Five overlapping futures for the agentic internet. Original Decentralised News illustration.
What Matters
By 2030, the agentic internet could combine open connections, concentrated distribution, jurisdiction-specific controls and several payment systems. Businesses should test whether identities, permissions, data and accepted outcomes survive a provider switch. DN’s scenario engine exposes your assumptions about five overlapping futures; it does not assign probabilities or predict a winning platform.
The Signal
The agentic internet is unlikely to become one uniform network by 2030. Open protocols such as MCP and A2A are pushing toward interoperability, while platform ecosystems, national regulation, sovereign infrastructure, identity rules and payment rails are pulling in different directions. DN therefore models five plausible futures instead of one prediction.
Five plausible agentic-internet futures
| Scenario | What it looks like | Who benefits | Main risk |
|---|---|---|---|
| Open | Agents move across interoperable tools, registries, identities and payment rails. | Developers, users, smaller providers, open ecosystems. | Security and trust complexity. |
| Closed | Large platforms control models, tools, identity, distribution and payments. | Integrated ecosystems and incumbents. | Lock-in and concentration. |
| Regulated | Agents operate through licensed, auditable, jurisdiction-specific frameworks. | Enterprises, governments, regulated sectors. | Slower innovation and compliance fragmentation. |
| Decentralized | Identity, compute, payments and discovery shift toward open networks and cryptographic coordination. | Permissionless ecosystems and cross-border machine commerce. | Governance, UX and abuse resistance. |
| Fragmented | Regional standards, incompatible ecosystems and policy blocs coexist. | Local champions and jurisdiction specialists. | High switching and integration cost. |
DN Agentic Internet 2030 Scenario Engine
Independent support points under your assumptions, from 0 to 6 for each scenario. These are not probabilities, forecasts or market shares. Scores can overlap and do not sum to 100. With all inputs set to the middle option, every scenario starts at 3 / 6 points.
Scenario 1: The Open Agentic Internet
In the open scenario, an agent can discover tools, authenticate, negotiate capabilities, pay and move between providers without being trapped inside one model vendor. MCP and A2A are early building blocks for this future because they separate capabilities from any single application.
The economic advantage is composability. The risk is that interoperability expands the attack surface unless identity, permissioning and provenance improve at the same pace.
Scenario 2: The Closed Platform Internet
Large model, cloud and device providers could absorb the agent stack vertically. In this future, the best user experience comes from staying inside one ecosystem because identity, memory, tools, billing and data are tightly integrated.
This is commercially attractive and operationally simple. It also creates concentration risk and makes switching costly.
Scenario 3: The Regulated Agent Internet
Finance, health, employment, education and public services may develop highly controlled agent layers. Agents could require certified identities, auditable logs, human appeal and jurisdiction-specific permissions.
The EU AI Act already shows the direction: provider and deployer duties become part of the infrastructure, not an optional add-on.
Scenario 4: The Decentralized Agent Internet
Open identity, decentralized compute, cryptographic credentials, stablecoins and internet-native payment standards could create a permissionless agent economy. Agents could discover services and settle value without bilateral integrations or centralized billing relationships.
The hard problems are trust, dispute resolution, UX and abuse resistance.
Scenario 5: The Fragmented Internet
The most realistic outcome may be partial fragmentation. Different regions could require different identity, data, model, payment and safety layers. An agent that works in one jurisdiction may not be authorized in another.
This raises integration costs but creates a new market for translation layers, policy engines and agent-routing infrastructure.
The structural variables to watch
- Interoperability: whether agents can move across models and tools.
- Identity: whether agents have portable, trusted credentials.
- Payments: whether machine settlement becomes open or platform-controlled.
- Regulation: whether compliance converges or fragments by jurisdiction.
- Distribution: whether a few ecosystems own the user relationship.
- Compute: whether infrastructure remains concentrated or becomes more portable.
How businesses should prepare without knowing which future wins
- Keep model and tool interfaces portable.
- Use open standards when they reduce switching costs.
- Separate identity and authorization from any one model vendor.
- Design payment layers so rails can change.
- Maintain policy engines that can adapt by jurisdiction.
- Track dependency concentration across cloud, models and payments.
- Preserve human appeal and auditability in consequential workflows.
DN Alpha Thesis: control is a stack, not one winner
A shared protocol can connect two services while one vendor still controls their identity, memory, distribution or billing. Equally, a managed platform can support open frameworks. AWS describes AgentCore as modular services that can work with different frameworks and models. That documentation supports a narrow interoperability claim; it does not independently demonstrate effortless migration of every production workflow. Primary source: AWS AgentCore overview.
DN inference: an organization may use open tool connections, a concentrated cloud runtime, regulated financial permissions and stablecoin settlement at the same time. These are overlapping operating layers. Treating them as five exclusive bets obscures the procurement decisions that matter.
