Standardizing the “Agent-to-Human” Connection

Hey everyone, it's Alan Schimmel. Welcome back here to Techstrong TV. I want to introduce you to Zach Hanif.

I hope I got that right, Zach. Yeah? You did indeed.

You did indeed. It's great to meet you, Alan. Nice to meet you, Zach, and thank you for joining us here on Techstrong TV.

Zach, by the way, is head of AI, ML, and data. That's a lot. Yeah.

A lot of stuff right there. There's a lot that goes in there, but they all work together. For Twilio.

Yep. Yeah, yeah. Well, it does.

And, and as I mentioned, Twilio, a company we've been following a long time here at Techstrong. Zach, that is a lot. You don't look that old.

You've been doing this, though, a while. Tell us, like, how did you wind up here? What's your path been like?

So one, I don't look that old because fortunately the camera I have lies a great deal on my behalf. You got a good camera. So that's, that's always the big benefit.

A good ca- That's a lesson out there for you people watching. A good camera always helps. But go ahead.

Uh, two, my path has, has gotten me here because ever since the early days of my career, the thing I knew I wanted to do was work at, with data and AI and ML at large scales. That was... It's been my entire career from the beginning.

I have been blessed with the ability to do that across a wide number of domains, everything from internet security to anti-money laundering to fraud. I've worked in financial institutions, fintech, I've worked for general purpose, large scale internet companies, all sorts of great stuff, right? Uh, I'm very grateful to be here at Twilio right now.

You get a huge amount of perspective and challenge inside of a space that up until this point I haven't, spent a lot of time with, and it's a wonderful opportunity to work directly with customers. Love it. I love it.

And i- you know what? Sometimes just things just ca- ca- you know, line up right. What a, what a great time for you personally with your background to be living right now in what this kind of revolution that we're, we're living through in terms of AI and, and everything that goes with it.

So- The last fifteen years have been very exciting. That's absolutely true. Absolutely.

Good for you, man. Um, talk to us about Twilio. How long you been at Twilio?

So, excuse me. So I'm coming up on just on about two years now, having been- Mm-hmm ... at Twilio, so not all that long as, as far as we kind of measure time and material like that.

Um, but the company's gone through, like, a multitude of, of, of evolutions, and it really feels like it's on a incredible path forward, over the la... you know, having been built up for the last couple of years and, and looking into the future. So most people, I think, actually have interacted with Twilio without knowing it.

One of the things I like to say is that if you've gotten a text message from, like, your dentist or from any kind of brand that you've worked with, if you've gotten an email, from an organization that you're subscribed to a mailing list on, statistically, it's pretty likely that within the last week you've gotten a couple of messages sent by someone who you want to interact with using Twilio. Twilio's o- one of the leading, if, if not the leading, companies that build out the capability to do communication between organizations and their audiences through channels, SMS, email, whole bunch of them. What most people don't realize is that we have an incredible amount of capability in the area of customer engagement in general, and have been building that out quietly for the last couple of years.

So Twilio's got this, great, union of things between the knowledge of what customers, what our customers want to communicate with their audiences about and how they're doing that communication. And so that means we've got the ability to have a, you know, very, very powerful customer engagement platform that allows our customers to have direct personal relationships with their audiences, no matter where those audiences are, and no matter how those audiences want to be communicated with. And it gives us the ability to make sure that as time goes on, and this is what drew me to Twilio, that puts us on a position where we can add intelligence to functionally every interaction and trust to every part of a customer's journey.

It's, it's an incredibly large lever that honestly most people never even think about, but has a huge ability to kind of influence how we do business on a day-to-day basis. I love it. Man, like I said, what a great time to just be alive though, huh?

Yeah. Um, Zach, of course, you know, as you mentioned, all, all the great things here about Twilio, but for people... I don't want...

We only have fifteen minutes, and Twilio is a company that's this wide- Yeah ... wider, you know, in terms of all the great things it does. But for people who want to just, you know, they don't have time right now, but w- where...

com, what would you suggest is, like, the best path for people to kind of digest all of the great things Twilio does? So I'd tell people if they want to engage with the company, you probably go to the website first, right? Twilio has been, since its foundation, a developer-friendly, developer-centric firm, that makes all of our capabilities available, over APIs, and that focus on accessibility shows up through all of our documentation and all of our work.

We've got a pretty extensive blog. We've got a bunch of material that's available. So if you wanna start engaging, I would tell you to, to go there.

The second thing that I would do is say it is incredibly easy and incredibly accessible to just start working with the tools. Um, entry is very easy, it's very accessible to play with, but most importantly, if you have someone or something you want to be able to communicate with through multiple channels or you wanna have a better understanding about-How your audiences are currently engaging with your content, Twilio is the place to be. Love it.

