Why Agentic Wallets™?
One thing about working alongside a visionary like Sir Tim Berners-Lee for long enough is that you get to see the rest of the world catch up to the vision.
When Tim and I founded Inrupt, I said we were heading into an “unclear but certain future.”
We saw — well, Tim did — how the web was going to end up. One giant stepping stone that’s fallen into our path earlier than we expected was the arrival of powerful artificial intelligence.
Eight years ago, Tim wrote about “Charlie,” the name he coined for a new kind of personal AI assistant. We reckoned we’d see Charlie in twenty years. It would be way more capable than the then state-of-the-art assistants like Siri and Alexa. In Tim’s vision, Charlie could do things in a very personal way. It could create recipes based on the user’s diet and recent restaurant orders. It could develop tailored workout regimens based on the user’s fitness and health data. It could arrange plans or book meetings based on the user’s calendars and transportation logistics. It could even make the necessary adjustments if things changed at the last-minute.
Sounding familiar?
2025: “The Year of AI Agents”
Large Language Models now have the raw intelligence it takes to make Charlie possible, a decade ahead of when we guessed. But that’s not enough.
These LLMs are mostly trained on publicly available data. Today’s consumer-facing products — ChatGPT, Claude and Perplexity for example — connect to search results on the open web to bring in current public information. Corporations, on the other hand, can train or fine tune models using their own proprietary data in order to create industry-expert AIs.
But, unlike corporations, consumers interacting with AI have no safe way to bring information relevant to them into the mix. Anything they need the AI to know about what they’re trying to accomplish, or how they like things done, they need to tell it, explicitly, mostly by typing into a little chat window.
Vital in Tim’s vision is that Charlie works for us, in ways that are personalized to us. Like any good human assistant, an AI agent needs more than smarts. It must be loyal, trustworthy and understand our particular wants and needs. To do that, individuals would have to give Charlie access to any and all of their personal data, so that it can deliver personalized experiences with immense utility. Charlie would be trusted to autonomously make decisions, initiate actions and follow through in a way that’s tailored to our unique needs.
But unlike the way we provide our everything to LLMs today, Charlie keeps you in control of what you let them see.
Introducing: Agentic Wallets™
Safe and personal are the two foundational elements of Charlie — and digital products more broadly — that Inrupt set out to build. And we’ve done it. We call them Agentic Wallets.
Agentic Wallets™ are built to integrate with any intelligence in order to create AI products that are smart, safe and personal. They bring the whole universe of data together via one channel, making it available to the AI in order to get the best outputs with the least effort:
This includes any information available on the web. But it also includes private data of all kinds. Buying preferences, medical records, IoT device data, pictures of your cat — whatever. And it integrates the transactions and credentials familiar from digital wallets today. Things like credit cards, boarding passes, tickets and increasingly, mobile drivers’ licenses (mDLs).
Bringing all this data together for each user via one channel is the crucial step in making AI personal and safe for customers. People work at the intersections of public, private and transactional data dozens of times a day in order to get things done online. But AIs can’t easily cross those boundaries. A simple instruction like “Make me a meal plan for this week” becomes much more powerful when the AI has access not only to internet recipes (public data), but also context-aware of my dietary restrictions (private data) and knowledge of what I bought at the grocery store yesterday (transactional data).
Privacy & Security for the Age of AI
Obviously there are privacy and security considerations here. We can’t let the AI treat my personal grocery receipt — let alone the heart rate data from my wearable — the same way it treats online recipes.
Data access and safe permissions across boundaries become more important as AI systems become more capable. Today’s consumer-facing AIs are mostly still text-based chatbots, but that’s rapidly changing. That’s why we’ve built Agentic Wallets™ from the beginning with transparent, flexible user controls.
Agentic Wallets™ allow the user to control which data can be shared with the AI, and for which uses. In some cases, users can get the benefit of personal AI outputs without giving up their underlying data at all. This is essential for building the trust and alignment between people and AI assistants that make capabilities like Charlie’s possible.
We’re starting to move up the value ladder of things that AI assistants can do for us. Not just answer our questions about recipes, but actually order groceries to make the meal. And soon, to look ahead at our calendars and volunteer a suggestion for the dinner party we’re hosting next Friday.
But there will be hurdles along the way. For now, OpenAI’s Operator agent, for example, can navigate a website and add items to your shopping cart, but it can’t log in or add your payment details for you.
Agentic Wallets™ are built to overcome hurdles like that, and others ahead, in a safe and intuitive way. More importantly, it works at scale.
The technology is certain, and the end state for consumers is becoming clearer — a web experience intermediated by safe and personal AI assistants that work to make our lives easier.
So how do we get this tech into the hands of billions of people?
What This Means for Businesses
An AI assistant that handles everything from scheduling to shopping to finance to household management and healthcare. In that world, every business on earth better figure out how to interact with their customers’ Agentic Wallets™, since it will be the only way their customers interact with them.
That world opens a massive opportunity for direct two-way engagement with each individual customer. Via the one channel where they organize their lives, make decisions and buy. No more platform intermediaries setting the rules of engagement and taking the lion’s share of the value for themselves.
Another possible end state — and at least an intermediary one that’s already underway — is that businesses deliver their customers AI assistants tuned with the functions and expertise of a particular industry. So an individual will have a “Charlie” for managing their travel and another for managing their money. These AI domain-limited assistants are the natural evolution of today’s apps.
But the methods businesses use today to “personalize” their app experiences (which if we’re being honest are pretty limited), won’t work for AI agents. Today, users sign up for app with, at most, a few basic pieces of information — name, age, address. Then the app maker infers their preferences by tracking scrolls and button clicks, and comparing them to the scrolls and button clicks of millions of other users.
In this world, the user is doing all the work. Clicking and scrolling, choosing and navigating their way through a limited set of options presented to them. All the while they keep their personal intentions and preferences in mind.
AI agents can’t work like that. The set of options and capabilities won’t be limited by an app UI. Personal preferences and intentions need to be directly expressed to the AI system if the business is going to have any hope of giving the customer what they want. Agentic Wallets™ are the way that businesses can receive the direct input they need from their customers.
As the digital landscape shifts, businesses that embrace this soonest will be the ones best positioned to lead in AI-driven customer experience and digital transformation.
The opportunity is immense. The future is open. Let’s build it together.
