WEBVTT
00:00:20.800 --> 00:00:27.280
Welcome to AI Speed, the show where AI-powered companies talk about what actually works in the market right now.
00:00:27.519 --> 00:00:30.320
Business doesn't move at internet speed anymore.
00:00:30.480 --> 00:00:37.039
It moves at AI speed, and the people who figure out how to turn models into money will own the next decade.
00:00:37.200 --> 00:00:37.840
I'm Evan J.
00:00:38.000 --> 00:00:42.479
Chalvin, founder of Luxhammer and growth partner to high-performing brands.
00:00:42.719 --> 00:00:47.520
Today, I'm thrilled to be joined by Mangit Ng, founder and CEO of Anvil.
00:00:47.920 --> 00:00:50.079
That's useanvil.com.
00:00:50.320 --> 00:00:58.880
Mangit is building a platform that's tackling a problem nearly every business deals with, but few have truly solved paperwork.
00:00:59.039 --> 00:01:11.359
By turning documents into structured data and programmable workflows, Anvil is helping companies move beyond PDFs and manual processes into a more automated, scalable way of operating.
00:01:11.680 --> 00:01:13.680
Mangit, thank you for being here.
00:01:14.000 --> 00:01:15.120
It's a pleasure to be here.
00:01:15.439 --> 00:01:19.120
So paperwork is one of those universal pain points.
00:01:19.200 --> 00:01:24.400
What was the moment where you realized this was a problem worth building a company around?
00:01:24.640 --> 00:01:30.239
Oh man, yeah, I think it it's certainly a universal dislike for everyone, I would say.
00:01:30.480 --> 00:01:34.159
For me personally, uh there was kind of a turning point.
00:01:34.239 --> 00:01:38.640
I was applying for a mortgage, and these bankers kept sending me PDFs to fill out.
00:01:38.799 --> 00:01:41.200
And all the PDFs had the same information.
00:01:41.359 --> 00:01:46.640
Um, they wanted me to submit over email, and you know, yeah, it's a mortgage application.
00:01:46.719 --> 00:01:54.079
So they're asking you for social security, they're asking you for your income, your bank account numbers, and it just felt really broken and really insecure.
00:01:54.239 --> 00:01:56.879
So that was kind of like my personal experience.
00:01:57.040 --> 00:02:12.560
But as I kind of peeled part this problem and looked deeper into it, the realization came that whatever pain I'm feeling as a kind of end user of paperwork, uh, the person that's providing information, that banker was feeling tenfold.
00:02:12.639 --> 00:02:24.159
They have operations teams, they have operations peoples internally, and their entire job is like send out PDFs, collect the PDFs, read the PDFs manually, and type it into their lending system, their CRM system.
00:02:24.319 --> 00:02:26.560
So, and they do that hundreds of times a day.
00:02:26.719 --> 00:02:32.080
They just don't even have the time to lift their heads up to think about how they can do it better or easier or more efficiently.
00:02:32.240 --> 00:02:44.639
And so that's when I realized there really is a business creating software and creating tools for other software companies to build out end-to-end automated paperwork processes.
00:02:44.960 --> 00:02:46.080
Wow, that's awesome.
00:02:46.319 --> 00:02:56.879
So, what makes Anvil's approach unique, particularly in how you're turning documents and into data and workflows instead of just digitizing PDFs?
00:02:57.120 --> 00:02:57.439
Yeah.
00:02:57.680 --> 00:02:59.439
Um I think a couple things.
00:02:59.599 --> 00:03:10.319
If you look at the landscape today, there are quite a few companies in what we call the OCR space, or more recently, um LLM kind of like data extraction space.
00:03:10.479 --> 00:03:14.479
And their main product is they take a library of existing PDFs.
00:03:14.560 --> 00:03:19.520
They're typically filled out, or they're always filled out, I should say, and they try to make sense of it.
00:03:19.599 --> 00:03:19.840
Right?
00:03:20.000 --> 00:03:31.599
They they read the data, they pull out the information, they turn it into what you would call structured data that can then be immediately used and consumed by other software systems.
00:03:31.840 --> 00:03:35.360
So that's that's like the majority of the landscape today, I would say.
00:03:35.439 --> 00:03:38.960
Uh, what AML does is actually we flip that problem on its head.
00:03:39.039 --> 00:03:41.680
Um, we actually think that's the backwards way to look at it.
00:03:41.759 --> 00:03:48.560
Um, the right way to look at it is to think of documents as a, let's call it, a data transportation layer.
00:03:48.639 --> 00:03:51.439
I'm a person, I need to apply for a bank mortgage.
00:03:51.599 --> 00:03:58.000
So I am putting my data onto a PDF and then transporting it to the banker so that they can then do something with that data, right?
00:03:59.039 --> 00:04:03.840
The fact that it's on a PDF is actually a terrible experience.
00:04:03.919 --> 00:04:07.599
It's a terrible experience for the individual, it's a terrible experience for the computers, right?
00:04:07.680 --> 00:04:14.159
It's it's very hard to take data off of PDFs because PDFs are not meant to be structured information.
00:04:14.240 --> 00:04:16.800
They're meant to be information that's read by humans.
00:04:16.959 --> 00:04:20.959
Um, so what Anvil does is we try to maintain structure.
00:04:21.360 --> 00:04:29.199
When you go to use Anvil, you're not taking a filled-out PDF eating, reading information off of it using character recognition.
00:04:29.360 --> 00:04:31.439
What you're doing is saying, I already have this information.
00:04:31.519 --> 00:04:36.399
It already lives in your CRM or already lives in your Google account or whatever.
00:04:36.560 --> 00:04:47.759
And I'm going to take that data, I'm going to keep it in a structured format and generate a PDF as kind of a byproduct that can be sent out so that it can be received by the receiving party.
00:04:47.920 --> 00:04:51.680
And typically that's because they have existing processes and that's how they work.
00:04:52.079 --> 00:04:59.360
But over time, the real goal here is to actually have that data flow directly into whatever third party you wish to share that information with.
00:04:59.600 --> 00:05:03.680
And the PDFs kind of become an artifact of that process.
