The internet runs on open protocols and open source. So will agentic AI.
That is the headline on the Agentic AI Foundation (AAIF) homepage, checked on 25 October 2026. It connects the Internet’s foundations with AAIF’s vision for agentic AI.
In this series, an agent means an AI application that selects steps and uses tools toward an assigned goal, within the permissions it has been given.
The second sentence remains a proposition to test. It does not establish that every AI product will be equally open, or that products will connect without adaptation. It does raise a useful question for anyone buying business software: if you change AI tools next year, how much of this year’s work can come with you?
Consider a hypothetical industrial equipment distributor. It wants AI to answer a customer’s question: is this order available to ship? The task sounds small. Behind it sit a purchase order, product codes, inventory reservations in the ERP and the sales team’s rules. Some of that is data, some is integration and some is company knowledge. An “open” label does not tell you which parts are portable.
This first article uses the Internet, the Web and Linux to separate those parts. Later instalments will follow the same business task into ERP connections and agent working instructions.
The Internet lets different systems agree on communication
On 1 January 1983, ARPANET switched to TCP/IP, an important milestone in the development of the Internet, as Internet Society records. The relevant lesson here is that different computers and networks can communicate through common agreements. They do not all have to belong to one product family.
A protocol sets rules for how communicating parties send, receive and interpret messages. Imagine two companies agreeing on purchase-order fields and how to report an error. Real protocols specify much more, but the distinction is useful.
An implementation is the software that puts those requirements into practice. Two developers can write different programs in different languages and sell different services around them. Following shared requirements gives their products a basis for communication. Testing still matters: a compatibility claim does not prove that both products support the same version or the same parts of it.
Openness also concerns how the rules are maintained. The IETF describes an open process with public documents and discussions, participation by interested individuals, and experience from working implementations informing agreement. Its standards are voluntary. It does not own or control the entire Internet.
For a business buyer, that suggests looking beyond the existence of API documentation. Who can change the requirements? How are changes announced? Can another company actually build a compatible product?
The Web added a way to access information over the network
The Internet and the Web are different things. The Web is one system that operates over the Internet. Tim Berners-Lee invented the World Wide Web at CERN in 1989. On 30 April 1993, CERN placed the Web software in the public domain and later made a release available under an open licence, according to CERN’s history.
There are two separate decisions here: rules that allow systems to communicate, and rights to use and develop software. They can reinforce each other, but they do different jobs.
HTTP provides an example. RFC 9110 defines semantics for requests, responses and status codes. It describes behaviour; it is not a ready-to-run web server. A company may therefore replace its web server while people continue using their existing browsers, provided the relevant agreements are still met.
HTTP does not know what “available to ship” means at the distributor. Is stock on hand enough? Must reservations be deducted? Does the customer’s credit position matter? A protocol helps carry the question and answer. Someone still has to define the business meaning.
That boundary matters when AI begins using real systems. Successful data transmission and correct interpretation need separate checks.
Linux illustrates the right to develop software further
Linux occupies another part of this story. It is a kernel, the core that manages a computer’s resources, rather than a network communication protocol. It should not be presented as the origin of the Internet. GNU’s documentation dates Linux’s start to 1991 and its change to free software to 1992, years after the TCP/IP milestone above. (GNU/Linux FAQ)
The project documents the Linux kernel’s licence as GNU GPL version 2, with stated exceptions. A complete operating system commonly called Linux also includes other programs, whose licences need their own consideration. (Linux kernel licensing rules)
“Open source” therefore concerns rights, not just visibility. Under the Open Source Initiative’s definition, access to code alone is insufficient. Distribution terms must meet criteria covering redistribution, modification and derived works, among other requirements.
For a buyer, those rights may provide a route to inspect code, modify it or engage another maintainer under the licence terms. They do not remove maintenance work. The company still needs people who understand the software, an update process and someone responsible when it fails. Open source also does not make the company’s customer or commercial data public; access to that data is a separate decision.
Openness has several dimensions. A product may support an open protocol without publishing all its code. Another may publish its code while making data export difficult. The relevant question is which dimensions matter for your work.
