The agentic layer: how autonomous agents reshape the web by 2030

Automating interaction, not emotion.

Hypothesis

By 2030, the transactional layer of the web will be dominated by the Agentic Layer: a distributed network of autonomous agents, interfaces, and protocols where brand-consumer interaction shifts from direct website engagement to mediated, context-aware Agentic Experience (AX).

Websites won’t disappear, but they’ll become backend API-feeds, quietly powering real-time negotiations between brands and agentic systems.

Brands, marketers, and publishers will optimise for Agentic Interface Protocols (AIP) rather than traditional metrics like traffic or conversion.

Two internets, one transition

We already live across two internets:

Layer	Purpose	Nature
Social Internet	Emotional engagement, identity, culture	Human
Functional Internet	Discovery, decision, transaction	Machine

The first will remain human. The second is becoming automated.

The internet itself, the pipes, servers, cloud, and code, stays. What’s changing is the interface between intent and action. That interface is becoming agentic.

Soon, when a person says “book my usual hotel” or “find me a local agency that does analytics and privacy,” that intent won’t hit a website. It’ll be parsed, evaluated, and executed through agents negotiating with brand APIs in real time.

Defining the Agentic Layer

The Agentic Layer is a distributed ecosystem of autonomous agents that interpret intent, negotiate value, and transact data, content, and commerce without continuous human mediation.

I use AIP here as my working name for the interface protocol I want this layer to have: a way to negotiate permission, attribution and the return of value to the people and organisations supplying it. It is a proposal, not a claim that these capabilities already exist as one adopted standard.

Core components

Human interfaces (UX), where people delegate and receive outcomes, and Agentic Experience (AX), how agents interact with the systems delivering them
Agentic Interface Protocol (AIP), the structured, authenticated standards that define how agents communicate, verify, and transact
Agentic Infrastructure, APIs, data vaults, and identity layers that make those transactions secure, auditable, and composable

The web doesn’t vanish; it just changes purpose. HTML is the legacy front-end. APIs are the new surface.

Historical parallels

Each technology wave imitates what came before until its true form emerges.

The early web mimicked magazines.
Social media mimicked social circles.
Early AI mimicked chatbots.

Now the native form is appearing: agency. Systems that act, not just answer.

Drivers of change

Technology

Patents: filings such as C3 AI’s patent application US20250124069A1, published 17 April 2025 describe orchestrated multi-agent systems where autonomous sub-agents execute coordinated tasks across domains.
Operating systems: Windows Copilot, Gemini on Android, and Apple Intelligence (flawed, but in development) are embedding agentic orchestration at the system layer. Meanwhile, Meta and OpenAI are racing toward different implementations of this vision: Meta through wearable interfaces that bring agentic assistants into daily life through glasses and potential brain-to-text technology; OpenAI through a hardware design collaboration effort aimed at redefining the post-smartphone era.
Browsers: Opera, SigmaOS and Perplexity (Comet) are releasing early agentic browsers, capable of acting on user intent inside the browsing flow.
Infrastructure: OpenAI’s GPTs, Anthropic’s MCP, and OpenAI and Stripe’s ACP are early building blocks for the governed exchanges I describe here.

Behaviour

People are shifting from searching to delegating. “Find me” becomes “Do it for me.”

Evidence:

Google AI Overviews now answer queries directly, collapsing the search journey.
Google Search AI Mode executes next steps (book, compare, calculate) without leaving results.
ChatGPT with browsing and shopping actions routes user intent straight to product databases.

Each of these is an early interface for delegation, intercepting the traditional brand click-path. AX describes the agent’s interaction with the services underneath it.

Economics

The economy is shifting from attention to execution. What matters is no longer “Who saw it?” but “Which agent completed it?”

Agentic commerce protocols are reaching early deployments, alongside established API platforms. I see the investment in agent tools and multi-agent orchestration as a path towards AIP and negotiated value exchange.

What changes, what doesn’t

What changes

Websites become backend endpoints.
UX/CX shifts from visual design to agentic negotiation.
SEO evolves into AEO, Agentic Experience Optimisation.
KPIs become about data quality, trust signals, and execution value.

What doesn’t

People still scroll.
Culture still forms in the social layer.
Emotion remains human.

Agents remove friction, not feeling.

Implications

For brands

Your front door is no longer your homepage, it’s your Agentic Endpoint.
You’ll define what data, content, and offers are available to agents via AIP.
Optimise for AX visibility: ensure agents can interpret, trust, and transact your data.

For consumers

The interface becomes conversational, multimodal, and goal-driven.
Trust shifts from “Do I trust the brand?” to “Do I trust my agent to choose well?”