00:05:03.920 --> 00:05:25.120
And so companies that use Anvil to build can collect information in a structured format, generate the PDFs that they need to generate and submit it to third parties outside of their organization, but they get to continue working with the structured data and treat the PDFs as a byproduct, as kind of an afterthought, as opposed to having the PDF be the core of their of their business.
00:05:25.519 --> 00:05:42.240
Anvil's really thinking of a data first approach, with PDFs being the output slash artifact, as opposed to saying PDFs are the core in the library and like what we you know where the information all lives and and and using that as the as kind of like the the source of truth.
00:05:42.319 --> 00:05:45.839
So that's that's the difference between Anvil and and OCR.
00:05:46.720 --> 00:05:48.480
Yeah, that makes a lot of sense.
00:05:48.639 --> 00:05:51.759
Um so where are you focusing most of your energy right now?
00:05:52.079 --> 00:05:52.639
Yeah.
00:05:52.959 --> 00:06:02.639
Um well, I think there's a lot of energy being placed on how do we leverage existing AI tools to make our process better.
00:06:02.879 --> 00:06:10.800
Again, because we think of the problem slightly differently, Phil is using AI differently internally in our product, right?
00:06:10.959 --> 00:06:15.360
So we use AI to find blank fields on a PDF.
00:06:15.519 --> 00:06:28.319
And then we use AI to label those fields, and then we use AI to help us provide kind of like a common schema that you, as a developer, as a software developer, or or as an organization that's somewhat technical, can interface with.
00:06:28.399 --> 00:06:35.680
And we can take your data, translate it into the structure of the PDF's data, and then we can put it onto the PDF for you.
00:06:35.920 --> 00:06:47.360
So that's one area that we're we're really investing in is how do we use AI to create this um remove the friction of moving data from one place to another place.
00:06:47.519 --> 00:06:54.720
Other areas that we're really focused on is how do we make Anvil's endpoints, API endpoints more consumable by AI agents.
00:06:54.959 --> 00:07:00.560
So, you know, we launched our MCP server, which allows agents to directly interface with Anvil.
00:07:00.800 --> 00:07:07.680
Now you can go to a Chat GPT or Claude and say something like, Hey, here's a PDF.
00:07:07.759 --> 00:07:09.279
I need you to set it up for signature for me.
00:07:09.360 --> 00:07:13.920
Maybe, maybe you say something like, hey, review that the terms are okay and then set it up for signature for me.
00:07:14.160 --> 00:07:23.120
So Claude will do that, and then it'll make a request out to Anvil and say, Hey, I need you to set up a signature packet with this field for this person.
00:07:23.360 --> 00:07:26.560
Um, and then immediately present that within Cloud for you to sign.
00:07:26.639 --> 00:07:30.560
So, and that's a legally binding, cryptographically secure e-signature.
00:07:30.720 --> 00:07:33.519
So those are kind of the areas that we're really exploring today.
00:07:33.680 --> 00:07:40.480
How people work, how people work with AI and how Anvil works in this new world where AI tooling is kind of everywhere.
00:07:40.720 --> 00:07:44.720
Yeah, those are really interesting uh use cases for AI.
00:07:44.800 --> 00:07:45.439
That's cool.
00:07:45.680 --> 00:07:47.120
So, who do you serve best?
00:07:47.199 --> 00:07:50.720
What's the ideal customer profile that gets the most value from Anvil?
00:07:51.040 --> 00:07:51.279
Yeah.
00:07:51.439 --> 00:07:56.319
Today, our ideal customer profile is typically somebody who's somewhat technical.
00:07:56.480 --> 00:07:58.079
Oftentimes it's engineers.
00:07:58.319 --> 00:08:08.800
Oftentimes, what their company is trying to do is take an existing manual process and create a more automated, online, connected process.
00:08:08.959 --> 00:08:12.319
So you can think of it as a sick insurance, for example.
00:08:12.480 --> 00:08:20.959
If you're an insurance broker, you used to sell insurance through face-to-face meetings and phone calls, and you'd call out to get a quote.
00:08:21.199 --> 00:08:25.360
Now with Anvil, you can actually do most of that online.
00:08:25.439 --> 00:08:34.399
Somebody can come to your website, provide the information, it populates the correct insurance documents to be submitted to the state farms, the all states of the world.
00:08:34.559 --> 00:08:36.240
They do whatever they do and they come back.
00:08:36.320 --> 00:08:38.320
It's like, great, here's the policy.
00:08:38.480 --> 00:08:43.600
And you can use Anvil to automatically, with code, set up a signature packet.
00:08:43.759 --> 00:08:48.639
Now, the nice thing is with all these coding agents these days, more and more people are able to do that.
00:08:48.720 --> 00:08:56.159
You know, you don't actually have to have a deep history of software development to be able to create software that does that automatically for you.
00:08:56.320 --> 00:09:16.480
And that's that's what Anvil is trying to do and broaden our ICP to not just software developers, but also operations people, uh, back office people, um, kind of people who report into the COO organization, those types of people that that are just inundated with with moving paperwork around all day.
00:09:16.720 --> 00:09:17.840
Yeah, that's great.
00:09:18.080 --> 00:09:21.279
So, what's been your biggest challenge growing Anvil this year?
00:09:21.600 --> 00:09:23.200
Biggest challenge growing Anvil this year.
00:09:23.279 --> 00:09:23.440
Wow.
00:09:23.679 --> 00:09:26.159
Just honestly the pace, the pace of everything.
00:09:26.320 --> 00:09:29.360
Every, every other day, there's a new model coming out.
00:09:29.519 --> 00:09:36.240
Every other day, there's a new person that vibe coded an application that was like, hey, I built DocuSign in a weekend.
00:09:36.399 --> 00:09:38.639
And Anvil, Anvil has an e-signature solution as well.
00:09:38.799 --> 00:09:40.320
So we're competing with them.
00:09:40.559 --> 00:09:47.279
And I think there's a lot of excitement, warranted excitement around how easy it is to build software now.
00:09:47.519 --> 00:09:57.360
But I think there's a lot of misplaced, um misplaced belief that just building software is enough to create a business, to create a company.
00:09:57.519 --> 00:10:04.480
Because when you think about a product like eSignatures, what you're selling is actually compliance and trust, not the ability to scribble your name.
00:10:04.639 --> 00:10:14.159
And that compliance and trust comes with layers of, you know, of like auditing, of logging, of setting up your signing certificate infrastructure.
