Maintained by Serafim Tech Limited · serafimtech.io Version 1.0 · 2026 · Licensed under Creative Commons Attribution 4.0 (CC BY 4.0)
Serafim Tech Limited · Company No. 12884078 · Registered in England & Wales
Open framework. This work is licensed under a Creative Commons Attribution 4.0 International License. You are free to share and adapt it for any purpose, including commercially — provided you give appropriate credit to Serafim Tech, link to the license, and indicate if changes were made. We encourage merchants, platforms, agencies, and tool-builders to use it, teach it, and build on it.
The Agent-Readiness Hierarchy is a simple, reusable model for answering one question: how ready is a store for AI agents that shop on a customer’s behalf?
As commerce shifts from humans browsing pages to agents acting on intent, “are we ready?” became a question every merchant needs to answer and almost none could measure. This framework turns that vague question into a diagnostic — a way to locate any store on a map, score it, and know exactly what to fix next, in priority order.
It is deliberately open. A shared, free vocabulary for agent-readiness is more useful to everyone — including us — than a proprietary one locked in a drawer.
| File | What it’s for |
|---|---|
| SCORECARD.md | A fillable template — score your store and read off bottleneck + readiness. |
| WORKED-EXAMPLE.md | A fictional store scored end-to-end, showing how scores become a decision. |
| LAYER-CHECKS.md | Objective pass/fail criteria for every score at every layer. |
| OPEN-QUESTIONS.md | The framework’s open questions, known limitations, and roadmap — where it’s unfinished and where it’s going. |
An AI agent acting for a shopper engages with five things, and lower layers weight the value of higher ones: a brilliant promotion an agent can’t parse (Layer 5 on a weak Layer 2) isn’t worth zero, but it’s worth far less than it should be. The layers don’t strictly block each other — a merchant can work several in parallel — but effort spent high in the stack pays off only as far as the layers beneath it support it. So the hierarchy is generally read — and prioritized — from the bottom up.
┌─────────────────────────┐
│ 5 · LEVERAGE │ Can I capture value & learn?
┌─┴─────────────────────────┴─┐
│ 4 · TRUST │ Can the transaction be trusted?
┌─┴─────────────────────────────┴─┐
│ 3 · TRANSACTABILITY │ Can an agent actually buy?
┌─┴─────────────────────────────────┴─┐
│ 2 · COMPREHENSION │ Can it understand what I sell?
┌─┴─────────────────────────────────────┴─┐
│ 1 · PRESENCE │ Can an agent find & reach me?
└─────────────────────────────────────────┘
build up ↑
| Layer | The question the agent is asking | What lives here | Mostly in your control? |
|---|---|---|---|
| 5 · Leverage | Can I (the merchant) capture value and learn? | Agent-readable promotions, loyalty & bundles; agent attribution & observability; new monetization; the data feedback loop | Yes |
| 4 · Trust | Can the transaction be trusted? | Delegated authorization, Know-Your-Agent posture, transparent returns & substitution, dependability signals | Partly — shared with payment networks, identity standards & regulators |
| 3 · Transactability | Can an agent actually buy? | Cart accepts agent line items, computes totals, returns a working checkout URL; payment handlers; fulfillment options | Mostly |
| 2 · Comprehension | Can it understand what I sell? | Catalog quality, structured data, variant resolution, semantic metadata, localization | Yes |
| 1 · Presence | Can an agent find and reach me at all? | Discovery profile reachable & parseable; live agent endpoint; agentic channels enabled | Partly — shared with your platform & the protocol layer |
The base is widest on purpose: presence is the broadest foundation that everything else leans on, and each layer narrows because fewer merchants get that far. Your weakest low layer is usually your real bottleneck — a brilliant Layer 5 sitting on a broken Layer 1 delivers a fraction of its potential in practice.