Very cool. Zach, if you don't mind, I wanna segue or pivot into a, a new launch from Twil-Twilio here. It's called A2H- Mm-hmm ...

Agent2Human. Uh, this is an open source protocol. Oh, everybody loves open source, right?

A spec that provides a single channel agnostic audible service surface for agents to communicate with the human principles. All right, I read the, the press release part of it. Talk plain English for us here, Zach, like we're gen AI people.

Um, what is it? Last year has been the big drive where using LLMs, building agents, and leveraging those agents has become more available, not just to those of us kind of in the industry, but to the general public, right? You're seeing that takeoff happen.

One of the things that's happening right now is because agents represent a new interface, a new way of interacting between humans and humans, and humans and their computers, right? We're all starting to explore a little bit more how do we actually drive that interaction? What are the things that are available to us where it wasn't there before?

If you remember back, right, when computers started, like we're going back to the '50s here, right? You interacted with these things with punch cards and actual physical levers and lights that you could see on a machine that was the size of a wall. Over time, we started to get CRT monitors.

First, you know, you effectively had just basic lines, dots, the outlines of a shape, and now, right now, I'm talking to you on a screen that's got, you know, eight thousand pixels or whatever the, the actual value is. But, like, we have rich content and media that we can engage with natively, right? We engage with things with a mouse.

There are keyboards all over the place. We have different ways of engaging. Okay, so with the agentic kind of push that we're starting to see now, everyone's seeing this new technology and asking themselves, "Is there a different way that we can interface with the computer?

Can I talk to it, for example? Is there a way that I can just type plain text and get that information back? Maybe I don't even need to pa- intentionally communicate with it at all.

" Lots of different ways that people are exploring with this. A2H, when you get right down to it, right, is a way that standardizes how an AI agent communicates with and requests permission to do things from humans. And it's that last part that's really important right now.

How do you get consent from humans if the agent wants to do something and it's not sure if it has permission? Historically, inside of, you know, functionally all of, computing, our ability to give the computer a permission has been very coarse. You can read something, you can write to something, or you could execute it, run it.

But as we're getting into a space where agentic judgment is starting to allow a much larger realm of what a machine can do, those previous kind of like three-part models start breaking down. And so A2H gives us-- excuse me, standardizes how computers talk to us when they're asking for more context, when they're asking for consent, and when they're giving us information, right? A2H, we hope, right, is the start of a kind of like interaction mechanism or a contract for this kind of new agentic era we're finding ourselves in, right?

Other protocols think about how agents talk to each other or they, how they talk to data or other technical resources. A2H puts that in the context of can we get a single standardized, ideally channel agnostic surface, could be via voice, it could be via text, doesn't matter what it is, right, for how these agents talk to us. One of the big underpinnings of it, I think I mentioned this before, is that you want that mechanism to drive that positive consent.

"Hi, I need to do something. Am I able to? " You want that to be auditable.

You want it to be secure. You want it to be very clear, so that way when an agent takes an action, the human who is accountable for that action doesn't wake up the next day and realize that they have something that they need to regret. Does that make sense, Alan?

Uh, it makes total sense- Yeah ... to me, but this is my life these days, Zach, so I, you know, they, you talk in my language, no pun intended. " Now, this is an open source spec.

Yes. Yeah? Um, what exactly does that mean?

Like, is it you put it out on GitHub? Mm-hmm. But I assume there's some hooks for Twilio that can be used if you want.

Absolutely. So n-- it's not just available for people to use without Twilio, we hope people use it independent of their use of Twilio, right? We also hope, and this is part of why we have an open spec, we want commentary on it.

What are the things that we may not have anticipated? What are the things that you've seen that cause us to begin to adjust this? This is an interface thing, right?

" And a subjective sense, "Hey, did that feel like a natural interaction? " Right? So we want to get that direct interaction there, right?

So when we think about it, both A2H, right, as well as t- the hooks that it has going back into Twilio, give us the ability to say, if this standard gets adoption, people are starting to look at it, we are adopting it as well, and we're able to drive that forward as a part of what we do. " That's the mechanism that we're going to be able to use for some of our own agentic capabilities, should users, leverage that. Got it.

Now, look, I'm, I'm playing with a lot of agentic stuff myself, my team is here. We've hooked up to some services that let us talk to an agent and it talks back to us- Mm-hmm. Like a, I don't know, voice-to-text kind of stuff, if you will.

But this sounds like something more than that, right? There's-- it's not just reading words or hearing words and converting it to text. It's...

There's, there's, as you called, hooks, right, in, in, into making things happen. Um, when you say it's a specification, what do I gotta do to get this working? Like, okay, this sounds great.

We'd like to use it. Yeah. I go to GitHub, I look up A2H from Twilio, I find it.