00:10:14.240 --> 00:10:21.200
So there's a lot of work that goes into it that is beyond a kind of like good MVP, shiny demo.
00:10:21.360 --> 00:10:22.399
So I think that's part of it.
00:10:22.639 --> 00:10:30.559
I think the other part of it is just like the transition from a search first world to maybe a AI suggestion world.
00:10:30.720 --> 00:10:32.000
That's that's been interesting.
00:10:32.159 --> 00:10:40.240
We are certainly seeing ourselves rank higher and higher amongst all of the chat GPTs, claws, and perplexities of the world.
00:10:40.320 --> 00:10:41.919
But it is a different mental model.
00:10:42.000 --> 00:10:43.840
So marketing has kind of shifted.
00:10:44.159 --> 00:10:49.120
I think people are figuring out what marketing means and trying to do that in real time as marketing is changing.
00:10:49.200 --> 00:10:51.679
So that's another interesting challenge that we're facing.
00:10:51.919 --> 00:10:53.279
Yeah, absolutely.
00:10:53.600 --> 00:11:02.639
Um, and how do you differentiate really like in a space where there's so many tools that claim to automate workflows?
00:11:02.879 --> 00:11:08.879
Yeah, and then then the ultimate tool, which is like, you know, the claude and Jack GPT is just automating it themselves internally.
00:11:10.000 --> 00:11:10.480
Right.
00:11:10.799 --> 00:11:22.080
I think one thing that we hear a lot about Anvil from our customers and from our prospects is the level of integration that Anvil provides, the flexibility that Anvil provides, right?
00:11:22.320 --> 00:11:32.799
So to design a system that is both really easy to use, but accommodating of different people's ways of working and flexible, that's pretty hard.
00:11:32.879 --> 00:11:34.879
And I think we've done a pretty good job of that.
00:11:35.039 --> 00:11:42.000
So today, you can go use Anvil as a school teacher that needs to collect field trip sign-offs from parents.
00:11:42.159 --> 00:11:52.960
But you could also use Anvil if you are a you know 20-year-old veteran software engineer building customized AI agents that need a document automation tool.
00:11:53.039 --> 00:12:03.440
And Anvil provides you, meets you where you are in terms of the level of complexity and flexibility that you need for for whatever you're trying to accomplish.
00:12:03.919 --> 00:12:06.000
So that's yeah, that's that's what I would say.
00:12:06.320 --> 00:12:14.000
And what creates hesitation for companies when moving away from traditional processes like PDFs and manual workflows?
00:12:15.039 --> 00:12:17.919
Yeah, uh I think it's just comfort a lot of times.
00:12:18.080 --> 00:12:19.120
That's a big part of it.
00:12:19.200 --> 00:12:26.559
I think it's just the the friction, the the inertia of this is how we've always done it.
00:12:26.720 --> 00:12:30.080
Uh I think there's a lot of companies that feel not empowered, right?
00:12:30.240 --> 00:12:35.919
They the the documents, the PDFs that they have to fill out are dictated by a third party.
00:12:36.080 --> 00:12:44.480
Maybe it's the state farm insurance claims claims forms that they have to fill out or the IRS government documents, right?
00:12:44.559 --> 00:12:49.279
You know, as a small business, you can't go to the IRS and say, hey, change your forms so it's easier for me.
00:12:49.519 --> 00:12:50.240
They don't care.
00:12:50.480 --> 00:12:52.399
So I think there's a little bit of that.
00:12:53.279 --> 00:12:58.480
But what we're trying to do is to show them that it actually is possible and very easy to do.
00:12:58.559 --> 00:13:10.480
When you simply upload that PDF into Anvil, we will tag it for you automatically and is ready for you to consume as part of a signature packet, as part of an Anvil workflow, whatever it may be.
00:13:10.720 --> 00:13:12.559
Um you can do it.
00:13:12.799 --> 00:13:14.159
Yeah, that's really cool.
00:13:14.399 --> 00:13:18.559
So, what challenges do you face to raising awareness about your company and its products?
00:13:18.960 --> 00:13:21.039
Anvil is a boring business.
00:13:21.120 --> 00:13:24.639
I think that's probably, you know, try to make Anvil exciting for people.
00:13:24.720 --> 00:13:26.720
Uh, try to make paperwork exciting for people.
00:13:26.879 --> 00:13:31.519
People don't typically sit around thinking, huh, I like thinking about paperwork.
00:13:31.600 --> 00:13:32.639
I like thinking about forms.
00:13:32.720 --> 00:13:33.840
I think it's like taxes, right?
00:13:33.919 --> 00:13:39.600
Like, fortunately for taxes, there's like a deadline that some government sets and they can like penalize you.
00:13:39.679 --> 00:13:43.679
But uh with paperwork more broadly, there there isn't that forcing function.
00:13:43.759 --> 00:13:47.519
Um, so I think that's a little bit of the challenge with bringing awareness.
00:13:47.679 --> 00:14:00.799
I think it's also a very broad problem in that it's hard to know exactly what are the triggers or the things that like allow Anvil to identify that you have this problem and you need to solve it, right?
00:14:00.879 --> 00:14:10.639
Like, so a classic example of this is like if a small business or online business knows you're having a child, they know that these are the things they need to sell to you.
00:14:10.720 --> 00:14:13.120
They know exactly what that's the triggering event.
00:14:13.279 --> 00:14:19.919
With Anvil, oftentimes these are just like long-learning processes, and you don't know who is even sitting there thinking about how I can make this better.
00:14:20.159 --> 00:14:22.080
Those are internal discussions to businesses.
00:14:22.240 --> 00:14:24.399
It's not advertised or broadcast out.
00:14:24.559 --> 00:14:36.960
So for us, what we really need to do is just be in be top of mind all the time for a very boring, for a very boring kind of like uh area of work.
00:14:37.279 --> 00:14:40.960
And so figuring out ways to do that is has been the big challenge.
00:14:41.200 --> 00:14:42.639
Yeah, how do you address that?
00:14:42.879 --> 00:14:45.600
Yeah, I think the first way we address that is to be helpful.
00:14:45.759 --> 00:14:55.039
Um, our content that we put out there is really geared towards being helpful, solving the problem that the people is that the person is searching for uh a solution to.
