zeroclick. / futures
Acts III–V · a path-built story of agentic commerce · 2029 → 2040

Commerce becomes an operating layer.

The buyer has learned to delegate. Now intent can travel through merchant capabilities, mandates, exchanges, pooled demand, commissioned supply, trust markets and physical systems—without assuming every seller becomes an agent.

Read the future in one pass. Build a path through four consequential forks, follow one need through four purchasing experiences, then expand the economic, protocol and agent-ecology layers.

How to read this simulation

The simulation begins with the bounded 2029 model, then asks what structural change makes the next commercial behavior possible.13 The resulting chapters are conditional narratives, not point predictions. Delegation is not treated as a ladder toward maximum autonomy: it remains bounded by stakes, reversibility and the work of articulating human preferences.1415161718 Sensitivity widens with time.

Act I · 2026 → 2027The buyer delegatesSoftware moves from ranking choices to receiving bounded authority.
Act II · 2027 → 2029 · institutions decide who can transact
Lens AOpen railsauthority becomes portableLens BFragmented webtranslation becomes the taxLens CWalled gardensdelegation becomes captive
Acts III–V · 2029 → 2040Commerce becomes persistentSeller agents, policies and physical execution turn purchases into an operating system.
One need · one readable arc

Buying stops being one kind of event.

The shift is not an agent clicking today’s checkout faster. Intent becomes portable authority, businesses become callable, specialized agents coordinate the market and purchases become policies that can reach the physical world.

The human does not disappear. Attention moves from assembling transactions to setting boundaries, reviewing exceptions and revoking authority. Read the purchasing story first; expand the economic, protocol and ecology layers afterward.

Build the path

Choose what becomes true—then live through the result.

Authority, market formation, assurance and execution are independent forks. The selected ski goggles need stays constant so the changing purchasing experience remains visible.

Build the path · four forks · 144 combinations

The future is a recipe, not a regime.

Choose one answer at each dated fork. The server maps the combination onto all nine structural assumptions, derives the closest institutional benchmark and reruns every purchasing chapter.

choices = authored scenario semantics
state trajectory = deterministic model
2029 · fork 1

Who can carry durable purchasing authority?

2032 · fork 2

Where does delegated demand meet supply?

2036 · fork 3

Who guarantees the outcome when agents are wrong?

2040 · fork 4

How does commercial intent act in the world?

reader-selected scenario recipe mapped to disclosed structural assumptions · shareable URL
Starting recipe preview · simulate to apply · Walled gardens is the nearest benchmark

Agent super-app empires

Integrated recourse improves while switching and cross-market reach shrink. The storefront survives as a business boundary even when its pages disappear. Trust is legible and concentrated in the operator that can remove access. Checkout becomes policy maintenance and the person moves to supervision.

2029 · authorityRuntime-bound authority
2032 · marketCallable merchants
2036 · assurancePlatform guarantee
2040 · executionStanding outcome policies
machine counterparties11%
continuous commerce3%
pooled demand2%
physical execution10%

Binding tension: Integrated guarantees compete with portability, switching and independent seller access.

Compare three institutional benchmarks—not the full possibility space
Published path · structural model

Walled gardens

Platform tokens are exclusive. Chat runtimes block third-party rails almost half the time, take rates run at a tenth or more, and the open protocols cost a merchant three times as much to integrate as the platform's own checkout. Agents still grow — 15% of online GMV by 2029Q2 — but on cards and platform checkout, with wallets stuck below a fifth of buyers and cross-garden purchases a coin flip.

2040 machine counterparties16%
continuous commerce4%
published 2040 p10–p908%–25%512 structural draws
The whole path in four changes

One need moves from request to operating policy.

story first · model layers after

Conditional annual ranges overlap and are not valuations. Role classes are functional activations above the disclosed threshold—not agent population or forecast headcount.

2029
assisted commerce · model-triggered authored event

Distribution turns checkout into a platform boundary.

Integrated runtimes make agent checkout coherent inside their own marketplaces while third-party reach remains conditional.

2029
Interface evolution · adaptive agent store · illustrative Ski goggles purchase · Walled gardens

One intent, one approval

State the goal, inspect the evidence and approve this purchase.

adaptive agent store · approval bundle · engine selected
Agent influenced$235B–$270B
Agent executed$76.7B–$119B
Merchant-routed after tolls + loss$63.0B–$98.5B
✦Your commerce agentSki goggles · Walled gardens · 2029
illustrative experience
Scene 1 of 4 · Intent wakes the agent

The purchase begins before a storefront appears.

Find ski goggles for my helmet. I ski in sunshine and flat light, so include a spare low-light lens. Keep the full order under $220.
I’ll act inside your single-use mandate, preserve Ask before changing frame fit, lens type or spending limit. and bring back only a meaningful decision.
Scene 1 of 4 · Intent wakes the agent
Open the model beneath this experienceeconomics · trust · failure · physical handoff
In the physical world · illustrative physical-world experience

Try the goggles with your helmet.

2029 · Walled gardens
managed marketplace
Site referral 83%
What the person experiencesThe visible moment

The skier compares the sun and low-light lenses, then checks the frame against their helmet.

What software does off-screenThe delegated work

Compare fit information, lens options, total price and delivery.