What do I need to make it work? Well, we're still in the very early days of the specification, so we've only, we've only just published it. Things that are starting to happen now are the development of libraries that make the adoption of this easier if you're a developer.

Mm-hmm. We're starting to look with and partner with teams who want to implement the specification inside of their own agentic frameworks, right? Okay.

And of course, we're consistently bringing in and, and accepting more commentary on the spec as well. So we're still in those very early stages where there's kind of this forming action around, hey, have we proposed something that's resonating with the larger community as a whole? We think so, right?

Has that proposal demonstrated that it's relatively easy to begin to implement? And for areas where we have easy-to-implement libraries, things that let you get started a little bit faster, a little bit more rapidly, are we watching that adoption and finding that people are able to go places with it and kind of bring their visions to life using that technology? So it's a little bit of all three of those things.

If you ask me again that question in six months, my quest-- my answer's going to be different to you because we'll have gotten through that initial phase of everyone kind of wrapping their heads around and beginning to interact for the first time with a lot of these capabilities. Very cool. Very cool.

Um, people watching out here, we've got developers, we've got geeks, as I call them, they're my people. What do you want to tell them, Zach? How did, how did they get involved?

How did they help? What did they... Well, and help themselves as well, right?

Yeah, I think that's right. So I think the first message that I'd probably want to send is the thi-- the realization that we came to that drove the creation of this specification, and it's this, right: conversation plus consent equals trust. And we came to the realization that with-- e- even-- no matter how powerful the technology is, without a verifiable trail of, you know, human approvals and, and authorizations, right, agent autonomy is gonna hit a ceiling in terms of how it gets adopted, right?

Think about it this way. You wouldn't trust something that isn't you with your credit card if there wasn't a way for that thing to call you up and say, "Hey, I'm about to do something with your money. It's a little ambiguous at the moment.

" You've already authorized them. We're both confident in that, but now you have a clear record of when it was you versus when it was something acting on your behalf and with your permissions. That communication plus consent and the, the resultant trust that you get out of that is, we believe, a big part of what needs to exist to be able to unlock more and more things in the agentic world, not just from the perspective of the early adopters and the believers, but from general people on the street who otherwise would have a, a trust issue because they are not as deep inside of the space as we are.

We've opted in, when you think about it that way, for anyone who comes on your show, for your yourself, for people who look like me, right? And so our premise here, both in A2H and in the larger perspective of kind of where Twilio is, is we want to give our customers two fundamental things. One, a unified platform that they can use to accomplish their goals.

If your goal involves communication, and tell me if there-- I don't think there's any that don't, right? But if your goal involves communication or taking action, right, we wanna give you a simple way of being able to do that. And A2HGives us the ability through a protocol to say, "Hey, if this gets traction, and this is the direction that we all go in, or it evolves into something that becomes that," right, then it just happens naturally.

That fluidity is important. But the second thing that's really important to us here is to make sure that you have that trust. Trust is the thing that lets everything function in a civilized world with a minimum of friction.

If we trust the interactions that we're having, if we know, for example, that, hey, if somebody is engaging with my agent, we can identify both my agent positively as something acting on my behalf, and we can identify the counterparty, the person interacting with it, as a clear, distinct, and known entity, there's trust there. If I'm a customer, and I'm using an agent, and I know and I can track that I'm getting value from that use, there's trust there. All of that is absolutely critical.

And so our goal with A2H is to give us the ability and to just deepen Twilio's commitment to making sure that we have low-friction, easily accessible capabilities in a platform that stretches across the entire life cycle of how an individual might wanna communicate with their audience and better understand that audience's needs, desires, and next steps while making sure that everyone inside of that three-part chain can trust their individual counterparties. We think it's powerful. I think it is too, man.

I think it is too. Very cool. Zach, we're about out of time, um- Fantastic.

You wouldn't happen to have the GitHub address where people can go find this, do you? We absolutely do. Um, it'd be a little weird for me to recite the whole GitHub repo- No, I know- verbally, but we'll strip it down to you ...

com backslash something. Just, you know what, we'll, we'll put it in the notes on, on this, but just make a note, w- if you could have your people send it over to us, we'll, we'll put it in there. You got it.

Absolutely. Sound good? Yes, it does.

All right. Hey, man, I'm sure you're excited. Our people here are so excited.

I got people spending weekends in the office 'cause they're working on their agentic stuff. I'm sure you're, you're seeing and hearing this too. It's an exciting time to be here.

It really is. Um, keep up the great work. Come back and keep us posted, okay?

Thanks, Alan. I appreciate it. Thank you for having me.

It's my pleasure. Zach Hanif, head of AI, ML, and Data over at Twilio here on Techstrong TV. We're gonna take a break.

We'll be back with more.