00:14:55.279 --> 00:14:59.279
We also make Anvil usage from our UI free.
00:14:59.440 --> 00:15:09.759
So if you want to send a signature packet, legally binding signature packet, you can from our user interface, you can log into Anvil, upload a PDF, 10 PDFs, however many PDFs you want.
00:15:09.840 --> 00:15:25.120
Um, you can collect information on that PDF, you can get it signed, and we don't charge you a dime because we think that the use case there is really for these one-offs, which is great marketing for us, but it's also like for maybe the, like I said, the school teacher that just needs to collect field trip waivers, right?
00:15:25.200 --> 00:15:27.200
Like we don't want to be charging them.
00:15:28.159 --> 00:15:30.799
Where we do make money is on the integrated usage, right?
00:15:30.879 --> 00:15:35.440
You have a thousand clients, and every year you need to collect a letter of engagement with them.
00:15:35.600 --> 00:15:36.159
Like, great.
00:15:36.320 --> 00:15:47.759
Use Anvil in an automated, integrated fashion, whether that's with your own code, whether that's with Claude or with Zapier or other N8 and no code tools, that's where we do charge.
00:15:48.159 --> 00:15:48.799
That's where we make money.
00:15:48.879 --> 00:15:51.759
So it's really bringing making your life easier.
00:15:51.840 --> 00:15:53.120
That's when that's when we really charge.
00:15:53.360 --> 00:15:54.720
Yeah, that makes a lot of sense.
00:15:54.960 --> 00:16:00.960
So, what trends are you seeing in automation, APIs, and AI that excite you the most?
00:16:01.519 --> 00:16:03.120
That excite me the most.
00:16:03.919 --> 00:16:08.720
I I think the capabilities of these models is just insane.
00:16:08.960 --> 00:16:14.080
You know, I think two years ago was like a fun toy.
00:16:14.320 --> 00:16:17.519
A year ago it was like, okay, I could see it kind of do stuff here.
00:16:17.679 --> 00:16:20.639
Um, but it's kind of like broken.
00:16:20.799 --> 00:16:29.919
Like I remember trying to have use Chat GPT a year ago to say, hey, I need you to go through the attendees of this conference and help me create a list.
00:16:30.080 --> 00:16:32.000
And it would poop out after like 20.
00:16:32.159 --> 00:16:33.519
And it would just be like, ah, I'm done.
00:16:33.600 --> 00:16:43.200
It was like a bad intern, you know, like um, and you would have to like kind of like berate it and be like, look, no, no, no, I need you to keep going until you've gone through all 3,000.
00:16:43.360 --> 00:16:45.519
If it was 20 people, I would have just done it myself.
00:16:45.840 --> 00:16:53.440
But now it's getting to a point where it is quite good, quite powerful.
00:16:53.600 --> 00:17:00.879
And on top of that, it has the focus, the the ability to focus on a single task for long enough to actually solve it.
00:17:01.039 --> 00:17:02.720
I think that's been that's been interesting.
00:17:02.879 --> 00:17:11.519
Now I can have clawed code go out and think about a problem for 15, 20 minutes, 30 minutes a day, and I'll come back with a solution.
00:17:11.680 --> 00:17:14.720
In the past, the the con the context windows were just too small.
00:17:14.799 --> 00:17:15.279
You couldn't do that.
00:17:15.440 --> 00:17:18.000
Now now it's actually pretty good at it.
00:17:18.400 --> 00:17:20.559
Yeah, that's really cool.
00:17:20.880 --> 00:17:29.920
Uh so if we were to have this conversation again in 12 months, what would need to happen for you to feel like it was a big win for the company?
00:17:30.640 --> 00:17:33.599
What would need to happen for me to feel like it's a big win?
00:17:33.759 --> 00:17:47.920
I think something that I'm really focused on right now is that when a developer happens upon Anvil and they're doing research on software that can help them automate or build software faster for documents, they tend to really, really prefer Anvil.
00:17:48.240 --> 00:17:49.599
We hear this all the time.
00:17:49.839 --> 00:17:54.160
You know, the the APIs of the other providers are so difficult to work with.
00:17:54.319 --> 00:18:02.000
Um there's they're very opinionated about how things are done, and I don't get the flexibility that I want to build the thing that I want.
00:18:02.160 --> 00:18:05.200
So we we're often the first choice of developers.
00:18:05.359 --> 00:18:14.640
Now, there's all these other decision makers in the process that may or may not derail a sales motion, but from a champion perspective, developers are our champion.
00:18:15.279 --> 00:18:18.240
I would like AI agents to be our champion as well.
00:18:18.400 --> 00:18:30.640
I would like an AI agent when asked, hey, I need you to build, I get order forms every day, and I need you to build a system that helps me review the order forms and then add a signature to it if it works.
00:18:30.799 --> 00:18:33.759
Like, right, like if the order form is good.
00:18:33.920 --> 00:18:44.640
I want agents who are just going out and figuring out how to do this as they're searching the web to find Anvil and prefer Anvil as that solution for them, right?
00:18:44.720 --> 00:18:46.720
I don't want them building it from scratch themselves.
00:18:46.880 --> 00:18:52.720
I don't want them to go spin up a whole nother service that is unrelated to the core task that's being accomplished.
00:18:52.880 --> 00:18:55.200
So I want them to use Anvil to do it.
00:18:55.359 --> 00:18:56.960
So I think that would be a huge win.
00:18:57.119 --> 00:19:06.319
I think maybe another message that I would like to get across for maybe all the developers out there is um the world runs on paperwork.
00:19:06.480 --> 00:19:15.920
As much as we live in a bubble that you're doing, uh we're working on AI and crypto, whatever it is you're working that's like fully digital, the world still runs on paperwork.
00:19:16.000 --> 00:19:23.599
And if you want your AI agent to be useful to do something for you, at some point they're going to have to interface with paperwork.
00:19:23.680 --> 00:19:27.200
And uh, you know, let Anvil help them do real work.
00:19:27.519 --> 00:19:32.799
So that that that would be the pitch for for humans out there that are building AI agents.
00:19:33.119 --> 00:19:35.920
Well, that's it for today's episode of AI Speed.