For the economy

The funnel collapses.
Transactions occur before a click.
Data accuracy and protocol compliance define competitiveness.

The death of robots.txt

Let’s be real: robots.txt is 90s technology trying to survive in a post-LLM world.

It was built for polite crawlers knocking on HTML doors. Agents can still browse and crawl today. My expectation is that the valuable exchange moves towards negotiation.

In the Agentic Layer, access is brokered through authenticated APIs. robots.txt expresses crawler preferences; it is not an access-control mechanism.

Each brand will operate an Agentic Endpoint, a dynamic interface defining what data, content, or commercial terms can be accessed, by which agents, and under what rules. That’s where the negotiation happens, live, under AIP standards.

The handshake replaces the crawl. Discoverability and commerce move from static instructions to real-time, policy-aware exchange.

Goodbye robots.txt. It belongs in the museum next to dial-up modems and Netscape Navigator.

Boundaries and critical notes

The Agentic Layer automates the functional internet, not the social one.
Adoption will vary: travel, finance, and logistics will move first.
Regulation and interoperability lag behind.
Without open AIP standards, silos will form.

Scenarios and confidence

Scenario	Description	p-score
Optimistic Convergence	Open AIP standards, interoperable AX ecosystems, seamless delegation	0.65
Siloed Ecosystems	Closed platforms, fragmented experiences	0.55
Hybrid Coexistence	AX dominates transactions; social remains the cultural core	0.80

p-score = my personal probability rating for this note, based on agent adoption, AIP progress, and brand readiness. These scenarios can overlap, so the ratings are not a distribution that must total one.

China note: as Kai-Fu Lee argues in AI Superpowers, China’s pragmatic, test-and-learn approach gives it an advantage. Expect China to run ahead on AX and AIP deployment, testing in the wild while the West debates governance.

Personal reflection

The value-exchange question is central: if an agent uses someone’s content, data or service to deliver an outcome, how does value return to the source? This is the line of thinking that developed into Odexa.

Websites won’t die, they’ll simply stop being for humans first.

They’ll live on as background infrastructure, machine-readable and machine-negotiated. The new UX/CX is the conversation between user intent and agentic execution. The interface is a negotiation layer, not a landing page.

We’re already seeing this through AI Overviews, Google Search AI Mode, and ChatGPT’s integrated browsing.

Search 1.0 (blue links) became Search 2.0 (AI summaries). The next phase, Discovery 2.0, is about delegation.

Brands that treat this as incremental optimisation will fall behind. Those that prepare for AX-based access will lead.

Closing note

This is a working hypothesis, not a conclusion. The direction feels clear even if the timing doesn’t.

The Agentic Layer is forming quietly, beneath the noise of social media. It won’t replace the web; it will rewrite its purpose.

When discovery becomes negotiation, the web stops being something we visit. It becomes something that acts on our behalf.

Working questions

I’m working out loud on this hypothesis, discovering new questions and uncovering blind spots as I go. Some answered below.

What is MCP?

MCP stands for Model Context Protocol. It is an open protocol connecting AI applications to tools, resources and prompts. The application and server implement authentication, permissions and safeguards; the protocol alone does not make an integration secure.

Think of it as the “universal translator” between an AI model and the digital services it wants to use.

What is ACP?

ACP stands for Agentic Commerce Protocol. It is the open commerce protocol introduced by OpenAI and Stripe. Its initial ChatGPT checkout lets an agent work with a merchant’s checkout and payment systems, with the customer confirming the purchase. It does not by itself provide general-purpose price negotiation or remove the need for human authorisation.

ACP makes parts of checkout accessible through machine-to-machine exchange. The merchant still handles the order and the systems must enforce the agreed terms.

How does the Agentic-Layer hypothesis differ from MCP and ACP?

MCP and ACP are building blocks, not the full architecture.

MCP focuses on connection and context (how a single model plugs into systems).
ACP focuses on commerce (how agents buy, sell, or trade value).
The Agentic Layer is the ecosystem that forms once those protocols scale.

In my view: the Agentic Layer sits above the internet, using frameworks like MCP and ACP underneath it. It’s not just about technical connectivity or payments; it’s about a new UX between humans and machines: intent → negotiation → execution.

MCP is the pipe, ACP is the contract, and the Agentic Layer is the environment where all interaction happens (between content creators/publishers and the agentic protocols).

Why does this matter for marketers and brands?

Because the UX is moving away from websites.

Consumers will delegate through interfaces that understand goals, not clicks. Brands will compete on the Agentic Experience (AX) they provide to those agents. Brand data will be surfaced and negotiated through Agentic Interface Protocols (AIP), not search results or landing pages.

If MCP and ACP are the technical rails, AIP and AX are the commercial surface where brands actually compete.