Average readiness across 14,616 sites. Payment is the weakest published layer at 6% average fill.2
Prologue · the person is the workflow
You assemble the transaction.
You search, compare, judge, authorize, fill the form, track and resolve exceptions. Software helps, but you remain the commercial operating system.
Human-led commercesoftware owns 1/8 steps
Purchasing experience · plays as it enters the story
You assemble the purchase across pages
illustrative experience · today · illustrative, not a forecast
youintentsearch pagescommercial handoffseller pagescommercial handoffmerchant formcommercial handoffcommute readyoutcome
searchCycleWorksevidencecomparisonformstatus
Scene 1 of 6 · Search across sellers
discoverdecidetransactresolve
After the purchase · Observed 2026 market · before the scenario
The market is visible. Machine agency is not.
The economic anchor measures the whole online retail market, not agentic commerce. A person still converts attention into a transaction, so agent influence and execution remain deliberately unestimated here.26
Surfacesearch results, review tabs, product page and checkout form
Authorityfresh human consent at every consequential click
What this unlocks nextThe first displacement is attention: agents compress research before they receive any authority to spend.
Journey lens · before Act I
Which need should this future keep trying to solve?
Choose one need now. The person, agent and market will keep confronting the same stakes as the purchasing interface changes.
the model runs after each handoff
Neednot chosen yet
Authoritynot chosen yet
Marketnot chosen yet
Assurance + executionnot chosen yet
Act I · the machine learns the purchase
Discovery becomes delegation.
You state the outcome. The agent researches and explains, but you still choose and commit.
Agent-assisted commercesoftware owns 3/8 steps
Purchasing experience · plays as it enters the story
The agent turns intent into a shortlist
illustrative experience · recommendation era · illustrative, not a forecast
youintentbuyer agentcommercial handoffshortlistcommercial handoffCycleWorkscommercial handoffcommute readyoutcome
assistantresearchevidencemerchantcheckout
Scene 1 of 5 · Describe the outcome
discoverdecidetransactresolve
After the purchase · 2026Q4 · shared scenario state
Attention moves before money does.
The agent influences which product reaches the cart, but the near-term engine counts only transactions it actually carries. The dollar view below is an annual-equivalent translation at the observed 2026 market size—not a new market forecast.
Surfaceevidence-backed shortlist that hands back to a seller site
Authorityresearch and recommendation; the person still chooses and commits
What this unlocks nextTo move from recommendation to execution, intent must become a bounded mandate that a seller and payment rail can verify.
Why the model moves here · 2026Q3 → 2026Q4
2026Q3
The 402 era
Agents can find products but mostly cannot pay for them. In the starting scenario, six in ten attempts die on a missing instrument — the person never linked anything the agent could spend. Purchases that do settle concentrate among checkout-capable or agent-native sellers already exposing a machine-readable rail. The seller-capability mix is a disclosed prior, not an inference from a readiness score.17911
What a 402 actually says
An agent requests a product and the seller answers 402 Payment Required with a machine-readable challenge: the price, the rails it accepts, and where to send proof. The agent pays on any rail it can and retries the same request with that proof. No cart, no login wall, no form — the handshake is the checkout. Every failure on this page is one of five ways that handshake does not complete.
2026Q4
Card tokens arrive
Card-network agent tokens start clearing real orders: a tokenised credential bound to one agent, one merchant scope, one consent. Platform checkouts inside chat products pick up the first holiday volume. In the model, sellers with any rail climb from 9% to 25%, because integration cost and visible failed demand accelerate adoption.57
Choice 1 of 4 · before Act II
When the agent earns permission, who can carry it?
This choice travels into every later chapter. It changes whether authority is portable, collectively governed or captive to one runtime.
the model runs after each handoff
NeedEveryday mobility
AuthorityPortable personal mandate
Marketnot chosen yet
Assurance + executionnot chosen yet
Act II · the machine earns permission to act
Checkout becomes a protocol.
You grant a bounded mandate. The agent verifies the seller and terms, pays, stores the receipt and returns only for exceptions.
Delegated commercesoftware owns 6/8 steps
Purchasing experience · plays as it enters the story
One approval replaces the checkout session
illustrative experience · delegated purchase · illustrative, not a forecast
youintentbuyer agentcommercial handoffmerchant checkout capabilitycommercial handoffCycleWorkscommercial handoffcommute readyoutcome
mandateverificationapprovalsettlementreceipt
Scene 1 of 5 · Grant a bounded mandate
discoverdecidetransactresolve
After the purchase · 2027Q2 · last shared quarter
Authority becomes an economic variable.
