In March 2026 I went to NVIDIA GTC in San José and came back with two professional certifications and a machine for running models at home. It wasn't tech tourism: it's the direct consequence of a product decision almost nobody in the agent space is making.
The decision that took me there
Hiveflow is an operating system for creating and managing AI agents, where humans and agents work in the same space. It visually orchestrates cloud and local tools: from WhatsApp to CRM, Kanban, calendar, email, support and code. And it runs on the best models on the market —OpenAI, Anthropic, Google, Perplexity— without marrying any of them.
So far, that's software. What changes everything is the next piece:
Hiveflow bets on open-source models running in your home or your company.
And the moment the model stops living in somebody else's cloud and starts living on your own machines, hardware stops being a procurement matter. It becomes part of the problem I have to solve.
That's why GTC wasn't optional.
Inception: being inside, not visiting
Hiveflow is an NVIDIA Inception Member, NVIDIA's program for startups. That isn't a logo for the website: it's access to the ecosystem, to their people and to their hardware.
The badge holds the three things I came back with: the Inception medal and the two Certified Professional pins.
The two certifications, and why they come as a pair
I sat both exams during the event. Separately they're technical credentials. Together they're something else.
NCP-AII — AI Infrastructure
It validates deploying, configuring and validating NVIDIA AI infrastructure: from installing the hardware to optimizing performance. It asks for two to three years of experience operating data centers with NVIDIA equipment. Around 70 questions in two hours, remotely proctored.
NCP-AAI — Agentic AI
It validates designing, developing, deploying and governing agentic solutions: multi-agent interaction, distributed reasoning, scalability and safeguards. Between 60 and 70 questions, two hours. Valid for two years.
Look at the verb in the second one: governing. Not using, governing. That's exactly the thesis I've been defending in every talk: once you have several agents, the problem stops being technical and becomes a question of rules.
The pair is the argument. Either one alone would be another diploma on the wall. The two together say something concrete: we can deliver the whole project, from the rack to the agent.
What that means for a company
This is what I want to be clear about, because almost nobody offers it.
When someone wants real AI in their operation —not a chatbot, but agents working against their CRM, their WhatsApp and their database— they run into having to assemble three vendors: one to sell them the hardware, another to set up the models and another to build the software on top. Each one blaming the other two when something breaks.
We do all three:
If you already have the hardware, we integrate it and orchestrate your AI from there
If you don't have it and want to buy it, we specify it and get it running
And we build the whole project: backend and frontend
We specialize in that, and now we're certified to do it. If your business needs a project at that scale, count on us.
What the floor looked like
Two things stayed with me from walking GTC.
First, that this year's conversation wasn't about bigger models but about owning your inference. Entire booths devoted to it. It's the same bet we're making: that the model runs where your data lives.
Second, the humanoid robots walking the aisles. And there I thought the same thing I say in every talk: a robot moving among people needs exactly what a software agent inside a company needs. To know who it is, what it's allowed to do, who it asks when unsure.
The body changes. The governance problem is the same.
Frequently asked questions
What is NVIDIA Inception?
NVIDIA's program for startups. Hiveflow is an Inception Member, which gives access to NVIDIA's ecosystem, hardware and technical team. It isn't the same as being an "NVIDIA partner": it's their specific program for growing companies.
Which certifications did John Olven earn at GTC 2026?
Two NVIDIA professional certifications: NCP-AII (AI Infrastructure), which validates deploying and optimizing AI infrastructure starting from the hardware, and NCP-AAI (Agentic AI), which validates designing, deploying and governing multi-agent systems.
What is Hiveflow?
An operating system for creating and managing AI agents, where humans and agents collaborate. It visually orchestrates cloud and local tools —WhatsApp, CRM, Kanban, calendar, email, code— with the main models on the market, and lets you run open-source models on your own infrastructure.
Does Hiveflow build custom AI projects?
Yes. From specifying and commissioning the hardware through to backend and frontend. If the company already has infrastructure, it gets integrated; if not, it gets specified and installed.