00:19:36.079 --> 00:19:50.000
A huge thank you to Mangit for sharing his invaluable insights into how Anvil is helping companies turn documents into data and workflows and for navigating the future of automation and APIs.
00:19:50.240 --> 00:20:01.279
If you're building or leading an AI native company or a service business that uses AI under the hood and you care about revenue adoption and market share, make sure to subscribe to AISpeed.
00:20:01.440 --> 00:20:06.160
Learn how the best AI operators ship faster, sell smarter, and stay ahead.
00:20:06.319 --> 00:20:07.279
Thanks for listening.
00:20:07.440 --> 00:20:12.559
Until next time, keep building, keep selling, and keep moving at AI speed.
00:00:20.800 --> 00:00:27.280
Welcome to AI Speed, the show where AI-powered companies talk about what actually works in the market right now.
00:00:27.519 --> 00:00:30.320
Business doesn't move at internet speed anymore.
00:00:30.480 --> 00:00:37.039
It moves at AI speed, and the people who figure out how to turn models into money will own the next decade.
00:00:37.200 --> 00:00:37.840
I'm Evan J.
00:00:38.000 --> 00:00:42.479
Chalvin, founder of Luxhammer and growth partner to high-performing brands.
00:00:42.719 --> 00:00:47.520
Today, I'm thrilled to be joined by Mangit Ng, founder and CEO of Anvil.
00:00:47.920 --> 00:00:50.079
That's useanvil.com.
00:00:50.320 --> 00:00:58.880
Mangit is building a platform that's tackling a problem nearly every business deals with, but few have truly solved paperwork.
00:00:59.039 --> 00:01:11.359
By turning documents into structured data and programmable workflows, Anvil is helping companies move beyond PDFs and manual processes into a more automated, scalable way of operating.
00:01:11.680 --> 00:01:13.680
Mangit, thank you for being here.
00:01:14.000 --> 00:01:15.120
It's a pleasure to be here.
00:01:15.439 --> 00:01:19.120
So paperwork is one of those universal pain points.
00:01:19.200 --> 00:01:24.400
What was the moment where you realized this was a problem worth building a company around?
00:01:24.640 --> 00:01:30.239
Oh man, yeah, I think it it's certainly a universal dislike for everyone, I would say.
00:01:30.480 --> 00:01:34.159
For me personally, uh there was kind of a turning point.
00:01:34.239 --> 00:01:38.640
I was applying for a mortgage, and these bankers kept sending me PDFs to fill out.
00:01:38.799 --> 00:01:41.200
And all the PDFs had the same information.
00:01:41.359 --> 00:01:46.640
Um, they wanted me to submit over email, and you know, yeah, it's a mortgage application.
00:01:46.719 --> 00:01:54.079
So they're asking you for social security, they're asking you for your income, your bank account numbers, and it just felt really broken and really insecure.
00:01:54.239 --> 00:01:56.879
So that was kind of like my personal experience.
00:01:57.040 --> 00:02:12.560
But as I kind of peeled part this problem and looked deeper into it, the realization came that whatever pain I'm feeling as a kind of end user of paperwork, uh, the person that's providing information, that banker was feeling tenfold.
00:02:12.639 --> 00:02:24.159
They have operations teams, they have operations peoples internally, and their entire job is like send out PDFs, collect the PDFs, read the PDFs manually, and type it into their lending system, their CRM system.
00:02:24.319 --> 00:02:26.560
So, and they do that hundreds of times a day.
00:02:26.719 --> 00:02:32.080
They just don't even have the time to lift their heads up to think about how they can do it better or easier or more efficiently.
00:02:32.240 --> 00:02:44.639
And so that's when I realized there really is a business creating software and creating tools for other software companies to build out end-to-end automated paperwork processes.
00:02:44.960 --> 00:02:46.080
Wow, that's awesome.
00:02:46.319 --> 00:02:56.879
So, what makes Anvil's approach unique, particularly in how you're turning documents and into data and workflows instead of just digitizing PDFs?
00:02:57.120 --> 00:02:57.439
Yeah.
00:02:57.680 --> 00:02:59.439
Um I think a couple things.
00:02:59.599 --> 00:03:10.319
If you look at the landscape today, there are quite a few companies in what we call the OCR space, or more recently, um LLM kind of like data extraction space.
00:03:10.479 --> 00:03:14.479
And their main product is they take a library of existing PDFs.
00:03:14.560 --> 00:03:19.520
They're typically filled out, or they're always filled out, I should say, and they try to make sense of it.
00:03:19.599 --> 00:03:19.840
Right?
00:03:20.000 --> 00:03:31.599
They they read the data, they pull out the information, they turn it into what you would call structured data that can then be immediately used and consumed by other software systems.
00:03:31.840 --> 00:03:35.360
So that's that's like the majority of the landscape today, I would say.
00:03:35.439 --> 00:03:38.960
Uh, what AML does is actually we flip that problem on its head.
00:03:39.039 --> 00:03:41.680
Um, we actually think that's the backwards way to look at it.
00:03:41.759 --> 00:03:48.560
Um, the right way to look at it is to think of documents as a, let's call it, a data transportation layer.
00:03:48.639 --> 00:03:51.439
I'm a person, I need to apply for a bank mortgage.
00:03:51.599 --> 00:03:58.000
So I am putting my data onto a PDF and then transporting it to the banker so that they can then do something with that data, right?
00:03:59.039 --> 00:04:03.840
The fact that it's on a PDF is actually a terrible experience.
00:04:03.919 --> 00:04:07.599
It's a terrible experience for the individual, it's a terrible experience for the computers, right?
00:04:07.680 --> 00:04:14.159
It's it's very hard to take data off of PDFs because PDFs are not meant to be structured information.
00:04:14.240 --> 00:04:16.800
They're meant to be information that's read by humans.
00:04:16.959 --> 00:04:20.959
Um, so what Anvil does is we try to maintain structure.
00:04:21.360 --> 00:04:29.199
When you go to use Anvil, you're not taking a filled-out PDF eating, reading information off of it using character recognition.
00:04:29.360 --> 00:04:31.439
What you're doing is saying, I already have this information.
00:04:31.519 --> 00:04:36.399
It already lives in your CRM or already lives in your Google account or whatever.
