scraping
- Anomaly detection and spend-control systems that operate at transaction speed
For Brand and Channel Teams
At Levels 4 and 5, the human never sees a product page. The agent selects, compares, and transacts entirely within structured data feeds. This fundamentally reorders what "brand visibility" means:
- Schema compliance becomes a distribution channel requirement, not an SEO nicety
- Agent-optimized product data structures will determine search ranking
- Trust signals move from visual design to verified data provenance
- Price and availability accuracy become real-time brand equity factors
For Transformation Advisors
The organizations that will struggle are those treating agentic commerce as an AI feature rather than an infrastructure program. The contraco perspective, developed across 28 years of watching organizations misread the requirements of each digital wave:
- Start by accurately diagnosing which level your infrastructure can actually support today
- The gap between Level 2 and Level 3 has ended more pilots than any capability limitation
- Governance architecture must precede technical deployment, not follow it
- The closed loop at Level 5 is not a destination; it is a regulated utility to be managed
Know exactly where you stand.
Then move deliberately.
contraco works with organizations navigating the transition between levels. Not to accelerate for its own sake, but to ensure the infrastructure, governance, and human readiness are in place before the commitment is made.
Start the Conversation → Explore the Resonance Method™ →