From Intent to Purchase Backed by @Betaworks
New York, NY
Joined June 2024
- Tweets591
- Following4
- Followers669
- Likes1.6K
Pinned Tweet
A glimpse of what Ophelia API can do across an @OpenTable reservation request in a headless false setup.
It’s the same execution everywhere.
Travel (@bookingcom), tickets (@Ticketmaster), fitness, transportation (@Uber, @lyft), and other experience + service supply.
Ophelia retweeted
intimacy is the most valuable data ever collected, and personal ai is about to collect it at scale.
@jmmoreno3003 and i tested Emma by @nuance_ai, and what became obvious was how personal ai merges two things: the same product starting to know you, can now also act for you.
for most of history those were separate companies.
putting those together concentrates a lot of power. the only real question is who holds it.
we think the end-user should.
your intent stays on your device. the world gets only what a task needs. every action is checked, and a person steps in when something breaks.
but i think the ai companies that matter most will be the ones that refuse to centralize you.
what you want stays on your phone.
nothing about your personal data follows you into the next app or interface, and when something goes wrong, an agent yes acts on behalf of the human, but at the core of each consumer company - a human authorizes all of it.
we’ve noticed this becomes especially true for financial or medical transactions. @opheliaapi
The first AI persona you’ll actually want to meet face-to-face.
Most AI personas stare blankly while you talk. Emma reacts while you’re still speaking.
After 18 months of research at @nuance_ai, this is Emma, built on Nuplex, our full-duplex audiovisual foundation model.
Ophelia retweeted
The upside I want is that more people can afford to try their own ideas.
I think the difficulty describing a positive AI future starts with how little time we spend understanding the people we’re building it for.
We automate customer service while we’re still learning why people need help. Then we lose the conversations that would have told us what to fix.
We build faster than we learn how the systems we’re replacing actually work.
There are people in those industries who could explain why something works that way, including which parts they’d love to change.
But if AI makes running a business much cheaper, someone with a very particular idea for a device, a car or a travel service has a better shot at starting it.
People still work because they care about what they’re making, and other people choose whether it’s worth their time and money.
Cheaper intelligence gives us more ways to use them. But ownership, data, and the right to say no still need to belong to people.
Thiel’s definite optimism asks us to be specific about the world we want to build. I think that requires spending considerably more time in the world people already inhabit.
The consumer products dressed up as enterprise companies are especially interesting to me. An enterprise signs the contract, and an everyday person lives with the result.
At our core building @opheliaapi so someone building their own platform or product can give it the ability to authorize, purchase, or result in an outcome connected to not only real businesses, but also the real-world.
There isn’t a feature or a decision I make without wanting to know what happens to the person on the other side.
Two things I’ve been thinking about regarding AI:
• People often highlight the risks involved in AI, which are surely very real. I’m struck, however, by how hard it is to understand the contours of the positive scenario. There are easy-to-perceive and frequently invoked elements of a positive vision — more diseases cured, more material abundance, etc. — which sound great, but these are by themselves just ingredients. I’m curious what the larger equilibrium is supposed to look like. Do people work? If so, why? What is their marginal contribution? If not, how are resources allocated? What is scarce? What is the ratio between humans and AIs? Are humans in control, whatever that means? One sometimes hears the Culture novels invoked as one scenario, but I’m not sure they depict a desirable outcome. (AI runs everything and humans compete in games? Iain Banks himself said that the humans in the Culture are “somewhere between passengers, pets and parasites.”) There is Richard Brautigan’s famous poem, but do we want to be “returned to our mammal brothers and sisters”? (Isn’t that rewinding Genesis?) We could say that it was hard to imagine the world of 1900 from the vantage point of 1700, and that’s assuredly true. But nobody in 1700 was trying to recast humanity itself within a couple of years — things necessarily unfolded slowly and by increments, and therefore with the possibility for error correction. Given how rapid the rate of progress in AI may be, it feels strange for the upside case to be so opaque.
• There is increasing discussion of whether AIs are conscious or how we would know whether they are, with conclusions there often assumed to be relevant to matters of moral worth. At first blush, this seems a rather arcane, benign, or even absurd issue. (Remember the response to Blake Lemoine’s fears.) Setting aside the terms of the debate itself, or the relevant demarcation criteria, it seems to me a quite high-stakes question, for an affirmative answer — that is, granting moral worth to AIs — would presumably mean that the aggregate worth of AIs would soon outweigh that of humans in any utilitarian calculus. The AIs will be exceptionally wise and intelligent, and vastly more numerous (especially once one considers the imminent future and the AIs not yet born). As such, even epsilon of moral worth will multiply to a lot once one considers the quadrillions that could soon exist. (This is effectively a distributed version of the Nozick’s utility monster.) If this conclusion is unpalatable, I think one of three propositions needs to be rejected: the premise AIs can have moral worth; utilitarianism itself; or the belief that there is even a possibility that they will be far more numerous than humans.