What changes in the worldThe physical handoff

Thursday delivery · charcoal frame, rose sun lens and amber low-light spare. Try the fit with your helmet before removing the return tags.

Control returns hereAsk before changing frame fit, lens type or spending limit.
What happens next · model-derived interpretation

The visible change is compression: product search, seller choice, service, payment and delivery arrive as one proposal. Human attention concentrates at the moment of authorization. At the 2029 handoff, 16% of demand is delegated while only 1% has a machine on both sides. The experience therefore begins as an approval bundle.

What remains contested: The managed marketplace makes routine execution coherent, but a runtime-bound account concentrates ranking, appeal and exit inside the runtime.

Active machine cast · model-derivedBuyer agent · Mandate agent · Settlement agent
Coordination burden3%range 2%–3%
Human roleoperatorowns multi-agent drift
Between purchasing chapters · model-derived experience transition

The structural handoff begins as an approval bundle.

2029Q2 → 2029
Interfaceapproval bundlestructural starting state
Human attentionapprove this purchasestructural starting state
Agent mandatesingle-use mandateautonomy starts at 5%
Market boundary46% reachablereach starts at 46% · exception risk starts at 30%
Model trace · why this chapter changes
Model trigger · branch-specific event at the near-term handoff15.5% of demand is delegated, while seller-agent adoption begins at 7.7% after the 2029Q2 handoff.Authored consequence: Merchant access to delegated demand now depends on platform distribution and ranking policy.
Event sensitivity · branch-fixed handoff event
Distribution turns checkout into a platform boundary.100%

512 range-covering structural draws · authored event selection coverage, not probability. Distribution turns checkout into a platform boundary. 100%

Model competence20%Can the agent reason through the commercial task?
Harness reliability18%Can tools, retries and checks finish the task safely?
Memory continuity6%Can preferences and outcomes persist with user control?
delegated demand16%
machine on both sides1%
continuous commerce2%
physical execution0%
pooled demand0%
adaptive supply0%
governance capacity37%
adversarial loss9%
2029Q2 → structural 2029model-derived transition · equations, not observed attribution
delegated demandThe structural layer carries the branch’s final agent-GMV share into its starting delegated-demand state.16%near-term agent GMV 16% → 16%
seller-agent footholdA disclosed 12% handoff factor converts rail-capable sellers into the initial seller-agent foothold.8%sellers answering a rail 65% → 8%
governance capacityPortable authority, liability clarity and branch coordination are averaged into the initial governance capacity.37%structural priors
physical execution footholdA disclosed handoff factor seeds the small share of agent demand that can already cause and verify physical execution.0%near-term agent GMV + structural prior 16% → 0%
pooled demand footholdDelegated demand and the pooled-mandate prior seed the initial ability to bargain across more than one buyer.0%agent GMV + pooled-mandate prior 16% → 0%
adaptive supply footholdA disclosed handoff factor seeds supply that can be configured from machine-readable demand rather than selected from a fixed catalog.0%seller-agent foothold
adversarial lossThe failed-attempt share and adversarial-pressure prior seed loss pressure; this is a structural mapping, not observed loss.9%failed attempts + pressure prior 83% → 9%

Arena lens: For ski goggles, repeatability 30%, reversibility 70%, consequence 45% and preference-articulation load 65% produce 26% modeled delegation fit. These are unanchored scenario priors, not measurements.

share of demand delegated starts at 15.5%; seller-agent adoption starts at 7.7%; share with machine counterparties starts at 1.2%; share negotiated starts at 1.2%; share run continuously starts at 1.6%.

No new actor class enters yet; adoption and institutional state drive this transition.

modeled from unanchored structural inputs · 17 published research hooks · ensemble ranges are modeled sensitivity, not measured probability

2032
assisted commerce · model-triggered authored event

Authority remains attached to the runtime account.

Runtime accounts become the default container for spending authority, commercial history and the platform guarantee.

2032
Interface evolution · live seller-presence market · illustrative Ski goggles purchase · Walled gardens

One intent, one approval

State the goal, inspect the evidence and approve this purchase.

live seller-presence market · approval bundle · engine selected
Agent influenced$408B–$666B
Agent executed$148B–$308B
Merchant-routed after tolls + loss$124B–$259B
✦Your commerce agentSki goggles · Walled gardens · 2032
illustrative experience
Scene 1 of 4 · Intent wakes the agent

The purchase begins before a storefront appears.

Compare complete goggle offers under $220, with helmet fit and a low-light spare preserved.
I’ll act inside your single-use mandate, preserve Ask before changing frame fit, lens type or spending limit. and bring back only a meaningful decision.
Scene 1 of 4 · Intent wakes the agent
Open the model beneath this experienceeconomics · trust · failure · physical handoff
In the physical world · illustrative physical-world experience

A complete lens bundle reaches the doorstep.

2032 · Walled gardens
managed marketplace
Site referral 72%
What the person experiencesThe visible moment

The skier approves a complete bundle with delivery and fit returns included.

What software does off-screenThe delegated work

Negotiate full-set pricing while keeping each skier’s fit constraints.

What changes in the worldThe physical handoff

Thursday delivery · charcoal frame, rose sun lens and amber low-light spare. Try the fit with your helmet before removing the return tags.