00:04:36.560 --> 00:04:47.759
And I'm going to take that data, I'm going to keep it in a structured format and generate a PDF as kind of a byproduct that can be sent out so that it can be received by the receiving party.
00:04:47.920 --> 00:04:51.680
And typically that's because they have existing processes and that's how they work.
00:04:52.079 --> 00:04:59.360
But over time, the real goal here is to actually have that data flow directly into whatever third party you wish to share that information with.
00:04:59.600 --> 00:05:03.680
And the PDFs kind of become an artifact of that process.
00:05:03.920 --> 00:05:25.120
And so companies that use Anvil to build can collect information in a structured format, generate the PDFs that they need to generate and submit it to third parties outside of their organization, but they get to continue working with the structured data and treat the PDFs as a byproduct, as kind of an afterthought, as opposed to having the PDF be the core of their of their business.
00:05:25.519 --> 00:05:42.240
Anvil's really thinking of a data first approach, with PDFs being the output slash artifact, as opposed to saying PDFs are the core in the library and like what we you know where the information all lives and and and using that as the as kind of like the the source of truth.
00:05:42.319 --> 00:05:45.839
So that's that's the difference between Anvil and and OCR.
00:05:46.720 --> 00:05:48.480
Yeah, that makes a lot of sense.
00:05:48.639 --> 00:05:51.759
Um so where are you focusing most of your energy right now?
00:05:52.079 --> 00:05:52.639
Yeah.
00:05:52.959 --> 00:06:02.639
Um well, I think there's a lot of energy being placed on how do we leverage existing AI tools to make our process better.
00:06:02.879 --> 00:06:10.800
Again, because we think of the problem slightly differently, Phil is using AI differently internally in our product, right?
00:06:10.959 --> 00:06:15.360
So we use AI to find blank fields on a PDF.
00:06:15.519 --> 00:06:28.319
And then we use AI to label those fields, and then we use AI to help us provide kind of like a common schema that you, as a developer, as a software developer, or or as an organization that's somewhat technical, can interface with.
00:06:28.399 --> 00:06:35.680
And we can take your data, translate it into the structure of the PDF's data, and then we can put it onto the PDF for you.
00:06:35.920 --> 00:06:47.360
So that's one area that we're we're really investing in is how do we use AI to create this um remove the friction of moving data from one place to another place.
00:06:47.519 --> 00:06:54.720
Other areas that we're really focused on is how do we make Anvil's endpoints, API endpoints more consumable by AI agents.
00:06:54.959 --> 00:07:00.560
So, you know, we launched our MCP server, which allows agents to directly interface with Anvil.
00:07:00.800 --> 00:07:07.680
Now you can go to a Chat GPT or Claude and say something like, Hey, here's a PDF.
00:07:07.759 --> 00:07:09.279
I need you to set it up for signature for me.
00:07:09.360 --> 00:07:13.920
Maybe, maybe you say something like, hey, review that the terms are okay and then set it up for signature for me.
00:07:14.160 --> 00:07:23.120
So Claude will do that, and then it'll make a request out to Anvil and say, Hey, I need you to set up a signature packet with this field for this person.
00:07:23.360 --> 00:07:26.560
Um, and then immediately present that within Cloud for you to sign.
00:07:26.639 --> 00:07:30.560
So, and that's a legally binding, cryptographically secure e-signature.
00:07:30.720 --> 00:07:33.519
So those are kind of the areas that we're really exploring today.
00:07:33.680 --> 00:07:40.480
How people work, how people work with AI and how Anvil works in this new world where AI tooling is kind of everywhere.
00:07:40.720 --> 00:07:44.720
Yeah, those are really interesting uh use cases for AI.
00:07:44.800 --> 00:07:45.439
That's cool.
00:07:45.680 --> 00:07:47.120
So, who do you serve best?
00:07:47.199 --> 00:07:50.720
What's the ideal customer profile that gets the most value from Anvil?
00:07:51.040 --> 00:07:51.279
Yeah.
00:07:51.439 --> 00:07:56.319
Today, our ideal customer profile is typically somebody who's somewhat technical.
00:07:56.480 --> 00:07:58.079
Oftentimes it's engineers.
00:07:58.319 --> 00:08:08.800
Oftentimes, what their company is trying to do is take an existing manual process and create a more automated, online, connected process.
00:08:08.959 --> 00:08:12.319
So you can think of it as a sick insurance, for example.
00:08:12.480 --> 00:08:20.959
If you're an insurance broker, you used to sell insurance through face-to-face meetings and phone calls, and you'd call out to get a quote.
00:08:21.199 --> 00:08:25.360
Now with Anvil, you can actually do most of that online.
00:08:25.439 --> 00:08:34.399
Somebody can come to your website, provide the information, it populates the correct insurance documents to be submitted to the state farms, the all states of the world.
00:08:34.559 --> 00:08:36.240
They do whatever they do and they come back.
00:08:36.320 --> 00:08:38.320
It's like, great, here's the policy.
00:08:38.480 --> 00:08:43.600
And you can use Anvil to automatically, with code, set up a signature packet.
00:08:43.759 --> 00:08:48.639
Now, the nice thing is with all these coding agents these days, more and more people are able to do that.
00:08:48.720 --> 00:08:56.159
You know, you don't actually have to have a deep history of software development to be able to create software that does that automatically for you.
00:08:56.320 --> 00:09:16.480
And that's that's what Anvil is trying to do and broaden our ICP to not just software developers, but also operations people, uh, back office people, um, kind of people who report into the COO organization, those types of people that that are just inundated with with moving paperwork around all day.
00:09:16.720 --> 00:09:17.840
Yeah, that's great.
00:09:18.080 --> 00:09:21.279
So, what's been your biggest challenge growing Anvil this year?
00:09:21.600 --> 00:09:23.200
Biggest challenge growing Anvil this year.
00:09:23.279 --> 00:09:23.440
Wow.
00:09:23.679 --> 00:09:26.159
Just honestly the pace, the pace of everything.
00:09:26.320 --> 00:09:29.360
Every, every other day, there's a new model coming out.
00:09:29.519 --> 00:09:36.240
Every other day, there's a new person that vibe coded an application that was like, hey, I built DocuSign in a weekend.
