The agentic economy market map

On this page

Chapter 3 of The Hitchhiker's Guide to the Agentic Economy: Australia 2026

Report hub · Previous: AI agent adoption and measurement · In this chapter: agent systems · commerce · actors and loss · service snapshot · procurement gate

The functional topology of agent commerce

The market for autonomous software is not a single product layer. It is an interconnected supply chain linking raw model inference to enterprise balance sheets.

A foundation model provides reasoning and token generation. But transforming raw inference into consequential economic action requires memory stores, durable runtimes, external tool connectors, identity registries, policy engines, checkout protocols, payment rails, and enterprise systems of record. When an enterprise deploys an agent, it is not simply purchasing software; it is establishing a chain of authority, expenditure, and legal liability across multiple independent technology vendors.

Understanding this commercial landscape requires looking beyond vendor marketing categories. A functional market map must answer four basic operational questions:

• Work: What specific task is delegated, and who verifies that the delivered output meets business requirements?
• Authority: What internal systems, external APIs, and budget limits is the software permitted to touch?
• Settlement: What financial rails and payment instruments move value, and how do they reconcile into accounting ledgers?
• Recourse: Who bears operational and financial loss when an autonomous run fails, duplicates actions, or corrupts production data?

The agentic economy organizes into four connected systems: work and demand, agent systems, exchange and commerce, and institutions and consequences. Across these systems, five distinct flows must remain coordinated: work, data, authority, money, and accountability.

Key Takeaways

• The market has four connected systems and more than a dozen distinct infrastructure layers.
• APIs, MCP, A2A and commerce protocols solve adjacent problems; they are not direct substitutes for every use case.
• Commercial completion requires discovery, authority, transaction, delivery, acceptance and records.
• Existing enterprise and public infrastructure remains part of the stack.
• Market value is likely to collect around scarce capability, trusted context, distribution, authority, systems of record and accepted outcomes.

Four interconnected market systems

Companies in this market frequently span multiple operational boundaries. Model providers offer hosted execution sandboxes and proprietary tool frameworks. Cloud platforms bundle inference compute, identity management, and observability tracing into managed suites. Payment networks extend commercial card rails to digital agents. Incumbent enterprise platforms adapt existing databases and ERP systems to ingest autonomous machine actions.

Categorizing participants by functional control boundaries clarifies how value and risk are distributed:

                          INSTITUTIONS AND CONSEQUENCES
          law | standards | public infrastructure | labour | capital
                                      |
                                      v
WORK AND DEMAND                AGENT SYSTEMS                 EXCHANGE AND COMMERCE
objective                      models + compute              discovery + offers
workflow        ----------->   context + memory   -------->  checkout + orders
buyer and user                 runtime + tools               credentials + payments
outcome                        identity + policy             delivery + acceptance
repeat use      <-----------   evaluation + security <----  billing + recourse
                                      |
                                      v
                         ENTERPRISE SYSTEMS OF RECORD
                ERP | CRM | finance | procurement | identity | audit

Across every connection: work | data | authority | money | accountability

This commercial architecture operates as a closed loop. Business demand originates an objective. Agent infrastructure executes the required steps. Commerce protocols manage exchange between independent buyers and sellers. Institutions set legal rules and statutory obligations. Enterprise systems of record retain authoritative financial and operational history. Operational evidence from completed jobs flows back to the operator and the market.

View one: work and demand

All software capability remains speculative until attached to a business task that someone funds and accepts:

Layer	Functional role	Operational evidence
Objectives and tasks	Translates business needs into bounded, executable jobs	Explicit deliverables, stopping rules, and failure constraints
Workflows and sectors	Embeds execution inside existing industry processes	Pre-existing operational workflows, legacy system hand-offs, and compliance boundaries
Buyers and users	Identifies budget holders, operators, and commercial beneficiaries	Sponsoring business unit, human operator, and accountable corporate principal
Outcomes and repeat use	Confirms whether work delivers verified commercial utility	Formal sign-off, cycle-time reduction, fully loaded cost, and repeat order rates

Demand is currently concentrated across software engineering, customer support, legal and financial research, sales operations, and administrative data processing. High utilization of conversational assistants does not indicate autonomous agent adoption. The distinguishing operational event is delegated execution inside an enterprise workflow, tracked via verifiable job-level records.

