The agentic web
Web3 gave the internet ownership.
Web4 gives it agency.
Web3 asked who owns the data. Web4 asks who is allowed to act on it - and that turns out to be a much harder question. loob.io follows the shift: autonomous AI agents, the identity and payment rails they need, and what ownership means once software transacts on its own.
One direction, four eras
Each generation of the web added a verb.
- Web1 · 1993ReadDocuments with addresses. Publishing became something anyone could do.
- Web2 · 2004WritePlatforms made participation trivial and quietly kept what was written.
- Web3 · 2015OwnKeys, ledgers and transferable claims. Ownership without a platform in the middle.
- Web4 · nowActSoftware with goals, tools and money. The user stops clicking and starts delegating.
What this site covers
Six threads that only make sense together
Agents, identity, payments, ownership: each is discussed as a separate field, usually by people who do not read each other. They are the same story told from different ends. These are the entry points.
Web4
What the term describes once the marketing is stripped out - and what it does not.
02AI Agents
Agentic AI, autonomous agents, multi-agent systems, computer use and browser agents.
03Voice AI
Real-time speech as an interface - the first agent surface most people will meet.
04Autonomous Commerce
Software that searches, compares, books and settles without a person in the loop.
05AI & Blockchain
Where the two actually meet: identity, payments, provenance and verifiable execution.
06Digital Ownership
Provenance, licensing and authenticity when most content is machine-made.
How the pieces connect
An agent that can only talk is a demo. An agent that can act is infrastructure.
For three years the interesting question about AI was how well a model could answer. That question is largely settled, and it turned out to be the easy one. The interesting question now is what happens when a model is given tools, a budget and a goal, and left to work - when it stops producing text and starts producing consequences.
The moment software acts on someone's behalf, four things that used to be implicit have to be made explicit. Who is this agent? Whose authority does it carry, and how far does that authority reach? How does it pay for what it uses, in amounts and at a frequency no card network was built for? And when it produces something - a design, a contract, a decision - who owns the result, and how would anyone prove it later?
None of those questions is answered by a better model. They are answered by protocols, credentials, ledgers and law. That is why a site about AI agents keeps ending up in territory that looks like blockchain infrastructure, and why a site that started around digital assets keeps ending up talking about AI. The two arrived at the same problem from opposite directions.
Blockchains spent a decade solving trust between parties that cannot vouch for each other. Autonomous agents are the first technology with that problem at scale.
The layers, and how far each one actually is
It is worth being precise about maturity. Some of this ships today and is boring. Some of it exists as a specification with two implementations. Some of it is genuinely unsolved, and the honest answer is that nobody knows yet.
- 08AccountabilityWho is liable when an autonomous agent is wrong. No settled answer in any jurisdiction.Unsolved
- 07SettlementPaying for goods and services, including refunds and disputes. Rails exist; consumer protection does not follow yet.Emerging
- 06DelegationProving an agent acts for a named principal, within a named scope. Standards drafted, adoption thin.Emerging
- 05IdentityA stable, verifiable identifier for a non-human actor. DIDs and verifiable credentials work; almost nothing consumes them.Emerging
- 04CoordinationAgents discovering and messaging other agents across organisational boundaries.Emerging
- 03Tool accessA model calling external systems through a described interface. Solved, widely deployed, boring.In production
- 02InterfaceText, voice and screen control as ways to receive an instruction. In production everywhere.In production
- 01ReasoningPlanning a multi-step task and recovering from failure. Good enough that it is no longer the bottleneck.In production
Read the stack from the bottom up and the pattern is clear: the closer a layer sits to money and liability, the less settled it is. Reasoning works. Speaking works. Paying is starting to work. Being answerable for the outcome does not work at all yet, and that single gap explains most of why autonomous systems are still kept on a short leash in production.
Insights
Recent writing
Longer pieces on specific questions - comparisons, status reports and a few arguments. No newsletter, no gated content.
Web3 vs Web4: What Actually Changes
Ownership was the Web3 question. Execution is the Web4 question.
AI AgentsHow AI Agents Will Change the Internet
Pages were built for eyes. What happens when the visitor has none?
AI AgentsWhat Is Agentic AI?
The difference is the loop, not the model.
InfrastructureDo AI Agents Need Blockchains? A Sober Look
Two of the four hold up. The other two are solved better elsewhere.
Frequently asked
Short answers
Is Web4 an official standard?
No. There is no ratified specification and no standards body that owns the term. Web4 is shorthand for a set of changes that are already happening - agents acting on behalf of users, machine-readable services, and identity and payment layers built for software rather than people. Treat it as a label for a direction, not a version number.
How is Web4 different from Web3?
Web3 was about ownership: who holds the keys, who controls the asset, who can transfer it without permission. Web4 is about agency: which software may act, on whose behalf, and within what limits. Web4 does not replace Web3 - it consumes it. The ownership and settlement primitives Web3 built are exactly what an autonomous agent needs in order to transact.
Are AI agents actually being used, or is this still a demo?
Both, depending on the layer. Agents that read, summarise, call APIs and hold a live phone conversation are in production and generating revenue. Agents that negotiate, purchase and settle across organisational boundaries without supervision are largely still pilots - not because the models are inadequate, but because identity, authorisation and liability are unresolved.
Why does a site about AI agents keep talking about blockchains?
Because the unsolved problems of autonomous agents are trust problems, and that is the problem blockchains were built for. An agent needs an identity nobody can forge, a way to prove what it was authorised to do, a payment rail that works in fractions of a cent, and a record of what it did that survives a dispute. Not all of those need a ledger - but two of them are hard to build any other way.
Who publishes loob.io?
loob.io is published from Vienna by NEOB Technology GmbH, a company that builds AI systems commercially. The site is editorial, not a product page: it exists because the questions are interesting and badly covered, and it says so where a commercial interest is relevant.