Score each layer 0–3. This is what turns the hierarchy from a diagram into a diagnostic. The scoring is still partly judgment-based today; a detailed, checkable rubric for each layer at each score is the priority for the next version (see Known limitations).
| Score | State | Meaning |
|---|---|---|
| 0 | Absent | The capability isn’t there; agents fail at this layer. |
| 1 | Basic | Present but incomplete or unreliable (e.g. catalog exists but variants don’t resolve). |
| 2 | Functional | Works correctly for a competent agent, end to end. |
| 3 | Optimized | Works — and is structured so agents prefer you over a comparable merchant. |
Reading the scores
These keep the framework strategic, not merely technical. They apply at every layer.
1 · Agent-ready, not agent-dependent. Expose your capabilities to agents while fighting to keep what matters: stay the merchant of record where you can, and protect identity, payment authorization, and the post-purchase relationship and data. The pressure toward disintermediation is real and you may not win every inch of it — but readiness should be a deliberate trade, not a surrender of the customer relationship by default.
2 · The same machinery serves humans and agents. Every layer, done well, also improves the human experience — clean data, a cart that always computes totals, transparent returns, exposed promotions. The work is therefore never wasted, even if agentic adoption arrives slower than predicted. Build for the agent; the human benefits too.
This is version 1.0 — a structured, reasoned model, not yet an empirically validated instrument. We have not yet published data proving that a high readiness score causes better agent conversion. We believe it does, and we’re committing to test it openly: as agentic traffic grows, we’ll publish whether readiness scores actually predict agent-driven revenue, and revise the framework where the data disagrees with the theory. Treat the score as a well-reasoned hypothesis to act on, not a law to trust blindly.
A note on our interest. Serafim Tech builds commercial tools in this space, and this framework maps onto problems our products help solve. We’ve published it openly, under a permissive license, precisely so it’s useful and verifiable independent of our products — you can apply the whole framework without ever touching anything we sell. We think being upfront about this makes the work more trustworthy, not less.
We publish the framework’s weaknesses alongside the framework, because a model you can argue with is more useful than one that pretends to be finished. The strongest objections, and our current responses:
If you think we’ve got a layer, an order, or a principle wrong, we want to hear it. The canonical version improves through challenge.
This is a canonical, openly-licensed framework, and it improves through use and challenge. Corrections, worked examples, and translations are all welcome — see CONTRIBUTING.md for how to propose them. Small fixes can go straight to a pull request; changes to the layers, the scale, or the principles start as an issue so we can discuss them first.
The Agent-Readiness Hierarchy is an original framework developed by Serafim Tech. It synthesizes and builds on publicly available thinking about agentic commerce, including McKinsey & QuantumBlack’s The Agentic Commerce Opportunity (October 2025) and Serafim Tech’s own diagnostic work. Those sources are cited as intellectual influences; the layered hierarchy, the maturity scale, and the governing principles are Serafim Tech’s own contribution, and it is this contribution that is offered here under CC BY 4.0.
Serafim Tech (2026). The Agent-Readiness Hierarchy: An open framework for agentic-commerce readiness, v1.0. Licensed under CC BY 4.0. Available at serafimtech.io.
Serafim Tech builds the capability layer for agentic commerce — composable, agent-readable commerce primitives that let merchants be found, understood, and transacted with by AI agents. We publish this framework openly because a shared standard for agent-readiness helps the whole ecosystem mature, merchants and builders alike.
Learn more at serafimtech.io.
Serafim Tech Limited Registered office: 124–128 City Road, Islington, London, EC1V 2NX, United Kingdom Company No. 12884078 · VAT 359617656 · Registered in England & Wales
Note: Serafim Tech Limited (agentic commerce, serafimtech.io) is unrelated to any similarly named gaming or hardware brand.
© 2026 Serafim Tech Limited. Licensed under CC BY 4.0. “The Agent-Readiness Hierarchy” name and the Serafim Tech mark may be used to refer to this framework and accurate adaptations of it.