Control returns hereAsk before changing frame fit, lens type or spending limit.
What happens next · model-derived interpretation

The visible change is compression: product search, seller choice, service, payment and delivery arrive as one proposal. Human attention concentrates at the moment of authorization. Since 2029, machine-counterparty share rises by 3.3 percentage points and continuous commerce falls by 0.3 percentage points. The experience remains an approval bundle, but its mandate and market scope deepen.

What remains contested: The managed marketplace makes routine execution coherent, but a runtime-bound account concentrates ranking, appeal and exit inside the runtime.

Active machine cast · model-derivedSeller agent · Buyer agent · Mandate agent · Settlement agent
Coordination burden4%range 2%–4%
Human roleoperatorowns multi-agent drift
Between purchasing chapters · model-derived experience transition

approval bundle remains visible while its mandate deepens.

2029 → 2032
Interfaceapproval bundleunchanged from approval bundle
Human attentionapprove this purchaseunchanged from approve this purchase
Agent mandatesingle-use mandateautonomy +3.3 pp to 8%
Market boundary46% reachablereach -0.0 pp to 46% · exception risk -0.0 pp to 30%
Model trace · why this chapter changes
Model trigger · identity portability < 35.0%Identity portability is 18.0% and machine counterparties reach 4.5%.Authored consequence: The guarantee reduces transaction friction while making suspension, appeal and exit economically consequential.
Event sensitivity · selected 31% · leading alternate 38%
Authority remains attached to the runtime account.31%
Platforms license selective credential bridges.38%
Regulation forces portable mandates out of the garden.31%

512 range-covering structural draws · authored event selection coverage, not probability. Authority remains attached to the runtime account. 31% · Platforms license selective credential bridges. 38% · Regulation forces portable mandates out of the garden. 31%

Model competence50%Can the agent reason through the commercial task?
Harness reliability32%Can tools, retries and checks finish the task safely?
Memory continuity15%Can preferences and outcomes persist with user control?
delegated demand24%
machine on both sides4%
continuous commerce1%
physical execution5%
pooled demand1%
adaptive supply3%
governance capacity37%
adversarial loss9%
2029 → 2032model-derived transition · equations, not observed attribution
delegated demandCommercial capability closes part of the gap toward the governance- and loss-constrained trust ceiling.+8.9 pp2029 16% → now 24%
seller-agent adoptionSeller-side diffusion follows the demand agents can address and the selected annual diffusion prior.+25.5 pp2029 8% → now 33%
machine counterpartiesBuyer and seller delegation meet at a rate shaped by persistent-agent cost and branch coordination.+3.3 pp2029 1% → now 4%
continuous commerceDelegated demand, machine counterparties and governance turn isolated transactions into standing policies.−0.3 pp2029 2% → now 1%
physical executionAgent-addressable operators diffuse toward a ceiling set by seller agents, machine counterparties and governance capacity.+5.2 pp2029 0% → now 5%
pooled demandPortable pooled mandates, branch coordination and market concentration determine how much delegated demand can bargain collectively.+1.0 pp2029 0% → now 1%
adaptive supplySeller agents, negotiated transactions and physical execution let supply respond to intent before a finished item exists.+2.8 pp2029 0% → now 3%
adversarial lossFraud and manipulation pressure compete with governance capacity as machine counterparties expand.−0.0 pp2029 9% → now 9%

Arena lens: For ski goggles, repeatability 30%, reversibility 70%, consequence 45% and preference-articulation load 65% produce 26% modeled delegation fit. These are unanchored scenario priors, not measurements.

share of demand delegated rises to 24.4% (+8.9%); seller-agent adoption rises to 33.3% (+25.5%); share with machine counterparties rises to 4.5% (+3.3%); share negotiated rises to 2.1% (+0.8%); share run continuously falls to 1.2% (-0.3%).

New modeled actors: Merchant capability — exposes discovery, checkout, fulfillment and support as callable operations without needing a human-facing storefront; Seller agent — sets price, inventory exposure, service terms and which buyers to trust; Authority network — proves what an agent may spend, disclose, sign and delegate onward; Open commerce router — routes portable intent across catalogs, logistics and payment providers without owning the buyer relationship.

modeled from unanchored structural inputs · 17 published research hooks · ensemble ranges are modeled sensitivity, not measured probability

2036
delegated commerce · model-triggered authored event

Platform guarantees keep approval in the loop.

Buyer and seller agents negotiate inside platform-run markets whose operator controls access, ranking and acceptable terms.

2036
Interface evolution · policy + exception surface · illustrative Ski goggles purchase · Walled gardens

Terms become negotiable

Set a bargaining envelope and review only counteroffers that cross a boundary.

policy + exception surface · negotiation room · engine selected
Agent influenced$650B–$1.3T
Agent executed$259B–$647B
Merchant-routed after tolls + loss$212B–$564B
✦Your commerce agentSki goggles · Walled gardens · 2036
illustrative experience
Scene 1 of 4 · Intent wakes the agent

The purchase begins before a storefront appears.

Keep the goggle set ready for the season within an $85 replacement-lens budget.
I’ll act inside your bounded negotiation mandate, preserve Ask before changing frame fit, lens type or spending limit. and bring back only a meaningful decision.
Scene 1 of 4 · Intent wakes the agent
Open the model beneath this experienceeconomics · trust · failure · physical handoff
In the physical world · illustrative physical-world experience

The lens set is ready before the trip.