With both of these matters (contours of future society; AI moral worth), there is the possibility of dramatic change. Today, people argue about datacenters. I’m curious how popular support for AI will evolve as more people come to understand these medium-term questions and the relevant stakes. What is the vision for AI that most people would be excited about?
Building with @alexyango + the Frida team keeps pushing us to make @opheliaapi better. Every new capability we ship meets a team eager to build with it. We love witnessing such a relentless in product development.
Watching that ambition turn into progress has been a highlight for our team. LFG
Frida is a personal agent designed around intuitive, native interfaces for everyday decisions, making the ordinary feel a little more delightful
Now available on the App Store in the US and Canada.
apps.apple.com/us/app/frida-…
Ophelia retweeted
one tiktok campaign away from becoming a unicorn
Frida is a personal agent designed around intuitive, native interfaces for everyday decisions, making the ordinary feel a little more delightful
Now available on the App Store in the US and Canada.
apps.apple.com/us/app/frida-…
Ophelia retweeted
So extremely proud of @alexyango + his team’s progress on Frida. It’s very rare nowadays finding builders who care about design, product, and clean code - especially in the consumer AI space. They’re also the fastest at expanding what Frida can do.
A lovely thing about building with them: every time @opheliaapi ships a new capability, they're eager to put it to work.
Frida is a personal agent designed around intuitive, native interfaces for everyday decisions, making the ordinary feel a little more delightful
Now available on the App Store in the US and Canada.
apps.apple.com/us/app/frida-…
Ophelia retweeted
Legit the opposite of agentic progress…
Making agents learn every merchant separately.
If agents need a separate integration for every merchant, you’re only deepening systemic fragmentation and becoming slower as a product.
It's pretty magical when your agent...just works
@instinct just partnered with @OpenTable, so it can now book and manage your restaurant reservations
No blocks, not TOS issues, and @link deals with the card holds, deposits, and other payment flows
Watch me make dinner plans 👇
Ophelia retweeted
The biggest mistake in agentic commerce is assuming the scarce resource is money processing.
Moving money from a bank account to a merchant is one of the most solved problems in software. A card, wallet, bank account --> they're all just different ways of getting money from one place to another.
So an “agentic card” doesn't fundamentally change that. It just gives the same payment system a new interface, and furthermore doesn’t make an agent autonomous.
It's also why I don’t believe agentic cards or wallets are the fundamental primitive of the agentic economy.
In 1,217 attempts across @Visa and @Mastercard cards, we’ve seen a 99.2% success rate completing purchases without persistent sessions, using a simple user request and our intent layer to guide model decisions across systems.
Because of that, it's why I believe authorization is going to matter more than the payment rail.
Who is allowed to do what, on whose behalf, across which systems?
It's the area between AND above intent and execution
This is also why I don't think browser use is the long-term answer. And why I’m skeptical of the industry's obsession with computer use too.
The next generation of software (more specifically consumer software) won't care whether a request came from an iPhone, a car, a voice command, or another agent.
It will care whether that request is authorized, and what that agent was authorized to do.
Agentic commerce should be built towards giving software the ability to act, not inventing another way to pay.
Thus, the future of agentic commerce will not be decided solely by Visa or Mastercard.
Ophelia retweeted
we're so early on ai that “buys things” for you.
today I asked Fo for one @reformation top in my size + color (even gave it the actual link to the item).
1 / then got redirected to their own login page, so i could signin AGAIN to Fo then again to Reformation
2 / then said I didn’t have a Reformation account (I use my login with Google for credentials + am VIP clientele)
2 / Amazon didn’t work - Fo said right away that it couldn't place any Amazon orders or add items to cart
as a previous consumer founder, the logins, third-party tools, and workarounds were seriously what wore me down the most.
the product i wanted to build was always waiting on plumbing. one more partnership deal, another integration, back-end build that slowed everything else down, it just never ended.
it’s why we began building @opheliaapi in the first place and pivoted later on.
we want the next wave of consumer founders + teams to get better tools than we had.
weeks spent on building (esp in consumer) should be spent on the part people fall in love with marketing, gtm, product, branding, etc.
not sure why or what it demonstrates about our market that teams are claiming to “do everything” and absorb everything in-house, it’s not scalable.
but more importantly it creates thinner, weaker technology.
Fo with Twilio 🤝
Wajo AI’s @hey__fo is transacting globally, and booking goods and services, with phone calls, texts, emails and humans in over a 100 countries since launch 🚀
Fo has emerged to be not only a productivity assistant, but the trust first, AI fulfillment layer to people's lives.