How the support points are calculated
Each control takes a declared value: 0 for weak or low, 1 for growing, partial or medium, and 2 for strong, portable or high. DN assigns a primary driver twice the weight of a secondary driver. This is an editorial sensitivity rule, not a coefficient fitted to market data.
| Scenario | Published formula | Interpretation |
|---|---|---|
| Open | 2 × interoperability + portable identity | Stronger common interfaces and portable identity support this scenario. |
| Closed | 2 × platform concentration + (2 − interoperability) | Concentrated distribution and weaker common interfaces support this scenario. |
| Regulated | 2 × regulatory intensity + portable identity | Stronger formal controls and usable identity support this scenario. |
| Decentralized | 2 × decentralized rails + portable identity | More usable open rails and identity support this scenario. |
| Fragmented | 2 × geopolitical fragmentation + (2 − portable identity) | More regional divergence and weaker portable identity support this scenario. |
Range: 0–6 points for each scenario. With all six inputs at their middle option, every output is 3/6. A zero is not a zero probability. Adding the outputs is not meaningful. The controls omit many real-world factors, including the economics of compute, data licensing, demand, enforcement capacity and integration quality.
Download the v1.1 scenario matrix and formulas (JSON) · Download CSV. These are DN-authored planning assumptions with linked documentation, not measurements of adoption or a statistically calibrated forecast.
Turn a 2030 scenario into a decision you can test today
| Exposure | Question to ask before committing | Evidence to keep | Continue the research |
|---|---|---|---|
| Open connections | Can the alternative provider discover the same capability and execute the same permitted task? | Protocol and tool versions, scope, result checks and errors. | Discovery and verification |
| Integrated platform | Can you export state, change providers and preserve approval boundaries? | Export format, rebuild work, renewal terms and a matched-task trial. | Runtime portability |
| Regulated workflow | Which role, jurisdiction and use case determines the actual obligations? | Scope assessment, applicable dates, oversight and appeal procedures. | Oversight and compliance cost |
| Open settlement | Can you distinguish payment, delivery and an accepted task outcome? | Terms, payment references, billable calls and accepted-result evidence. | The real API bill |
| Regional fragmentation | Does the service remain available and authorized on the second route? | Regional terms, dependencies, fallback limits and a migration test. | Concentration risk |
Beginner: map one useful workflow and its human approval point. Builder: run the same bounded task on an alternative stack, recording cost, accepted outcomes and permission differences. Institution: add contractual exit rights, data-retention rules and jurisdiction-specific review. A standards logo is not a substitute for those observations.
What would change DN’s hybrid thesis?
- Independent matched-workload tests show that state, permissions and accepted outcomes move reliably between providers without material rebuild work.
- Documented distribution and purchasing data show one platform family displacing competing routes across several sectors.
- Implemented cross-border identity and policy systems remove important regional authorization gaps.
- Evidence shows that open settlement consistently works with clear delivery, recovery and dispute terms at useful cost.
These are review questions, not findings from a DN market census. A future update should attach dated evidence and distinguish a publisher announcement from a tested result.
For a documented portability and dependency assessment, the DN Research Desk offers paid, scoped research. Define the workload, evidence and unresolved questions before commissioning work. Payment does not buy a favorable conclusion.
Methodology and limitations
Version 1.1, 2 November 2026. DN synthesizes current evidence from agent interoperability standards, platform concentration, AI regulation, identity and machine-payment infrastructure into five scenario families.
Limitations: scenario planning is not forecasting. Multiple scenarios can coexist across industries and regions. The engine is designed to expose assumptions, not predict one definitive 2030 outcome.
Commercial disclosure: no protocol, cloud, model or payment provider paid for inclusion.
Primary sources
Additional primary references reviewed on 2 November 2026: A2A project, 27 August 2026; AWS AgentCore overview. These documentation sources establish described features and announcements, not independent performance results.
- Model Context Protocol, July 2026 specification
- Linux Foundation, A2A adoption milestone
- European Commission, AI Act obligations
- x402, HTTP 402 and payment headers
FAQ
What is the agentic internet?
An internet where AI agents can discover services, authenticate, use tools, transact and coordinate with limited direct human intervention.
Will one standard dominate?
Not necessarily. Different standards may dominate discovery, tool use, identity, communication and payments.
Will the agentic internet be open?
Open standards are gaining traction, but platform integration and regulation could keep large parts of the ecosystem semi-closed.
Why might the internet fragment?
Different jurisdictions may impose incompatible rules for data, identity, safety, payments and model access.
What role does crypto play?
Crypto and stablecoin rails can support internet-native machine payments, but they are only one part of the broader agent stack.
What is the biggest concentration risk?
Many apparently independent agents may still rely on the same model, cloud, identity or payment provider.
How should companies prepare?
Build portable interfaces, open-standard support, flexible identity and payment layers, and jurisdiction-aware policy controls.
Which scenario is most likely?
The tool cannot estimate which scenario is most likely. DN’s working thesis is that several structures may coexist across sectors and regions; this remains an editorial inference.
About the Author
