I am offering my services to the world. To the United Nations.
When I was a kid in Canada in the 1980s, Halloween meant going door to door collecting candy. It also meant collecting money for UNICEF.
Pennies. Small change. Whatever people could give.
The way I understood it as a child was simple: there were starving children somewhere else in the world, and the coins we collected could help feed them. I could walk up to a neighbour’s door and do something that mattered beyond my own neighbourhood.
I was a child with very little to give. The contribution was small, but the invitation was enormous: you could help someone you would never meet.
Many Canadians will remember that same tradition. UNICEF Canada’s history describes generations of children collecting candy and coins with its orange Halloween boxes. Canadian families began participating in 1952, and the movement helped bring about UNICEF Canada’s opening in 1955. I was taking part in something that had already connected children across generations. UNICEF Canada’s account.
I cannot trace my particular pennies to a particular meal or child. I can tell you what collecting them taught me: distance did not remove my responsibility to care.
It is 2026 now. I have something different to contribute.
I have more than twenty years of experience in technology. I know how to build networks, automate infrastructure, teach engineers, and help people turn an idea into a working system. I have spent years sharing that knowledge through books, code, demonstrations, classrooms, and workshops.
I am offering my services to the world. To the United Nations. To people working to expand opportunity in their own communities.
I am offering my time and technical expertise voluntarily, without fees, honoraria, or expectation of monetary compensation. My particular interest is helping developing countries, especially least developed countries and underserved communities, strengthen their own ability to build, operate, and govern AI.
This is a personal offer. I would be honoured to help through technical reviews, policy briefings, working groups, workshops, or mentorship, wherever my experience is useful.
And an AI helped me find the opportunity to make it.
My OpenAI Dot brought the PNAI consultation to my attention, unsolicited. I had not asked it to find this opportunity. It surfaced a place where my experience might contribute to a much larger conversation.
I acted on it. I read the draft, developed a comment around work I could actually show, and put this offer of service into words.
That sequence matters to me. An AI helped me notice an opening. I then had to decide whether to step forward, what I could responsibly contribute, and which claims I could support. The responsibility for that contribution is mine.
I would like more people to experience that possibility: using AI to discover where their knowledge might be useful, then doing the human work of participating.
PNAI stands for the Policy Network on Artificial Intelligence, an open network within the United Nations Internet Governance Forum. It brings together government, civil society, academia, business, the technical community, and intergovernmental organisations. Its work is supported by volunteers, and participation is open. About PNAI.
Its November 2026 consultation concerns Bridging the AI Governance Implementation Gap: Beyond Compliance Towards Human-Centred Systems. This is a provisional draft for public comment. Its findings are not a final PNAI position or a United Nations endorsement. Draft and status.
The draft examines 13 cases: 12 from 11 countries and one international initiative. Its lenses are implementation capacity, accountability, and interoperability with adaptation to context. The cases were selected for diversity, not statistical representativeness; only four draw on independent measurement of outcomes. Draft, Chapters 1–3.
Section 3.6, “Conditions and capacities across contexts,” identifies limited public reporting as the most common constraint. It also examines where responsibility becomes fragmented: media provenance depends on platforms retaining information; public AI registers depend on other organisations supplying records. Installing a safeguard and demonstrating benefit to people are different achievements. Chapter 4’s recommendations remain to be developed after consultation. Draft, §3.6 and Chapter 4.
As an engineer, I recognise the problem. A design can look sound inside the boundary of one system. Then the work passes to another organisation, another platform, another operator. Someone else controls the next decision. What happens to the original permission? Where does the evidence go? Who can stop the operation?
That is the contribution I am preparing for §3.6, with relevance to interoperability and the recommendations being developed for Chapter 4.
I develop NetClaw, an open-source AI network engineering agent. With Dr. Nicholas Calcutti and Byrn Baker, I co-authored the NetClaw-to-NetClaw Federation Protocol, or NCFED. It is an individual Internet-Draft with intended Experimental status: work in progress, without IETF endorsement. Internet-Draft and current status.
The question behind it is practical: how can agents operated by different people cooperate while each operator retains authority over their own infrastructure?
NCFED reuses familiar network identifiers and maps MCP and A2A semantics for tools and delegated tasks. An NCFED implementation can distinguish agent traffic from BGP routing traffic on a shared TCP listening port and send each to its protocol engine. Conventional BGP routers do not become agent-aware. External federation requires mutual operator consent; remote tool calls require explicit grants for that peer, which can carry approval requirements. Authentication establishes identity. Authorization governs what that identity may do. Protocol specification, §§4, 9 and 12.
For a reader outside networking, the important idea is that cooperation should preserve the authority of the people responsible for each system. I want that principle to remain visible when the participants are software agents.
One recorded experiment shows both the possibility and the limits. A Hermes agent requested a read-only inspection of another operator’s Cisco Modeling Labs environment through the federation. The returned observations said the server was reachable and licensed. A mismatch between the remote tool’s expected data structure and the server’s response prevented a complete lab inventory. The agent reported that failure and named what it could not confirm. The recorded session reports no lab-state changes. Recorded session.
That incomplete result belongs in the story. Anyone evaluating the workflow should be able to see where it stopped being reliable. A confident summary that quietly omitted the failed inventory would be less useful to the next engineer and to anyone reviewing the work.
The project preserves terminal captures and a packet-capture file in a SHA-256 evidence manifest. Those artifacts offer material to inspect. Matching a hash establishes file integrity relative to the manifest; it does not establish the truth of every statement on a screen. Encrypted traffic does not independently reveal the authorization decision. A read-only trace does not demonstrate a human approving a production change.
These are developer-provided experimental observations. Their value is in making a bounded claim available for scrutiny.