• Be their EA on their morning commute - calendar prep, chasing lose ends
• Manage their household - in group chats, and coordination
• Book and manage reservations, travel, shopping, messy government forms
• Sit on hold with call centers and customer service
• Navigate healthcare systems
• Negotiate with car dealerships, hunt apartments
We are excited to enable this over the phone stack with our partnership with @Twilio that powers @wajo_ai’s @hey_fo's telephony stack for our users and we're proud to be building this alongside their team with more to come!
Sign up and let Fo get sh*t done for you today: wajo.ai
Ophelia retweeted
Simon is a goat.
But you can admire someone without agreeing with every conclusion.
I think reducing agentic commerce to cards, wallets, or new payment rails misses the bigger problem, and is a huge mistake.
Payment is one part of a transaction. It doesn’t solve how an LLM figures out what someone wants, decides where to get it, navigates the business, and actually completes the purchase.
So it’s neither inference or only execution, not even solely wallets or only cards.
I also don’t love the question “how do agents pay?” It assumes the payment relationship has to change just because the interface does.
Interfaces vary - and most will die.
The end customer is still a person. Their card, account, and authorization don’t suddenly disappear bc “software” or an agent is doing the work.
Look at what Meta’s Muse and Grok are doing with shopping. They’re asking users for their existing payment credentials.
That’s telling.
We’re already seeing that the customer doesn’t necessarily need a new wallet or a new financial instrument. The hard part is everything the agent has to do before and around the transaction.
And bolting on a separate wallet, payment rail, identity layer, routing, access API, and agent framework for every new use case sounds like a pretty messy way to build this.
I myself am even getting overwhelmed.
I can already tell most are making the mistake of taking infrastructure designed around human-driven behavior and systems, then assuming “agents” will simply “slot” into it.
Bubble behavior. And the even bigger mistake is assuming agentic commerce has to fit into the categories we already know.
We’re still figuring out what this market becomes. I’d be careful about deciding what it looks like before we’ve built it.
It assumes the future will look like the present, just with agents added.
I wouldn’t be so quick to put it in any box, yet.
Even more importantly, I love Simon and have nothing but respect for his opinions and views.
Let the Gods decide 😂
Ophelia retweeted
Amex knows.
Very excited to collaborate with the @AmericanExpress team and work on AI consumer experiences!!
We’re spending time on how to turn user intent into a great experience consumer, focused on helping platforms understand what users actually want, act on that intent, and keep the transaction in-app.
Built a way that feels native, empowering developers to innovate and users to authorize.
Excited to keep building together.
Grateful for the time, ideas, and conversation. Thank you @betaworks for letting us take over the legendary Fridenza :)
Ophelia retweeted
Very excited to collaborate with the @AmericanExpress team and work on AI consumer experiences!!
We’re spending time on how to turn user intent into a great experience consumer, focused on helping platforms understand what users actually want, act on that intent, and keep the transaction in-app.
Built a way that feels native, empowering developers to innovate and users to authorize.
Excited to keep building together.
Grateful for the time, ideas, and conversation. Thank you @betaworks for letting us take over the legendary Fridenza :)
Ophelia retweeted
a real-world request is almost never: “execute API call X.”
it's: “handle this for me.”
when an agent's understanding of systems gets better at directing future actions from actual behavior -->
autonomous products can become more than a smarter chatbot.
developers want infrastructure for turning open-ended human intent into authorized decisions and completed actions.
tools that create magic.
99% of the time - the smartest most talented ones are actually picky about what becomes part of their stack.
they don't want to build their company around whatever AI abstraction happens to be popular this month.
they choose the infrastructure that gives them more control, not another layer between them and their product.
that's why they choose us.
Ophelia retweeted
The obvious bets are:
1 / Models get smarter.
2 / Agents get more capable.
3 / Businesses expose more APIs.
We're betting that none of those solves the problem of deciding what should happen on behalf of a person.
The mistake is assuming the internet will be rebuilt for agents.
1 / Amazon will make Amazon agentic
2 / Shopify will make Shopify agentic
Every major platform will build its own agent interface.
But the more interesting question is there are millions of businesses that will never build an API for every model, agent, device and platform that wants to interact with them.
Someone has to bridge that gap.
Our bet is that the internet doesn't get rebuilt for agents, rather agents learn to operate on the internet that already exists.
Everyone assumes the future of software will be built inside the largest platforms.
We think the opposite: the most valuable economic activity will eventually happen across a long tail of businesses that no single platform controls.
Ophelia retweeted
websites are its own interface
which is why browser-use is not the solution.
intent should be actionable for agents.
The mistake is assuming the internet will be rebuilt for agents.
1 / Amazon will make Amazon agentic
2 / Shopify will make Shopify agentic
Every major platform will build its own agent interface.
But the more interesting question is there are millions of businesses that will never build an API for every model, agent, device and platform that wants to interact with them.
Someone has to bridge that gap.