View two: agent systems and execution infrastructure

Agent infrastructure transforms a high-level objective into monitored, multi-step actions in external environments:

Layer	Functional role	Representative participants or approaches	Commercial question
Models and inference	Reasoning, multimodal processing, planning, and function selection	OpenAI, Anthropic, Google, Meta, Mistral, Cohere; open-weight ecosystems	What capability, latency, context size, and price per token fit the task?
Chips, networks, data centres and energy	Physical compute capacity, interconnects, and electrical power	NVIDIA, AMD, hyperscalers, data-centre operators, power utilities	Where do physical compute bottlenecks, geographic latency, or energy costs limit scale?
Compute and hosting	Managed runtime deployment, GPU orchestration, and scaling	AWS, Microsoft Azure, Google Cloud, specialized inference clouds	Where does compute availability or margin compression constrain profitability?
Context, memory, data and provenance	Information retrieval, persistent state, embeddings, and licensing	Relational databases, vector stores, web search providers, enterprise knowledge graphs	Is context current, permissioned, attributable, and isolated from untrusted input?
Runtimes and orchestration	Execution state machines, checkpointing, retries, and recovery	OpenAI Agents SDK, Google ADK, LangGraph, Microsoft Agent Framework, AWS Bedrock AgentCore	What operational state survives a worker crash, network partition, or long-running pause?
Tools, browsers and connectors	Interaction with external software, DOMs, and service endpoints	REST APIs, headless browser platforms, SaaS connectors, MCP servers, code sandboxes	Can the agent execute actions reliably without triggering anti-bot or data integrity failures?
Agent coordination	Protocol discovery, sub-task delegation, and messaging across agents	A2A protocol and proprietary multi-agent messaging frameworks	How do independent agent systems hand off tasks and share verifiable execution traces?
Identity, authority and policy	Principal mapping, privilege scoping, budget caps, and revocation	Enterprise IAM (Entra, Okta), OAuth token vaults, AP2 mandates, policy engines	Who authorized the agent, what budget limits apply, and how is access instantly revoked?
Evaluation and observability	Tracing, regression testing, quality metrics, and incident logging	Observability platforms, LLM tracing tools, synthetic evaluation suites	Can the enterprise detect, explain, and audit unexpected execution behavior?
Security and recovery	Sandboxing, prompt-injection defense, and operational restoration	Isolated container runtimes, input guardrails, human escalation queues	How is the operational blast radius contained when an agent encounters malicious input?
Implementation and operations	Systems integration, workflow redesign, and human oversight	System integrators, IT consultancies, BPO providers, internal operations	Who manages the manual reviews, exceptions, and data cleaning outside the software boundary?

Large vendors increasingly package several horizontal layers into vertical suites. An orchestration framework may embed telemetry; a cloud hyperscaler may package identity, sandboxes, and models into an agent service. Architectural risk concentrates at the boundaries where these components hand off execution state.

Specialized interfaces across agent boundaries

Emerging protocols address different integration surfaces across the execution lifecycle. Supporting multiple protocols is standard practice because each addresses a different operational boundary:

Interface	Primary problem	Functional boundary	What it does not establish
REST / gRPC API	Executes a defined software operation	Application to application	Agent discovery, delegated corporate authority, or commercial payment terms
Model Context Protocol (MCP)	Standardizes how an AI model connects to local/remote tools and context	Host application, client runtime, and MCP server	Server security posture, payment settlement, delivery acceptance, or vendor terms
Agent2Agent (A2A)	Standardizes discovery, task delegation, and artifact return between agents	Client agent runtime and remote agent service	Shared runtime memory, tool internals, legal mandate authority, or payment clearing
ACP / UCP	Standardizes merchant product catalogs, carts, and order checkout states	Agent platform, consumer wallet, and merchant backend	Universal corporate identity, payment rail interoperability, or post-purchase dispute rules
AP2	Standardizes intent, mandate, and payment-authority semantics	Human principal, agent software, merchant, and payment rail	Delivery quality, ERP reconciliation, statutory tax compliance, or legal agency
x402 / MPP	Initiates and settles request-time machine payments over HTTP	Autonomous client, web server, and payment facilitator/blockchain	Physical/digital fulfillment, customer acceptance, tax invoicing, or chargeback recourse
Payment Cards & Scheme Tokens	Supplies payment credentials, network routing, and clearing	Consumer/corporate payer, merchant acquirer, card issuer, card scheme	Verification of the underlying business deliverable or task authorization

MCP and A2A serve complementary roles. MCP standardizes how a single agent accesses tools and external data sources within its runtime environment (Model Context Protocol documentation). A2A standardizes how independent agent systems locate each other, negotiate task parameters, track asynchronous execution, and return deliverables across corporate networks (Google Developers, "Developer's Guide to AI Agent Protocols").