A purchase can now move because software holds a one-use budget, compatible instrument and machine-readable receipt. Channel economics appear: agent-carried GMV can be separated from what reaches the merchant after the modeled take.
Surfacemandate, verification trace and one approval instead of checkout
Authorityone-use spend ceiling, required terms, expiry and explicit exceptions
What this unlocks nextOnce execution works, institutions compete to decide whether authority is portable, translated across seams or captive to a runtime.
Why the model moves here · 2027Q1 → 2027Q2
2027Q1
One in twenty
Success passes 7%. Merchants that publish machine-readable catalogs and a checkout API see agents complete orders; the rest see agents leave. Failed attempts become visible in server logs as a stream of challenges nobody answered, turning lost machine demand into an adoption signal.47
Why seller capability gates the curve
The engine multiplies each reachable attempt by a disclosed completion assumption: 15% for a page-only seller, 45% with machine-readable discovery but no dependable checkout, 80% with programmatic checkout and 100% at the agent-native ceiling. The starting distribution across those states is a wide scenario prior—not inferred from a readiness score. Sellers move upward as lost agent demand makes investment worthwhile.
2027Q2
Rails stop being exotic
Four in ten sellers expose a rail and one in eight buyers holds a wallet. UCP still carries the plurality of settled agent orders, but x402 and MPP together clear more than four in ten; card tokens and platform checkout carry the rest. This is the final shared quarter before market structure changes the rail mix.347
Three market structures, mechanically
A wallet resolves a 402 on x402 or MPP; a card token resolves it on the card networks or UCP; a platform account resolves it only inside that platform's checkout. Coordinated open rails bridge those families, fragmented openness leaves partial adapters, and the walled scenario lets runtimes drop outside rails nearly half the time. Those policy and cost levers are the only values that differ.
Your path so far · 2027Q3
The first choices now happen inside the story.
Agent purchases approach 1% of online GMV. That is the moment the platforms that host the agents decide whether a buyer's instrument is theirs or the buyer's — and whether open protocols coordinate or merely coexist. The simulation runs three answers from here with every parameter shared except nine lever values: interoperability, runtime gatekeeping, take rate, the relative cost of each rail, and which rail the incumbents push.68
Choice 2 of 4 · threshold to the structural future
Where should delegated demand meet supply?
Choose the market formation that grows from the bounded purchase. The next page carries this choice forward and lets you complete assurance and execution.
the model runs after each handoff
NeedEveryday mobility
AuthorityPortable personal mandate
MarketCallable merchants
Assurance + executionnot chosen yet
Compare the three institutional benchmark lenses
These remain useful reference regimes, but they no longer stand in for the full choice space.
Scenario A
Coordinated open rails
Any agent, any instrument, any storefront.
24%settle22%of GMV34%hold a wallet
Scenario B
Fragmented open web
The protocols stay open; the market still has to assemble them.
21%settle19%of GMV26%hold a wallet
Scenario C
Walled gardens
Agents buy reliably — inside the garden that hosts them.
17%settle15%of GMV19%hold a wallet
Open the 2026–2029 model audit and branch comparison
Forecast audit · what actually moves the curve
A story you can try to break.
The ribbons now come from 4,096 seeded Latin-hypercube runs, so every assumed range is covered densely and evenly. This low→high view moves one assumption at a time and holds the rest at the published coordinated-open defaults; it is sensitivity, not a probability.
What creates the open-vs-walled merchant outcome gap
Starting from the walled ending, each bar switches one condition to its open value and measures merchant-retained online GMV in 2029Q2. The effects interact, so they do not add up.
What ecosystem coordination adds beyond open protocols
Starting from the fragmented-open scenario, each bar switches one market condition to the coordinated-open value. The result measures a system capability, not the value or market share of any present company.
Portable authorizationStandards, networks, runtimes or independent routers could bridge a buyer instrument to more than one seller rail.
Shared seller surfacePlatforms or neutral infrastructure could make one integration reachable by many kinds of buying agent.
Visible failed demandMarket-wide capability and settlement signals could turn abandoned attempts into an adoption signal instead of leaving them invisible.
ZeroClick is deliberately outside this causal loop. After the simulation, these unmet system needs can inform what it might build, partner on, replace or stop doing; its current rails are not projected forward as the answer.
Rerun a custom counterfactual in the scenario lab →CoordinatedFragmentedWalled
Ribbons are the 10th–90th percentile across 4,096 seeded Latin-hypercube runs spanning every stated assumption range. Coordinated open rails reach 22% by 2029Q2; fragmented openness shows what protocols achieve without ubiquitous coordination, while the walled scenario shows the cost of proprietary runtime control.67