Our bet is that the internet doesn't get rebuilt for agents, rather agents learn to operate on the internet that already exists.
Everyone assumes the future of software will be built inside the largest platforms.
We think the opposite: the most valuable economic activity will eventually happen across a long tail of businesses that no single platform controls.
The mistake is assuming the internet will be rebuilt for agents.
1 / Amazon will make Amazon agentic
2 / Shopify will make Shopify agentic
Every major platform will build its own agent interface.
But the more interesting question is there are millions of businesses that will never build an API for every model, agent, device and platform that wants to interact with them.
Someone has to bridge that gap.
Our bet is that the internet doesn't get rebuilt for agents, rather agents learn to operate on the internet that already exists.
Everyone assumes the future of software will be built inside the largest platforms.
We think the opposite: the most valuable economic activity will eventually happen across a long tail of businesses that no single platform controls.
Ophelia retweeted
This is what it looks like to “think”. Rather than build on top of yesterday’s ideas, why not start with first principles in light of new technology?
Connectors feel like the obvious way to make the internet agentic.
I think they’re a dead end.
The internet won because you could go anywhere without asking the destination to integrate with your browser first.
Agents should work the same way.
If every service needs its own connector, you haven’t democratized access, you’ve created a toll booth in front of every website.
The internet doesn’t need to become API-native for agents to use it. And merchants shouldn’t have to change anything for software to transact with them.
Ophelia retweeted
Connectors feel like the obvious way to make the internet agentic.
I think they’re a dead end.
The internet won because you could go anywhere without asking the destination to integrate with your browser first.
Agents should work the same way.
If every service needs its own connector, you haven’t democratized access, you’ve created a toll booth in front of every website.
The internet doesn’t need to become API-native for agents to use it. And merchants shouldn’t have to change anything for software to transact with them.
Meta's Muse is getting shopping carts powered by connectors. Instacart is going to be first. There is a long list of retailers including Walmart that said theirs are coming soon, too.
This is different from ChatGPT/Claude plugins. Those render the plugin's visual output as an in-chat mini app. Instacart's connector on Muse will only handle data exchange - Muse will render the shopping cart and other UI using Muse-native widgets. (screenshot from the connector's docs)
Ophelia retweeted
If you focus on intent —> payment, payment methods, cards, browsers, servers, type of model, none of it matters.
100% successful outcomes are possible when the model can learn in real-time from intent
Ophelia retweeted
is your api infra propagating intent to every merchant accepting transactions and every issuer authorizing transactions? is your api infra deciding who takes the liability?
yall are doing amazing work (check out @opheliaapi!) but the core misalignment problem is so massive one product isn’t going to fix it!
Ophelia retweeted
Here's my issue with most agentic products right now:
Everyone is racing to give agents access: virtual cards, API credentials, prepaid balances, tokens, proxies.
But I think we're getting ahead of ourselves.
I'm noticing many teams spending a lot of time figuring out how agents get into existing systems, without really addressing what happens once they're there.
I'm not convinced the problem is simply how an agent "pays" or checks out.
I think the harder question is how an LLM understands what a person actually asked it to do (intent), what the underlying system permits it to do (execution), and whether those two things are even the same (benchmark + auditing).
It reminds me a lot of the early EDI days with Amazon and the supply chain in the 90s. Getting different systems connected was one problem, but getting them to actually understand and coordinate with one another is an entirely different one.
A few things worth noticing:
1 / payment can clear and still be the wrong purchase
2 / credential can be valid and the transaction can still be wrong
3 / access doesn't equal authorization + authorization doesn't equal intent
I don't think a card or another API wrapper gets us much closer to solving this.
What concerns me is that we're approaching a systemic problem through a collection of isolated solutions.
You have to understand the entire transaction lifecycle: the consumer, the merchant, the financial rails, permissions, exceptions, failure modes, and what happens when something inevitably goes wrong.
This is also why our team has been spending so much time on the underlying financial infrastructure, particularly for the next generation of consumer products that will increasingly act on behalf of people.
There's a broader point here about engineering that I think gets lost in all the excitement around AI.
It's never been easier to generate software, except I don't think that means we've made engineering any less difficult.
If you've actually spent time architecting and maintaining software, you know how quickly seemingly insignificant decisions compound.
A shortcut in one layer becomes a problem somewhere else, an abstraction that works in a demo becomes a constraint in production.
Eventually, someone has to understand the whole thing well enough to fix it.
The engineers I respect use AI to move faster, audit their work, test assumptions, and challenge their own thinking. But they aren't outsourcing their understanding of the system to a model.
You can use AI to get there faster, you cannot throw tokens at your way into great architecture.
Infrastructure is one of the few areas of technology where the underlying engineering still has to be deliberate, considered, and built to last.
I think that's going to become increasingly obvious as this category matures.
Great products are built on great infrastructure.