View three: exchange and commerce

Exchange infrastructure bridges technical execution and commercial trade, allowing autonomous buyers and sellers to discover services, negotiate pricing, settle transactions, and manage disputes:

Layer	Functional role	Current market implementations	Unresolved commercial friction
Discovery and representation	Makes products, services, and agent tools machine-readable	Search engines, directory endpoints, A2A Agent Cards, software marketplaces	Who controls ranking, verifies seller claims, and prevents malicious spoofing?
Offer and quote	Establishes explicit scope, pricing, deadlines, and delivery terms	Machine-readable pricing feeds, dynamic quoting endpoints, fixed API terms	How do dynamic quotes account for unexpected execution retries and variable compute costs?
Checkout and order	Captures commercial commitment and generates an order record	UCP, ACP, and proprietary merchant checkout endpoints	Which party maintains state machines for cancellation, stock timeouts, and order amendments?
Authority and credential	Proves the agent is authorized to commit financial resources	AP2 signed mandates, virtual card tokens, scoped OAuth credentials	How does corporate spending authority propagate across multi-hop sub-agent delegations?
Payment and settlement	Transfers monetary value across designated financial rails	Card networks, NPP/PayTo, x402 over Base/Solana, MPP HTTP payment	Which rail balances execution speed against enterprise invoicing and chargeback recourse?
Delivery and acceptance	Transmits the deliverable and confirms it satisfies scope	API responses, merchant fulfillment webhooks, human sign-off	What immutable evidence proves that the customer received and approved the deliverable?
Billing and finance	Tracks consumption, generates tax invoices, and reconciles ledgers	Usage billing engines (Stripe, Metronome), ERP systems (SAP, NetSuite)	How are granular, high-frequency machine calls aggregated into three-way matched invoices?
Support and recourse	Manages error remediation, service credits, and formal disputes	Card scheme dispute arbitration, contractual SLAs, support desks	Who carries financial liability when an agent executes unintended or corrupted actions?

Three distinct commerce operating models

Commercial activity is developing across three different environments:

1. Consumer agentic commerce: Conversational assistants helping consumers locate products, compare merchant pricing, and complete checkout. Projects supported by OpenAI, Google, Shopify, Stripe, and PayPal operate largely in this space (Universal Commerce Protocol; Agentic Commerce Protocol). The merchant manages inventory, payment capture, physical fulfillment, and customer service; the assistant functions as a conversational discovery channel.
2. Request-time machine payments: Software directly paying for atomic digital resources (web scraping, real-time data feeds, specialized model inference) at the moment of consumption. Protocols like x402 and MPP remove the overhead of account onboarding, API subscription tiers, and pre-funded balances. Settlement is executed per-call, frequently via stablecoins.
3. Enterprise supplier commerce: Autonomous software procuring goods and services within established B2B vendor relationships. Transactions operate under master services agreements (MSAs), pre-approved purchase orders, volume commitments, and monthly invoicing. Enterprise ERP and procurement platforms maintain the authoritative commercial record; agent interfaces streamline ordering and service consumption without bypassing accounting governance.

A consumer digital wallet can issue an agent a temporary virtual card. An enterprise agent can use an x402 micropayment to purchase a research dataset, subsequently submitting the transaction receipt into a corporate expense management system. Architecture depends on transaction size, counterparty trust, regulatory requirements, and failure recourse.