2036 · Walled gardens
managed marketplace
Site referral 59%
What the person experiencesThe visible moment

The skier reviews the season’s remaining replacement budget.

What software does off-screenThe delegated work

Watch equipment condition and supplier availability within permission.

What changes in the worldThe physical handoff

Alpine Optics · compatible amber replacement lens · $55 delivered before the trip

Control returns hereAsk before changing frame fit, lens type or spending limit.
What happens next · model-derived interpretation

The cart stops being the center of the transaction. Buyer and seller agents exchange terms inside a mandate, while the person sees the boundary of the bargain rather than every offer. Since 2032, machine-counterparty share rises by 6.3 percentage points and continuous commerce rises by 0.8 percentage points. That state transition moves the purchasing interface from approval bundle to negotiation room.

What remains contested: The managed marketplace makes routine execution coherent, but a runtime-bound account concentrates ranking, appeal and exit inside the runtime.

Active machine cast · model-derivedSeller agent · Buyer agent · Mandate agent · Settlement agent · Physical operator · Routing agent · Market operator · Supply agent
Coordination burden6%range 3%–8%
Human roleapprover and recourse ownerowns multi-agent drift
Between purchasing chapters · model-derived experience transition

The interface moves from approval bundle to negotiation room.

2032 → 2036
Interfacenegotiation roomfrom approval bundle
Human attentionreview boundary crossingsfrom approve this purchase
Agent mandatebounded negotiation mandateautonomy +4.4 pp to 13%
Market boundary46% reachablereach -0.1 pp to 46% · exception risk +0.9 pp to 31%
Model trace · why this chapter changes
Model trigger · negotiated transactions < 6.0%4.7% of transactions are negotiated and seller-agent adoption reaches 52.7%.Authored consequence: Negotiation becomes efficient inside the garden, but the operator can observe and shape both sides of price formation.
Event sensitivity · selected 80% · leading alternate 20%
Platform guarantees keep approval in the loop.80%
Runtime agents bargain inside captive markets.20%
Platform auctions internalize machine negotiation.0%

512 range-covering structural draws · authored event selection coverage, not probability. Platform guarantees keep approval in the loop. 80% · Runtime agents bargain inside captive markets. 20% · Platform auctions internalize machine negotiation. 0%

Model competence74%Can the agent reason through the commercial task?
Harness reliability47%Can tools, retries and checks finish the task safely?
Memory continuity24%Can preferences and outcomes persist with user control?
delegated demand33%
machine on both sides11%
continuous commerce2%
physical execution15%
pooled demand2%
adaptive supply9%
governance capacity37%
adversarial loss10%
2032 → 2036model-derived transition · equations, not observed attribution
delegated demandCommercial capability closes part of the gap toward the governance- and loss-constrained trust ceiling.+8.9 pp2032 24% → now 33%
seller-agent adoptionSeller-side diffusion follows the demand agents can address and the selected annual diffusion prior.+19.4 pp2032 33% → now 53%
machine counterpartiesBuyer and seller delegation meet at a rate shaped by persistent-agent cost and branch coordination.+6.3 pp2032 4% → now 11%
continuous commerceDelegated demand, machine counterparties and governance turn isolated transactions into standing policies.+0.8 pp2032 1% → now 2%
physical executionAgent-addressable operators diffuse toward a ceiling set by seller agents, machine counterparties and governance capacity.+9.7 pp2032 5% → now 15%
pooled demandPortable pooled mandates, branch coordination and market concentration determine how much delegated demand can bargain collectively.+0.9 pp2032 1% → now 2%
adaptive supplySeller agents, negotiated transactions and physical execution let supply respond to intent before a finished item exists.+5.8 pp2032 3% → now 9%
adversarial lossFraud and manipulation pressure compete with governance capacity as machine counterparties expand.+0.9 pp2032 9% → now 10%

Arena lens: For ski goggles, repeatability 30%, reversibility 70%, consequence 45% and preference-articulation load 65% produce 26% modeled delegation fit. These are unanchored scenario priors, not measurements.

share of demand delegated rises to 33.4% (+8.9%); seller-agent adoption rises to 52.7% (+19.4%); share with machine counterparties rises to 10.8% (+6.3%); share negotiated rises to 4.7% (+2.6%); share run continuously rises to 2.0% (+0.8%).

New modeled actors: Agent market operator — matches machine demand and supply while setting access and ranking rules; Assurance network — underwrites disputes, provenance, delivery and adversarial behavior; Physical operator agent — commits robots, vehicles, workers and facilities to machine-negotiated physical outcomes; Metered capability — quotes and settles small units of digital or physical work as they are consumed; Trust and escrow registry — makes identity, reputation, job state and conditional release portable between unfamiliar counterparties; Demand coalition — pools bounded mandates so households or organizations bargain without surrendering individual constraints.

modeled from unanchored structural inputs · 17 published research hooks · ensemble ranges are modeled sensitivity, not measured probability

2040
delegated commerce · model-triggered authored event

Persistent agents remain bounded by platform guarantees.

Persistent runtime agents manage recurring needs through captive accounts, guarantees and preferred seller capacity.