My recommendation is a common evidence record for work delegated across agents and organisations. For a representative operation, I want a reviewer to be able to answer five questions:
- Who held authority? Identify the participating operators and the system responsible for permitting or refusing each action.
- What autonomy was allowed? State what the agents could observe or execute and what required a human decision.
- What permission actually applied? Record the grants and approvals relevant to that operation, including their scope and validity.
- What happened? Distinguish attempted, completed, refused, failed, and uncertain steps.
- What can someone else verify? Identify the available evidence and tests, including denial, revocation, and failure cases.
A public summary can protect credentials and sensitive infrastructure details while explaining what evidence is withheld and how an authorised reviewer could examine it. For a team with limited resources, I would start with one representative workflow and a clear record it can maintain. The documentation needs to be usable by the people doing the work.
Then comes the harder question: did that workflow improve anything for the people it was meant to serve? We would need appropriate outcome measures, a basis for comparison, and the perspectives of those affected. I would apply that expectation to my own projects.
This is also why I care about who gets to build AI in the first place.
When I say I want AI in the hands of the developing world in 2026, I mean access to the knowledge and practical means to make decisions about it. People should be able to investigate a tool, adapt it to their language and circumstances, understand its costs, challenge its output, and decide whether to use it.
The people who understand a community’s needs should help define the work and judge its value. My contribution would begin by listening to the engineers, educators, public servants, and community organisations already there. I would want local instructors able to continue teaching and local operators able to maintain what we build together.
There are material conditions to address. ITU’s Facts and Figures 2026 estimates that two billion people remain offline. It puts Internet use at 93 per cent in high-income countries, 32 per cent in low-income countries, and 41 per cent in least developed countries. Those last two categories are distinct. ITU’s 2026 connectivity estimates.
That brings the ambition directly into my profession. Reliable networks, affordable connectivity, available devices, electricity, and people who can maintain systems all matter. So do language, accessibility, ongoing operating costs, and control over data. An AI workshop needs to account for the conditions in which its participants will keep working after it ends.
The United Nations’ Global Digital Compact already calls for AI capacity-building partnerships, education and training, access to open resources, and local solutions that reflect linguistic and cultural diversity. I see a place for practical teaching and open-source engineering within that work. Global Digital Compact, Annex I, §§60–62.
UNDP’s 2025 Human Development Report places people’s choices at the centre of the AI conversation. That is the ambition I want to serve: expanding what people can choose and accomplish in their own lives. UNDP, A matter of choice.
And UNICEF, the organisation I collected coins for as a child, is part of this present-day work too. Its joint Giga initiative with ITU works toward connecting every school to the Internet, alongside the wider need for skills, devices, and useful content. ITU on Giga.
UNICEF Innocenti’s September 2026 report Skills for an AI World draws on consultations with 97 children aged nine to seventeen in five African countries. The children raised concerns about false information, privacy, and dependence on AI. They valued critical thinking and learning skills themselves. These are the perspectives of the children consulted, rather than a survey representing an entire continent. UNICEF’s report and findings.
I find that connection moving. The child who collected pennies has become an adult who teaches technology. Children today are already telling us what they need to navigate that technology. They deserve to be heard in the decisions about it.
I have some relevant experience to bring. I spent more than eight years serving Canada’s House of Commons in infrastructure architecture, technical advisory, and senior IT planning roles. I later worked as a Cisco AI Technical Leader, and I now serve as Head of AI and Developer Relations at Itential. Public-sector responsibility and practical infrastructure work have shaped how I think about the consequences of technical decisions. Professional background.
I am the author of Automate Your Network and co-author, with Danny Wade, of Cisco pyATS—Network Test and Automation Solution. My first book is freely available to read and download. I publish code and educational material openly because access to knowledge has been central to my own career. Books and free reader.
At my most recent AutoCon workshop, approximately 75 engineers came together to build with AI. It was my third consecutive sold-out workshop. What stays with me is watching people discover that they could make something themselves, then consider what else they could do with that ability. Workshop account and materials.
I want to help make that experience available in more places. A useful starting point might be a workshop designed with local instructors, a small lab that engineers can run themselves, or a technical review that helps a public institution understand an agent’s permissions and limits. The right project would come from the people asking for help.
My offer to PNAI, other United Nations bodies and agencies, and interested Member States is to bring that experience in a personal, voluntary capacity. This is where my offer stands today: I am putting myself forward to help, and I would welcome a conversation about a concrete need.
You can do something with your own expertise, too.
Use AI to help find a consultation you can contribute to, locate an open-source project that needs your skills, or understand a problem raised by a local organisation. Read the original material. Talk with the people involved. Choose a contribution you can stand behind. Teach someone how to check the result, and leave them with something they can use and maintain.
For this particular opportunity, the PNAI comment deadline is Friday, November 27, 2026. The official participation page explains how to comment through the review platform or by email, with both routes given equal weight. It also explains how to join the network’s ongoing work. Check that page for updates if you are reading this later.
The report’s own AI-use note keeps responsibility for accuracy and integrity with contributors. That is a useful discipline to carry into any contribution: understand the process, verify the evidence, and own what you submit. PNAI’s AI-use note.
I cannot promise where my offer will lead. I can decide to make it, be specific about what I can do, and follow through when there is a useful way to contribute.
In the 1980s, I went door to door with candy to collect and pennies to give. I learned that helping could begin with what was already within my reach.
In 2026, I have knowledge, experience, tools, and time. AI helped me find a door worth knocking on.
I am offering what I have.
John Capobianco
Head of AI and Developer Relations, Itential · Creator of NetClaw · Author and educator
For this personal, voluntary offer: ptcapo@gmail.com