Tracking actors and financial exposure

The term agent obscures the human and corporate entities carrying economic risk. A single automated transaction traverses an extended chain of participants:

principal -> orchestrator -> agent -> tool or service -> merchant or provider
          -> payment participants -> delivery -> acceptance
          -> finance and reconciliation -> correction or dispute

At every hand-off, operational governance requires five explicit data points:

• Authority: Who granted the system permission to act?
• System of record: Which database or ledger holds authoritative state?
• Funds routing: Which party receives the payment, and through which intermediary?
• Loss allocation: Who absorbs the cost if the deliverable is corrupted or execution fails?
• Remediation: Who holds the technical and legal authority to halt the run, reverse the transaction, or demand a refund?

An economic principal is the person or corporate entity whose balance sheet, legal standing, and contractual obligations sit behind the action. A software agent holds no legal standing; it operates entirely on borrowed authority.

View four: institutions, governance and consequences

Agent systems function within established legal frameworks, regulatory regimes, and economic institutions:

Layer	Institutional role	Operational governance question
Enterprise systems	Maintain customer, vendor, financial, and compliance records of truth	Which database remains legally authoritative when an autonomous agent updates a record?
Firms and operating models	Structure organizational delegation, authority limits, and management	Which business decisions can be automated, and where is human sign-off legally required?
Labor and workforce	Provide domain expertise, exception handling, and manual remediation	Does autonomous software eliminate tasks, or increase demand for skilled verification?
Standards and governance	Establish technical interoperability, security baselines, and data formats	Which emerging protocol specifications have achieved production stability and cross-vendor support?
Law and regulation	Govern contractual enforceability, statutory liability, and sector compliance	Who is held legally liable when an autonomous agent causes financial harm or data exposure?
Public infrastructure	Provide national clearing houses, digital identity, and tax networks	How do autonomous agents interface securely with public banking, identity, and taxation rails?
Capital and market structure	Fund technology development and determine pricing power	Which infrastructure layers maintain pricing power versus facing rapid commoditization?
Insurance and risk transfer	Underwrite, price, and insure operational technology risks	Are autonomous software errors, unauthorized payments, and data breaches covered by existing policies?
Professional services	Design workflows, integrate legacy systems, and handle complex exceptions	What volume of human operational labor is required to keep autonomous systems functioning?
Physical robotics and IoT	Translate software decisions into kinetic action across warehouses and transit	What real-time fail-safes prevent physical damage when an autonomous agent commands machinery?

The Australian regulatory and operating environment

Australian enterprise deployments operate on global infrastructure while complying with domestic statutory requirements:

• Privacy and data handling: Guidance from the Office of the Australian Information Commissioner (OAIC) mandates rigorous due diligence, transparency, and human oversight when deploying commercial AI products (OAIC, privacy and commercially available AI products).
• Cyber security: The Australian Signals Directorate (ASD) provides technical frameworks for hardening agentic AI harnesses, focusing on credential isolation, data boundary enforcement, and malicious prompt containment (ASD, agentic AI harnesses).
• Financial services governance: The Australian Securities and Investments Commission (ASIC) actively scrutinizes AI governance among Australian Financial Services licensees. ASIC Report 798 evaluated 23 licensees across 624 AI use cases, identifying significant oversight gaps in board reporting, risk management, and consumer protection (ASIC, REP 798).
• Payment mandates: Australian Payments Plus operates PayTo, providing real-time infrastructure to establish legally enforceable account-to-account debit agreements with explicit transaction limits (Australian Payments Plus, PayTo).
• Structured invoicing: Australia's Peppol eInvoicing network standardizes structured, machine-readable invoice exchange directly between enterprise accounting platforms, eliminating manual invoice processing (Australian Taxation Office, eInvoicing).

These domestic institutions provide the legal and technical rails required to govern autonomous agency. Production deployment requires linking these systems into an organization's existing internal approval hierarchies and enterprise resource ledgers.

Five flows connecting the market

A functional market map is defined by how resources move across institutional boundaries. Five distinct flows must remain coordinated for an agentic transaction to complete successfully:

1. Work: The objective, execution plan, task state machine, sub-agent handoffs, and returned deliverable, bound to a single immutable job ID.
2. Data: Ingestion of source materials, system instructions, retrieval context, tool outputs, and execution logs. Each participant must hold verified legal rights to access, process, and retain that data.
3. Authority: Cryptographic credentials, enterprise roles, spending mandates, policy decisions, and instant revocation mechanisms linking the software actor back to the sponsoring principal.
4. Money: Price quotation, payment authorization, asset transfer, clearing finality, tax invoicing, and dispute remediation.
5. Accountability: Structured trace logs, evidence cards, validation decisions, human sign-off records, and contractual liability.