2040
Interface evolution · ambient agent surface · illustrative Ski goggles purchase · Walled gardens

Terms become negotiable

Set a bargaining envelope and review only counteroffers that cross a boundary.

ambient agent surface · negotiation room · engine selected
Agent influenced$909B–$2.0T
Agent executed$396B–$1.1T
Merchant-routed after tolls + loss$316B–$932B
✦Your commerce agentSki goggles · Walled gardens · 2040
illustrative experience
Scene 1 of 4 · Intent wakes the agent

The purchase begins before a storefront appears.

Use opted-in equipment condition and trip plans to prepare compatible lenses; ask before changing fit or budget.
I’ll act inside your bounded negotiation mandate, preserve Ask before changing frame fit, lens type or spending limit. and bring back only a meaningful decision.
Scene 1 of 4 · Intent wakes the agent
Open the model beneath this experienceeconomics · trust · failure · physical handoff
In the physical world · illustrative physical-world experience

An equipment signal brings the choice forward.

2040 · Walled gardens
managed marketplace
Site referral 50%
What the person experiencesThe visible moment

The skier sees a compatible replacement proposal before packing.

What software does off-screenThe delegated work

Connect an opted-in equipment signal with a compatible replacement offer.

What changes in the worldThe physical handoff

Alpine Optics · compatible amber replacement lens · $55 delivered before the trip

Control returns hereAsk before changing frame fit, lens type or spending limit.
What happens next · model-derived interpretation

The cart stops being the center of the transaction. Buyer and seller agents exchange terms inside a mandate, while the person sees the boundary of the bargain rather than every offer. Since 2036, machine-counterparty share rises by 5.6 percentage points and continuous commerce rises by 1.6 percentage points. The experience remains a negotiation room, but its mandate and market scope deepen.

What remains contested: The managed marketplace makes routine execution coherent, but a runtime-bound account concentrates ranking, appeal and exit inside the runtime.

Active machine cast · model-derivedSeller agent · Buyer agent · Mandate agent · Settlement agent · Physical operator · Routing agent · Market operator · Supply agent · Assurance agent
Coordination burden9%range 5%–10%
Human roleapprover and recourse ownerowns coordination capture
Between purchasing chapters · model-derived experience transition

negotiation room remains visible while its mandate deepens.

2036 → 2040
Interfacenegotiation roomunchanged from negotiation room
Human attentionreview boundary crossingsunchanged from review boundary crossings
Agent mandatebounded negotiation mandateautonomy +3.7 pp to 17%
Market boundary45% reachablereach -0.1 pp to 45% · exception risk +1.2 pp to 32%
Model trace · why this chapter changes
Model trigger · continuous commerce < 12.0%3.6% of commerce runs continuously while market concentration is 78.8%.Authored consequence: Execution feels continuous to the buyer while switching costs accumulate in identity, history, reserves and redress.
Event sensitivity · selected 98% · leading alternate 2%
Persistent agents remain bounded by platform guarantees.98%
Runtime subscriptions absorb the checkout session.2%
Platform reserves become private outcome markets.0%

512 range-covering structural draws · authored event selection coverage, not probability. Persistent agents remain bounded by platform guarantees. 98% · Runtime subscriptions absorb the checkout session. 2% · Platform reserves become private outcome markets. 0%

Model competence86%Can the agent reason through the commercial task?
Harness reliability58%Can tools, retries and checks finish the task safely?
Memory continuity30%Can preferences and outcomes persist with user control?
delegated demand40%
machine on both sides16%
continuous commerce4%
physical execution23%
pooled demand3%
adaptive supply14%
governance capacity37%
adversarial loss11%
2036 → 2040model-derived transition · equations, not observed attribution
delegated demandCommercial capability closes part of the gap toward the governance- and loss-constrained trust ceiling.+6.4 pp2036 33% → now 40%
seller-agent adoptionSeller-side diffusion follows the demand agents can address and the selected annual diffusion prior.+7.9 pp2036 53% → now 61%
machine counterpartiesBuyer and seller delegation meet at a rate shaped by persistent-agent cost and branch coordination.+5.6 pp2036 11% → now 16%
continuous commerceDelegated demand, machine counterparties and governance turn isolated transactions into standing policies.+1.6 pp2036 2% → now 4%
physical executionAgent-addressable operators diffuse toward a ceiling set by seller agents, machine counterparties and governance capacity.+7.6 pp2036 15% → now 23%
pooled demandPortable pooled mandates, branch coordination and market concentration determine how much delegated demand can bargain collectively.+0.7 pp2036 2% → now 3%
adaptive supplySeller agents, negotiated transactions and physical execution let supply respond to intent before a finished item exists.+5.2 pp2036 9% → now 14%
adversarial lossFraud and manipulation pressure compete with governance capacity as machine counterparties expand.+1.2 pp2036 10% → now 11%

Arena lens: For ski goggles, repeatability 30%, reversibility 70%, consequence 45% and preference-articulation load 65% produce 26% modeled delegation fit. These are unanchored scenario priors, not measurements.

share of demand delegated rises to 39.7% (+6.4%); seller-agent adoption rises to 60.7% (+7.9%); share with machine counterparties rises to 16.3% (+5.6%); share negotiated rises to 7.9% (+3.1%); share run continuously rises to 3.6% (+1.6%).

