Who Owns the Network (Agentic AI)

I had dinner with my friends last night and we were talking about where the future of agents/AI goes — wanted to share my thoughts on this for mass critical review.

Here goes -- so, Uber worked because your phone could encapsulate your context (i.e. your location) and you could express your preferences (i.e. destination). That's a clean and simple point to point. One request, one transaction with one supplier, in this case the driver.

Agentic AI does this on steroids -- it can encapsulate far more complex context and preference data and execute on a greater scale; the classic example is the AI bot that goes and books your whole travel itinerary for you. BUT that only works well because travel is so heavily digitised in modern tourist cities (e.g. Booking.com). The main thought bubble for me is - what does this look like in less digitised areas of our life?

Which raises a bigger question most service businesses haven't asked yet: who owns that network? If an agent can coordinate across the whole thing, the physical supply — the hotels, the subcontractors, the care providers, the specialist trades — is what actually matters. Whoever aggregates it becomes the network those agents rely on. That's a fundamentally different business from being good at talking to customers.

James Van Geelen from Citrini Research put it well:

"We had overestimated the value of human relationships. Turns out that a lot of what people called relationships was simply friction with a friendly face."

That's uncomfortable, but I think it's right. Travel agencies are the clearest example — an AI travel agent can read your preferences, query every provider at once, and produce a personalised itinerary in seconds. The thing that justified the agency's existence is disappearing.

There's probably a subset of people who genuinely care about human recommendations and are willing to pay a premium for them — and for that group, human service providers will likely stick around. But who knows how many years that lasts. My guess is it fades out with the current generation.

OK, so what actually gets automated?

Think about every touchpoint your business owns — sales calls, onboarding, project updates, service queries. They all sit somewhere on a spectrum between two positions.

On one end: execution work. Scheduling, reporting, standard customer service, booking management. AI will probably absorb most of this — my guess is within the next couple of years or so. On the other end: oversight work. Complex negotiations, high-stakes advisory, accountability when things go wrong. I think this becomes more valuable as the execution end gets absorbed.

But that only answers what gets automated. It doesn't tell you where to go instead.

So should my business care about any of this?

Yeah but probably not for the reason you think.

There's a third position most businesses aren't considering. Underneath both ends of that spectrum, someone still has to aggregate the actual suppliers. The hotels, the subcontractors, the care providers. That network is probably the infrastructure AI agents will end up querying. If you've built it, it's likely hard to displace. If you haven't, it could end up just as exposed as the execution layer.

Think about online payments. Ten years ago, taking a card payment online meant negotiating directly with banks, navigating compliance, building custom integrations. Now business owners just use Stripe. Stripe already did all of that — it aggregated the complexity, and no one goes around Stripe to negotiate their own deals with Visa and Mastercard. The question for your business is probably: can you become the Stripe equivalent of your niche? Can you aggregate the suppliers, the relationships, the local knowledge — so that AI agents just plug into you, rather than going around you?

Ok Dhanish — I'm going to go aggregate supply for my business!

Hold up a second though. I think it only matters if supply in your domain is genuinely hard to aggregate. Think construction — you can't onboard a reliable subcontractor with a listing page. You need to know who actually shows up, who meets code, who holds up when the timeline compresses. That takes years of working alongside people. An AI agent can do the scheduling, but it can't build that network for you.

Contrast that with travel. Airbnb already did the aggregation — self-serve signup, listing page, done. No deep knowledge required. If your suppliers can be captured with a website and a credit card, you don't have a defensible position. You just have another version of the execution layer.

Why I think this is coming faster than people expect

The execution window is short. You may already have good supplier relationships — that's actually your head start. But the risk is that AI-native startups are already moving in this space — home services, construction, healthcare — forming supplier relationships from scratch and building the coordination layer on top. They're not encumbered by a legacy execution business to protect. By the time the execution layer is visibly crumbling, those networks could already be claimed.

Supply networks probably compound. Whoever aggregates first likely earns a data advantage — more transactions, better pricing, richer inventory. a16z's 2026 analysis argues the winners in agentic AI will be the networks those agents depend on, because supply aggregation compounds in ways that raw AI capability doesn't. I think that's roughly right.

Trust could concentrate as the middle hollows out. When every competitor can automate the execution layer, the question shifts to: who do you actually trust when something goes wrong? My read is you'll need to earn that position before your execution revenue disappears — not after.

What does this actually look like in practice?

Back to my dinner conversation... Uber showed what happens when you make it easy to coordinate with one supplier at a time. AI agents can go MUCH further. i.e. they coordinate across an entire network at once. And that's where the less digitised areas of our life come in.

Take aged care. Imagine the agentic AI now coordinates an aged care network. If you're the one who needs care, you can specify "I want help from a cleaner on Monday, a nurse on Wednesday, and a meal delivery on Friday" — and the AI goes and books all of that for you. You've moved from point to point coordination (the Uber example) to coordinating across a network. A part time carer who was idle three days a week now gets a more even spread of work. A meal prep service running at half capacity suddenly has demand it couldn't reach before. Latent capacity across an entire resource/network set gets put to use.

The same logic can apply to household bills. An AI agent querying across every provider could renegotiate your insurance, energy, and subscriptions continuously, switching the moment a better deal exists. More money stays in consumers' pockets, and providers who were previously invisible suddenly have customers.

So the pattern is likely to be... when coordination costs drop to near zero, genuine economic value gets created by connecting what's already there in a far more efficient manner. (or at least let's hope so!)

So what do I or my business do about it?

Every interaction your business owns sits somewhere on that spectrum. Some of it is execution work you should automate yourself, before someone else does it to you. Some of it is genuine oversight you should protect. And some of it might be worth less than you think — friction with a friendly face.

My guess is the ones who get left behind will be the ones defending the execution layer right up until it disappears. The ones who come out ahead will probably have moved early — down into supplier networks or up into genuine expertise — whilst there was still time.

So that leaves us with this question --- how much of what your team calls "relationships" is actually expertise, and how much is friction?

---

Disclaimer: Thoughts are my own and do not represent any other parties.