00:09:36.399 --> 00:09:38.639
And Anvil, Anvil has an e-signature solution as well.
00:09:38.799 --> 00:09:40.320
So we're competing with them.
00:09:40.559 --> 00:09:47.279
And I think there's a lot of excitement, warranted excitement around how easy it is to build software now.
00:09:47.519 --> 00:09:57.360
But I think there's a lot of misplaced, um misplaced belief that just building software is enough to create a business, to create a company.
00:09:57.519 --> 00:10:04.480
Because when you think about a product like eSignatures, what you're selling is actually compliance and trust, not the ability to scribble your name.
00:10:04.639 --> 00:10:14.159
And that compliance and trust comes with layers of, you know, of like auditing, of logging, of setting up your signing certificate infrastructure.
00:10:14.240 --> 00:10:21.200
So there's a lot of work that goes into it that is beyond a kind of like good MVP, shiny demo.
00:10:21.360 --> 00:10:22.399
So I think that's part of it.
00:10:22.639 --> 00:10:30.559
I think the other part of it is just like the transition from a search first world to maybe a AI suggestion world.
00:10:30.720 --> 00:10:32.000
That's that's been interesting.
00:10:32.159 --> 00:10:40.240
We are certainly seeing ourselves rank higher and higher amongst all of the chat GPTs, claws, and perplexities of the world.
00:10:40.320 --> 00:10:41.919
But it is a different mental model.
00:10:42.000 --> 00:10:43.840
So marketing has kind of shifted.
00:10:44.159 --> 00:10:49.120
I think people are figuring out what marketing means and trying to do that in real time as marketing is changing.
00:10:49.200 --> 00:10:51.679
So that's another interesting challenge that we're facing.
00:10:51.919 --> 00:10:53.279
Yeah, absolutely.
00:10:53.600 --> 00:11:02.639
Um, and how do you differentiate really like in a space where there's so many tools that claim to automate workflows?
00:11:02.879 --> 00:11:08.879
Yeah, and then then the ultimate tool, which is like, you know, the claude and Jack GPT is just automating it themselves internally.
00:11:10.000 --> 00:11:10.480
Right.
00:11:10.799 --> 00:11:22.080
I think one thing that we hear a lot about Anvil from our customers and from our prospects is the level of integration that Anvil provides, the flexibility that Anvil provides, right?
00:11:22.320 --> 00:11:32.799
So to design a system that is both really easy to use, but accommodating of different people's ways of working and flexible, that's pretty hard.
00:11:32.879 --> 00:11:34.879
And I think we've done a pretty good job of that.
00:11:35.039 --> 00:11:42.000
So today, you can go use Anvil as a school teacher that needs to collect field trip sign-offs from parents.
00:11:42.159 --> 00:11:52.960
But you could also use Anvil if you are a you know 20-year-old veteran software engineer building customized AI agents that need a document automation tool.
00:11:53.039 --> 00:12:03.440
And Anvil provides you, meets you where you are in terms of the level of complexity and flexibility that you need for for whatever you're trying to accomplish.
00:12:03.919 --> 00:12:06.000
So that's yeah, that's that's what I would say.
00:12:06.320 --> 00:12:14.000
And what creates hesitation for companies when moving away from traditional processes like PDFs and manual workflows?
00:12:15.039 --> 00:12:17.919
Yeah, uh I think it's just comfort a lot of times.
00:12:18.080 --> 00:12:19.120
That's a big part of it.
00:12:19.200 --> 00:12:26.559
I think it's just the the friction, the the inertia of this is how we've always done it.
00:12:26.720 --> 00:12:30.080
Uh I think there's a lot of companies that feel not empowered, right?
00:12:30.240 --> 00:12:35.919
They the the documents, the PDFs that they have to fill out are dictated by a third party.
00:12:36.080 --> 00:12:44.480
Maybe it's the state farm insurance claims claims forms that they have to fill out or the IRS government documents, right?
00:12:44.559 --> 00:12:49.279
You know, as a small business, you can't go to the IRS and say, hey, change your forms so it's easier for me.
00:12:49.519 --> 00:12:50.240
They don't care.
00:12:50.480 --> 00:12:52.399
So I think there's a little bit of that.
00:12:53.279 --> 00:12:58.480
But what we're trying to do is to show them that it actually is possible and very easy to do.
00:12:58.559 --> 00:13:10.480
When you simply upload that PDF into Anvil, we will tag it for you automatically and is ready for you to consume as part of a signature packet, as part of an Anvil workflow, whatever it may be.
00:13:10.720 --> 00:13:12.559
Um you can do it.
00:13:12.799 --> 00:13:14.159
Yeah, that's really cool.
00:13:14.399 --> 00:13:18.559
So, what challenges do you face to raising awareness about your company and its products?
00:13:18.960 --> 00:13:21.039
Anvil is a boring business.
00:13:21.120 --> 00:13:24.639
I think that's probably, you know, try to make Anvil exciting for people.
00:13:24.720 --> 00:13:26.720
Uh, try to make paperwork exciting for people.
00:13:26.879 --> 00:13:31.519
People don't typically sit around thinking, huh, I like thinking about paperwork.
00:13:31.600 --> 00:13:32.639
I like thinking about forms.
00:13:32.720 --> 00:13:33.840
I think it's like taxes, right?
00:13:33.919 --> 00:13:39.600
Like, fortunately for taxes, there's like a deadline that some government sets and they can like penalize you.
00:13:39.679 --> 00:13:43.679
But uh with paperwork more broadly, there there isn't that forcing function.
00:13:43.759 --> 00:13:47.519
Um, so I think that's a little bit of the challenge with bringing awareness.
00:13:47.679 --> 00:14:00.799
I think it's also a very broad problem in that it's hard to know exactly what are the triggers or the things that like allow Anvil to identify that you have this problem and you need to solve it, right?
00:14:00.879 --> 00:14:10.639
Like, so a classic example of this is like if a small business or online business knows you're having a child, they know that these are the things they need to sell to you.
00:14:10.720 --> 00:14:13.120
They know exactly what that's the triggering event.
00:14:13.279 --> 00:14:19.919
With Anvil, oftentimes these are just like long-learning processes, and you don't know who is even sitting there thinking about how I can make this better.
00:14:20.159 --> 00:14:22.080
Those are internal discussions to businesses.