When commercial implementations fail, the root cause is almost invariably a disconnect among these five flows. A tool execution without verified authority creates a security breach. A payment settling without a verified delivery leaves financial loss. An output transmitted without formal acceptance cannot be recognized as commercial revenue.

Operational failure modes at the boundaries

Deploying autonomous software exposes critical failure modes across system hand-offs:

Operational failure mode	Control evidence required
Untrusted web content or prompt injection redirects execution	Verified instruction boundaries, sandbox containment, strict tool whitelisting
A credential is leveraged outside its authorized purpose	Principal binding, audience restrictions, dollar caps, short-lived expiry, token revocation
An agent executes excessive or catastrophic actions	Hard budget limits, write-action approval gates, policy decision logs
A network error triggers duplicate billing or double-purchasing	Stable job IDs, distributed idempotency keys, automated reconciliation routines
A malicious actor impersonates an agent service	Mutual TLS, cryptographic task signing, verifiable service endpoint records
Sensitive corporate data leaks across third-party tools	Zero-retention agreements, regional data residency, auditable data disposal logs
Delivered work is inaccurate, hallucinatory, or incomplete	Formal acceptance tests, human review escalation queues, contractual refund terms
Execution crashes or stalls after funds have moved	Linked order, payment, and delivery state machines managed by a single reconciliation owner

Verifying these controls requires an auditable operational record:

Control parameter	Operational requirement
Accountable owner	Named human role and the exact software subsystem where policy is enforced
Test methodology	Documented, reproducible test fixtures covering boundary conditions and negative test cases
Pass/fail criteria	Explicit thresholds for acceptable latency, error rates, and catastrophic failure tolerance
Audit cadence	Scheduled pre-deployment, periodic runtime, and incident-triggered reviews
Retained evidence	Immutable execution logs, cryptographic receipts, and signed decision records
Escalation path	Defined operational protocols for containing live failures and issuing refunds

Market value concentration

Six strategic hypotheses describe where defensible enterprise value will accumulate across the agent stack:

Strategic position	Source of commercial scarcity	Operational test
Proprietary model capability	Specialized reasoning architectures, domain tuning, and efficient inference	Does the model achieve higher first-pass acceptance rates on specific business workflows?
Governed enterprise context	Clean, permissioned, proprietary corporate data that competitors cannot scrape	Does the provider hold exclusive, attributable data essential for fulfilling the task?
Distribution and workflows	Deep integration into daily enterprise workflows, browsers, and operating systems	Can the vendor reach corporate buyers directly without paying high toll fees to platform gatekeepers?
Authority and identity rails	Robust cross-organizational delegation, mandate enforcement, and liability management	Does the platform verifiably reduce enterprise financial and legal risk?
Systems of record	Authoritative corporate databases that store operational, customer, and financial history	Does the agent layer strengthen the core system of record, or does it bypass it?
Guaranteed accepted outcomes	End-to-end integration of models, software scaffolding, human review, and financial recourse	Can the vendor deliver completed work profitably under fixed-fee or outcome-based commercial terms?

As foundation model reasoning commoditizes, value will tilt toward proprietary enterprise context, systems of record, and managed outcome delivery. The defining commercial metric will be whether a provider can deliver verified, accepted work at a sustainable gross margin.

Empirical service snapshot

The operational gap between an available API endpoint and a completed business job is evident in production pricing. The following records document live, reachable x402 endpoints checked as of 7 September 2026. Each represents a functional, machine-payable interface; none by itself completes an accepted business task:

Service route	Purchased asset	Published price	Endpoint access	Remaining customer integration work
Tavily direct x402 search	Single web search query with optional raw page content	0.01 USDC per call on Base	Public POST /search requiring x402 header	Raw search results require validation, deduplication, and synthesis into a deliverable
Browserbase direct x402 session	Headless browser execution container time	0.03 USDC for 15 minutes (0.12 USDC per hour)	Public session initialization endpoint	Buyer's agent must script the browser, handle CAPTCHAs, extract data, and verify accuracy
Firecrawl through StableEnrich	Web search or markdown scraping via gateway	US$0.0252 per search; US$0.0126 per scrape (USDC)	Public gateway endpoint	Gateway terms, caching rules, and uptime SLAs differ from direct vendor procurement

These micro-services settle programmatically in milliseconds. But converting them into enterprise value requires substantial application scaffolding: context boundaries, error handling, quality verification, and accounting reconciliation.