New modeled actors: Resource agent — continuously allocates money, energy, compute, inventory or capacity against a policy; Asset twin — turns a machine, package, building or product lifecycle into an authenticated source of commercial intent; Production agent — configures or commissions supply after machine demand forms, before a finished product necessarily exists.

modeled from unanchored structural inputs · 17 published research hooks · ensemble ranges are modeled sensitivity, not measured probability

Expand the forces behind this futureeconomic envelope · six protocol paths · full agent ecology
Economic envelope · annual U.S. retail e-commerce flows

A dollar can be influenced, executed, retained—or captured as infrastructure.

The engine separates commerce touched by an agent from transactions actually committed by one, then estimates where value can be retained, tolled, lost or created as machine-payable work. Every range moves with this branch’s structural draws plus five disclosed economic priors.

512 paired range-covering draws
one observed 2026 anchor
not a valuation or probability
agent influencedagent executedmerchant-routed after channel value + loss
2029$1.6T
addressable baseline
Agent influenced GMV
$235B–$270Bdiscovery or decision shaped
Agent executed GMV
$76.7B–$119Bauthority reaches commitment
Merchant-routed value
$63.0B–$98.5Bafter modeled channel tolls + loss; not profit
Agent-channel value
$4.4B–$11.1Baccess, orchestration + distribution pool
Risk leakage
$4.8B–$13.2Bfraud, dispute, failure + exception loss
Machine-service revenue
$137M–$496Mnew metered work; may overlap other flows
2032$1.9T
addressable baseline
Agent influenced GMV
$408B–$666Bdiscovery or decision shaped
Agent executed GMV
$148B–$308Bauthority reaches commitment
Merchant-routed value
$124B–$259Bafter modeled channel tolls + loss; not profit
Agent-channel value
$9.4B–$27.2Baccess, orchestration + distribution pool
Risk leakage
$8.4B–$31.2Bfraud, dispute, failure + exception loss
Machine-service revenue
$306M–$1.3Bnew metered work; may overlap other flows
2036$2.4T
addressable baseline
Agent influenced GMV
$650B–$1.3Tdiscovery or decision shaped
Agent executed GMV
$259B–$647Bauthority reaches commitment
Merchant-routed value
$212B–$564Bafter modeled channel tolls + loss; not profit
Agent-channel value
$17.0B–$55.6Baccess, orchestration + distribution pool
Risk leakage
$12.6B–$71.2Bfraud, dispute, failure + exception loss
Machine-service revenue
$674M–$3.5Bnew metered work; may overlap other flows
2040$3.1T
addressable baseline
Agent influenced GMV
$909B–$2.0Tdiscovery or decision shaped
Agent executed GMV
$396B–$1.1Tauthority reaches commitment
Merchant-routed value
$316B–$932Bafter modeled channel tolls + loss; not profit
Agent-channel value
$25.5B–$92.2Baccess, orchestration + distribution pool
Risk leakage
$20.1B–$130Bfraud, dispute, failure + exception loss
Machine-service revenue
$1.2B–$7.0Bnew metered work; may overlap other flows
Observed anchor. $1.4T annualized from Census Q2 2026 seasonally adjusted U.S. retail e-commerce sales; future growth is not observed.26External comparison, not calibration. McKinsey estimated $900B–$1T of U.S. B2C retail revenue could be orchestrated by agentic commerce in 2030. Its definition and horizon differ from these engine flows.27 Overlapping values above must not be added.
Agentic commerce paths · 2029 → 2040

Six futures can grow from the same need.

These are paths, not end-state options. Choose a mechanism, then move through the years to see what must become true and how purchasing changes before the 2040 destination. Seller-to-seller negotiation is only one possible mechanism. Replay this ski goggles need through embedded merchant capabilities, portable mandates, open commerce networks, metered services, trust markets or object-initiated coordination.

four stages per path
activation signals can coexist
experiences = illustrative
From the 2029 handoff to 2040

How this future becomes possible.

Follow the mechanism through bridge, formation, scale and infrastructure. Activation is model-derived support; milestones and purchasing moments are conditional narration.

largest modeled acceleration · 2032
conditional protocol-future path, not a forecast
bridge · 2029 · conditional path
Stores ship their first bounded agents

A merchant-native shopper agent joins search, product reasoning, cart and care in one context while typed tools keep price, availability and policy grounded in store systems.

What purchasing feels like hereThe person asks the store for an outcome, sees a progressively rendered comparison and cart, then applies the prepared checkout from inside the conversation.
Conditional direct annual value pool$2.5B$767M–$5.9B sensitivity · based on agent-executed GMV at a 2.5% scenario take. Non-additive.
What must become true nextThe harness grounds every claim, renders only server-issued records and keeps payment behind a real approval surface.
How this path can stallThe agent becomes a polished chat layer over missing inventory, weak policy data or manipulative ranking.
walled institutionThis path scales fastest inside integrated platforms; the experience is coherent, but identity, evidence and exit remain captive to the runtime.
Range sensitivity16% of 512 draws select this as the strongest mechanism at 2029; not a probability.
2040 destination · complete illustrative transaction
2040 · Ski gogglesmerchant-owned agent harness
A different purchasing world · illustrative protocol-enabled purchasing experience

Every store becomes an agent

“Use opted-in equipment condition and trip plans to prepare compatible lenses; ask before changing fit or budget.”
A merchant-native shopper agent opens a live ski goggles workspace, remembers the relationship and renders product, comparison, delivery and approval surfaces from authoritative store systems.

