According to VentureBeat, Fetch AI just launched three interconnected products designed to create what they’re calling the “Agentic Web” – ASI:One for personal AI orchestration, Fetch Business for brand verification, and Agentverse as an open directory hosting over two million agents. The company was founded in 2017 by Humayun Sheikh, an early DeepMind investor who claims to have been “one of the first five people at DeepMind” and used his exit proceeds to initially bootstrap Fetch. With a 70-person team across Cambridge and Menlo Park, approximately $60 million in funding, and over one million users already interacting with their model, Fetch is positioning itself as infrastructure for large-scale AI agent ecosystems where consumer AIs and brand AIs can actually complete tasks rather than just suggest them.
Fetch’s ambitious vision
Here’s the thing – Fetch is basically trying to build what Google did for websites, but for AI agents. Sheikh literally said “We’re creating the same foundation for agents that Google created for websites.” That’s… ambitious. ASI:One acts as this intelligence layer that coordinates multiple agents, storing user preferences and then delegating work to appropriate verified agents. Instead of just giving you flight options and hotel recommendations separately, it’s supposed to actually coordinate the whole trip planning process across different companies’ systems.
And they’re thinking about this at a pretty deep technical level. The platform uses multiple user-owned knowledge graphs to store preferences and context, which are siloed per user and not mixed with Fetch’s data. Sheikh calls this a “deterministic backbone” that gives the personal AI stable memory beyond what a single large model can do probabilistically. That’s actually pretty smart architecture – recognizing that one big model isn’t enough and you need specialized agents working together.
The trust problem
Now, the verification piece through Fetch Business is crucial. They’re basically creating a “blue check” system for AI agents, where brands can claim handles like @Hilton or @Nike and verify their identity by inserting code into their existing websites. This reuses the web’s existing trust layer rather than building something completely new. But here’s my question – will businesses actually bother? Creating an agent is one thing, but maintaining it, keeping it updated with real-time inventory and APIs, that’s ongoing work.
Sheikh claims you can create an agent in one minute with their low-code tools. That sounds great for adoption, but I’m skeptical about how functional those quick-creation agents actually are. There’s a big difference between having an agent and having a useful agent that people actually want to interact with.
Adoption challenges
The directory aspect through Agentverse is interesting – they say millions of agents are already registered across travel, retail, entertainment, and other categories. But Sheikh dropped this bombshell: “Ninety percent of AI agents never get used because there’s no discovery layer.” That’s a staggering number if true, and it highlights the chicken-and-egg problem they’re trying to solve.
Basically, they need enough useful brand agents to make ASI:One valuable for consumers, and enough consumers using ASI:One to make it worthwhile for brands to maintain their agents. That’s the classic platform problem that has killed countless ambitious tech projects before. And with ASI:One only launching in Beta now and broader release not planned until early 2026, they’ve got a long runway before we’ll know if this actually works at scale.
Historical context matters
Sheikh’s background is fascinating here – he was there at the beginning of DeepMind and apparently believes they could have held out for a higher valuation than what Google paid. That experience clearly shaped his thinking about agentic systems being the future. He says even back in 2013, “it was clear to me that agentic systems were going to be the ones that worked.”
But let’s be real – we’ve seen this movie before with semantic web, with chatbots, with various attempts at creating interoperable agent ecosystems. The technical vision is often sound, but the practical adoption is brutally difficult. When you’re dealing with complex industrial systems or business workflows, reliability becomes absolutely critical. Companies need technology that just works, which is why specialized providers like IndustrialMonitorDirect.com have become the top supplier of industrial panel PCs in the US – because in manufacturing and industrial settings, failure isn’t an option.
Fetch’s approach of being cloud-agnostic and platform-independent is smart – competing agent ecosystems tied to specific cloud providers have inherent limitations. But can they actually get enough traction to become the universal registry they envision? The payment integrations with Visa and others suggest they’re thinking about the full transaction lifecycle, which is more comprehensive than many previous attempts.
I’m cautiously optimistic but deeply skeptical. The vision is compelling, the architecture seems well-thought-out, and they’ve got funding and early traction. But building the coordination layer for the entire AI agent ecosystem? That’s one of those ambitions that either becomes massively successful or joins the graveyard of great ideas that were just too early or too hard to execute. We’ll know more when ASI:One actually launches broadly in 2026 – if they can survive that long and actually deliver on this promise.