The enterprise adoption boundary

Before procuring autonomous software, enterprise IT, legal, and finance teams evaluate ten operational boundaries:

1. Job definition: What exact business deliverable is produced, and who formally reviews and approves it?
2. Authority boundary: What internal systems, external APIs, and financial budgets is the software authorized to access?
3. Data governance: Where are prompts, context, and outputs processed, how long are they retained, and which system maintains the authoritative record?
4. Auditability: Does the platform export immutable job IDs, execution traces, payment receipts, and delivery logs?
5. Quality standards: What defines first-pass acceptance, what constitutes rework, and what are the tolerated failure thresholds?
6. Remediation: Who is contractually obligated to correct errors, issue service credits, or provide refunds when execution fails?
7. Operating limits: What spend ceilings, rate limits, availability SLAs, and support response times apply?
8. Human oversight: Which specific actions require synchronous human approval before side-effects are committed?
9. Financial reconciliation: How do transaction events map to general ledger accounts, purchase orders, and statutory tax invoices?
10. Termination: What performance thresholds or security near-misses trigger mandatory suspension or contract termination?

Commercial vendor evaluation requires analyzing the complete contractual boundary:

Contractual parameter	Procurement requirement
Service scope	Explicit definitions of the task, output format, verification criteria, and support coverage
Commercial pricing	Base platform fees, token consumption rates, overage tiers, and currency/tax terms
Data residency	Jurisdictional boundaries for data processing, backup storage, and sub-processor access
Service availability	Uptime commitments, scheduled maintenance notice, and degradation remedies
Platform stability	Mandatory advance notice before deprecating models, APIs, or authentication methods
Portability	Export formats for system logs, fine-tuning artifacts, and enterprise context
Liability and remedy	Indemnification for intellectual property infringement, data leaks, and unauthorized transactions
Exit strategy	Immediate credential revocation, verifiable data deletion, and operational cutover support

An infrastructure provider supplying containerized sandboxes may deliver excellent execution isolation and trace logs while leaving merchant verification, payment authority, and deliverable validation entirely to the customer. Enterprise procurement must evaluate the complete operating path rather than mistaking an isolated technical tool for an end-to-end commercial platform.

The evolution of market boundaries

The market for autonomous software is expanding horizontally across models, runtimes, tool protocols, and payment mechanisms. However, commercial governance remains anchored in established institutional infrastructure: corporate identity providers, merchant gateways, enterprise ERP ledgers, and national banking rails.

The most valuable commercial platforms will emerge at the connection points: linking discovery to binding commercial quotes, binding authority to external execution, and reconciling machine payments with accepted business delivery. Organizations that make these hand-offs transparent, secure, and auditable will define the operating architecture of the agentic economy.

Part II analyzes each technical layer of the execution path, tracing a single delegated job from model reasoning to accepted delivery.

Sources and method

• Primary sources: Technical specifications for MCP, A2A, UCP, ACP, AP2, x402, and MPP, vendor product documentation, Australian regulatory publications, and empirical service pricing checked as of 13 September 2026. Service snapshot checked on 7 September 2026.
• Analytical models: The four-system market architecture, the five-flow coordination model, and the enterprise adoption boundary are original analytical frameworks developed to structure market relationships.
• Evidence boundaries: Specifications establish interface design intent and stated capability. Interface comparisons evaluate architectural mechanics rather than third-party conformance testing. Service prices reflect public documentation on the checked date; no live commercial payments were executed.
• Institutional context: Regulatory materials from the OAIC, ASD, ASIC (REP 798, covering 23 licensees as of December 2023), Australian Payments Plus, and the Australian Taxation Office document Australian legal and financial infrastructure.
• Editorial record: Published by Agentic Economy, Perth, Western Australia. Updates and corrections are logged via the editorial record.

Back to the report hub · Next: models, inference and compute.