YouReturn to the relationshipDescribe the ski goggles result; inspect which preferences the store remembered and correct any stale ones.
AIPlan with one contextUse search, research, planning, cart and care skills without handing the conversation across lossy domain agents.
BizRender authoritative UITyped tools return server-validated product, comparison, cart, order and policy components.
YouApply the staged purchaseThe agent may propose and prepare; a person or explicit policy applies the consequential commitment.
BizOperate and careThe merchant agent watches inventory and service, stages operational changes and keeps the seller accountable.
Physical resultAlpine Optics · compatible amber replacement lens · $55 delivered before the trip

What the person gets:The shopper agent completes a merchant-owned order and stays for service, while the merchant agent watches inventory, pricing and operations behind a separate approval boundary.

From one assistant to an agent ecology

More agents create a new coordination problem.

A purchase gains specialized role classes as authority, market access and physical execution spread. Activation means the structural state can support that function in a representative purchase—it is not an agent count, population forecast or claim that every purchase uses every role.

active threshold 8%
model-derived role activation and coordination sensitivity, not agent population or forecast headcount
20293 role classes
range 3–3
Buyer agent16%
Mandate agent10%
Settlement agent10%
coordination depth
1%range 1%–1%
human becomes
operator
binding risk
multi-agent drift
20324 role classes
range 4–5
Seller agent33%
Buyer agent24%
Mandate agent16%
Settlement agent15%
coordination depth
3%range 2%–4%
human becomes
operator
binding risk
multi-agent drift
20368 role classes
range 5–9
Seller agent53%
Buyer agent33%
Mandate agent22%
Settlement agent21%
Physical operator15%
Routing agent12%
+2 additional active role classes
coordination depth
7%range 4%–10%
human becomes
approver and recourse owner
binding risk
multi-agent drift
20409 role classes
range 6–10
Seller agent61%
Buyer agent40%
Mandate agent26%
Settlement agent25%
Physical operator23%
Routing agent17%
+3 additional active role classes
coordination depth
11%range 6%–16%
human becomes
approver and recourse owner
binding risk
coordination capture

The forecast is not “more bots.” It is a shift from one assistant operating a human workflow to a governed ecology of principals, counterparties, infrastructure, collectives and physical operators. The burden rises whenever coordination outruns authority, evidence and recourse.

Inspect alternate interaction forms and delegation boundaries
Beyond the checkout ladder · model-derived interaction activation

Eight forms can coexist—and diverge.

These are independent activation signals from the same structural state, not stages, market shares or probabilities. A seller site can persist while pooled demand, commissioned supply or physical orchestration grows beside it.

strongest form outlined
physical implications authored
interaction form2029203220362040
Site referralPhysical fulfillment remains downstream of a conventional order.
83%
72%
59%
50%
Delegated checkoutThe seller fulfills an agent-authored order rather than a person-built cart.
13%
20%
25%
28%
Bilateral agent tradeInventory, service capacity and delivery can be reserved machine-to-machine.
1%
3%
8%
13%
Terms marketThe physical plan changes before checkout because timing and capacity are part of the bargain.
1%
1%
3%
6%
Standing outcome policySupply, repair, transport or care coordination continue without a new shopping session.
2%
1%
2%
4%
Collective procurementNeighborhoods, fleets or firms reshape routes, inventory and capacity through coordinated demand.
0%
1%
2%
3%
Commissioned supplyProduction schedules, components and facilities respond to intent rather than forecasted shelf demand.
0%
3%
9%
14%
Embodied orchestrationThe purchasing interface becomes a live map of actions, safety boundaries and revocable authority.
0%
5%
15%
23%
Delegation frontier · published path

Where autonomy stops depends on the need.

The aggregate market state is identical down each column. Only the disclosed arena priors change. Each cell reports its preset surface and the full interface spread across 512 range-covering structural draws; segment widths are sensitivity coverage, not probability.

engine-selected interface + full range coverage · unanchored priors · illustrative UI
commerce arena2029203220362040
Ski gogglesselected lensapproval100% selectedapproval88% selectednegotiate61% selectednegotiate59% selected
Household replenishmentexplore this needapproval100% selectedapproval33% selectedpolicy61% selectedpolicy59% selected
Everyday mobilityexplore this needapproval100% selectedapproval88% selectednegotiate61% selectednegotiate59% selected
Complex travelexplore this needapproval100% selectedapproval100% selectedapproval95% selectedapproval68% selected
Care coordinationexplore this needapproval100% selectedapproval100% selectedapproval95% selectedapproval68% selected
approvalnegotiatepolicyambient
Change the assumptions behind this story
Server simulationEvery submitted counterfactual is recomputed by the deterministic horizon engine; the URL preserves the complete run state.
Long-horizon counterfactual · structural assumptions

Change what becomes possible.

These inputs rerun the selected path from its 2029 scenario state. They do not alter the near-term engine or any dated observation.

published preset

A seasonal gear purchase shaped by personal fit, lens preferences and changing conditions. The arena changes interface thresholds and human attention, not the aggregate market path.

repeatability30%reversibility70%consequence45%preference load65%
unanchored scenario prior · modeled category lens
8% → 24% · unanchored research range
15% → 45% · unanchored research range
6% → 24% · unanchored research range
10% → 90% · unanchored research range
10% → 90% · unanchored research range
8% → 35% · unanchored research range
2% → 18% · unanchored research range
5% → 75% · unanchored research range
2% → 18% · unanchored research range
Only the selected structural path is replaced; the comparison lines remain the published presets.
Benchmark the result

Three regimes become reference points, not destiny.

The reader-built path can combine mechanisms that the old labels kept apart. Coordinated-open, fragmented-open and walled presets remain below as transparent comparison anchors for concentration and interoperability.

Coordinated open rails

agent markets

56% of commerce has a machine on both sides; 25% runs continuously against a policy. Conditional 2040 p10–p90: 18%–50% machine counterparties.

Fragmented open web

delegated commerce

23% of commerce has a machine on both sides; 7% runs continuously against a policy. Conditional 2040 p10–p90: 12%–35% machine counterparties.

Walled gardens

delegated commerce

16% of commerce has a machine on both sides; 4% runs continuously against a policy. Conditional 2040 p10–p90: 8%–25% machine counterparties.

Machine counterparties · published presets with p10–p90 structural sensitivity
0%15%30%45%60%Coordinated 56%Fragmented 23%Walled 16%202920362040
CoordinatedFragmentedWalled

Structural sensitivity ribbons. Each p10–p90 band covers 512 paired, seeded combinations of the nine disclosed unanchored structural ranges, conditioned on that branch’s published 2029 handoff. The line is the published preset; the ribbon is not a confidence interval or probability. A preset can sit outside its ribbon because the coverage design spans every range stratum rather than centering on that preset.19

Beyond the site. Embodied agents, autonomous mandates, agent trust registries and transaction-lifecycle protocols demonstrate partial building blocks—not adoption rates or a complete physical economy. Pooled demand, commissioned supply and embodied orchestration remain disclosed structural possibilities with unanchored priors.20212223

New merchant-agent evidence. Anthropic's 2026 Commerce Agents blueprint makes a governed shopper-agent plus merchant-agent stack implementable today. The model uses that architecture to sharpen one structural mechanism, but does not treat the launch or its vendor-reported business outcomes as an adoption prior.31

Research appendix: new actors and assumptions
Actors the 2029 model does not contain

The cast changes with the market.

These actors enter only when the structural layer reaches the relevant epoch.

Merchant capabilityenters 2032exposes discovery, checkout, fulfillment and support as callable operations without needing a human-facing storefrontnew actor
Seller agententers 2032sets price, inventory exposure, service terms and which buyers to trustnew actor
Authority networkenters 2032proves what an agent may spend, disclose, sign and delegate onwardnew actor
Open commerce routerenters 2032routes portable intent across catalogs, logistics and payment providers without owning the buyer relationshipnew actor
Agent market operatorenters 2036matches machine demand and supply while setting access and ranking rulesnew actor
Assurance networkenters 2036underwrites disputes, provenance, delivery and adversarial behaviornew actor
Physical operator agententers 2036commits robots, vehicles, workers and facilities to machine-negotiated physical outcomesnew actor
Metered capabilityenters 2036quotes and settles small units of digital or physical work as they are consumednew actor
Trust and escrow registryenters 2036makes identity, reputation, job state and conditional release portable between unfamiliar counterpartiesnew actor
Demand coalitionenters 2036pools bounded mandates so households or organizations bargain without surrendering individual constraintsnew actor
Resource agententers 2040continuously allocates money, energy, compute, inventory or capacity against a policynew actor
Asset twinenters 2040turns a machine, package, building or product lifecycle into an authenticated source of commercial intentnew actor
Production agententers 2040configures or commissions supply after machine demand forms, before a finished product necessarily existsnew actor
Research hooks

What the next evidence pass must update.

capabilityLearningannual improvement in agents completing multi-step commercial workprior range 0.08–0.24unanchored
inferenceCostDeflationannual decline in the cost of persistent commercial agencyprior range 0.15–0.45unanchored
sellerAgentDiffusionannual diffusion of seller-side pricing, inventory and negotiation agentsprior range 0.06–0.24unanchored
identityPortabilityhow portable identity, authority, reputation and receipts become across runtimesprior range 0.1–0.9unanchored
liabilityClarityclarity about who bears loss when an autonomous transaction goes wrongprior range 0.1–0.9unanchored
adversarialPressureannual pressure from fraud, collusion, manipulation and synthetic supplyprior range 0.08–0.35unanchored
physicalExecutionDiffusionannual diffusion of agent-addressable robots, vehicles, workers and facilitiesprior range 0.02–0.18unanchored
collectiveMandatePortabilityability to pool bounded purchasing mandates across people or organizationsprior range 0.05–0.75unanchored
adaptiveSupplyDiffusionannual diffusion of supply agents that can configure or commission capacity from demandprior range 0.02–0.18unanchored

Enter one illustrative agent-to-agent transaction →