00:14:22.240 --> 00:14:24.399
It's not advertised or broadcast out.
00:14:24.559 --> 00:14:36.960
So for us, what we really need to do is just be in be top of mind all the time for a very boring, for a very boring kind of like uh area of work.
00:14:37.279 --> 00:14:40.960
And so figuring out ways to do that is has been the big challenge.
00:14:41.200 --> 00:14:42.639
Yeah, how do you address that?
00:14:42.879 --> 00:14:45.600
Yeah, I think the first way we address that is to be helpful.
00:14:45.759 --> 00:14:55.039
Um, our content that we put out there is really geared towards being helpful, solving the problem that the people is that the person is searching for uh a solution to.
00:14:55.279 --> 00:14:59.279
We also make Anvil usage from our UI free.
00:14:59.440 --> 00:15:09.759
So if you want to send a signature packet, legally binding signature packet, you can from our user interface, you can log into Anvil, upload a PDF, 10 PDFs, however many PDFs you want.
00:15:09.840 --> 00:15:25.120
Um, you can collect information on that PDF, you can get it signed, and we don't charge you a dime because we think that the use case there is really for these one-offs, which is great marketing for us, but it's also like for maybe the, like I said, the school teacher that just needs to collect field trip waivers, right?
00:15:25.200 --> 00:15:27.200
Like we don't want to be charging them.
00:15:28.159 --> 00:15:30.799
Where we do make money is on the integrated usage, right?
00:15:30.879 --> 00:15:35.440
You have a thousand clients, and every year you need to collect a letter of engagement with them.
00:15:35.600 --> 00:15:36.159
Like, great.
00:15:36.320 --> 00:15:47.759
Use Anvil in an automated, integrated fashion, whether that's with your own code, whether that's with Claude or with Zapier or other N8 and no code tools, that's where we do charge.
00:15:48.159 --> 00:15:48.799
That's where we make money.
00:15:48.879 --> 00:15:51.759
So it's really bringing making your life easier.
00:15:51.840 --> 00:15:53.120
That's when that's when we really charge.
00:15:53.360 --> 00:15:54.720
Yeah, that makes a lot of sense.
00:15:54.960 --> 00:16:00.960
So, what trends are you seeing in automation, APIs, and AI that excite you the most?
00:16:01.519 --> 00:16:03.120
That excite me the most.
00:16:03.919 --> 00:16:08.720
I I think the capabilities of these models is just insane.
00:16:08.960 --> 00:16:14.080
You know, I think two years ago was like a fun toy.
00:16:14.320 --> 00:16:17.519
A year ago it was like, okay, I could see it kind of do stuff here.
00:16:17.679 --> 00:16:20.639
Um, but it's kind of like broken.
00:16:20.799 --> 00:16:29.919
Like I remember trying to have use Chat GPT a year ago to say, hey, I need you to go through the attendees of this conference and help me create a list.
00:16:30.080 --> 00:16:32.000
And it would poop out after like 20.
00:16:32.159 --> 00:16:33.519
And it would just be like, ah, I'm done.
00:16:33.600 --> 00:16:43.200
It was like a bad intern, you know, like um, and you would have to like kind of like berate it and be like, look, no, no, no, I need you to keep going until you've gone through all 3,000.
00:16:43.360 --> 00:16:45.519
If it was 20 people, I would have just done it myself.
00:16:45.840 --> 00:16:53.440
But now it's getting to a point where it is quite good, quite powerful.
00:16:53.600 --> 00:17:00.879
And on top of that, it has the focus, the the ability to focus on a single task for long enough to actually solve it.
00:17:01.039 --> 00:17:02.720
I think that's been that's been interesting.
00:17:02.879 --> 00:17:11.519
Now I can have clawed code go out and think about a problem for 15, 20 minutes, 30 minutes a day, and I'll come back with a solution.
00:17:11.680 --> 00:17:14.720
In the past, the the con the context windows were just too small.
00:17:14.799 --> 00:17:15.279
You couldn't do that.
00:17:15.440 --> 00:17:18.000
Now now it's actually pretty good at it.
00:17:18.400 --> 00:17:20.559
Yeah, that's really cool.
00:17:20.880 --> 00:17:29.920
Uh so if we were to have this conversation again in 12 months, what would need to happen for you to feel like it was a big win for the company?
00:17:30.640 --> 00:17:33.599
What would need to happen for me to feel like it's a big win?
00:17:33.759 --> 00:17:47.920
I think something that I'm really focused on right now is that when a developer happens upon Anvil and they're doing research on software that can help them automate or build software faster for documents, they tend to really, really prefer Anvil.
00:17:48.240 --> 00:17:49.599
We hear this all the time.
00:17:49.839 --> 00:17:54.160
You know, the the APIs of the other providers are so difficult to work with.
00:17:54.319 --> 00:18:02.000
Um there's they're very opinionated about how things are done, and I don't get the flexibility that I want to build the thing that I want.
00:18:02.160 --> 00:18:05.200
So we we're often the first choice of developers.
00:18:05.359 --> 00:18:14.640
Now, there's all these other decision makers in the process that may or may not derail a sales motion, but from a champion perspective, developers are our champion.
00:18:15.279 --> 00:18:18.240
I would like AI agents to be our champion as well.
00:18:18.400 --> 00:18:30.640
I would like an AI agent when asked, hey, I need you to build, I get order forms every day, and I need you to build a system that helps me review the order forms and then add a signature to it if it works.
00:18:30.799 --> 00:18:33.759
Like, right, like if the order form is good.
00:18:33.920 --> 00:18:44.640
I want agents who are just going out and figuring out how to do this as they're searching the web to find Anvil and prefer Anvil as that solution for them, right?
00:18:44.720 --> 00:18:46.720
I don't want them building it from scratch themselves.
00:18:46.880 --> 00:18:52.720
I don't want them to go spin up a whole nother service that is unrelated to the core task that's being accomplished.
00:18:52.880 --> 00:18:55.200
So I want them to use Anvil to do it.
00:18:55.359 --> 00:18:56.960
So I think that would be a huge win.
00:18:57.119 --> 00:19:06.319
I think maybe another message that I would like to get across for maybe all the developers out there is um the world runs on paperwork.
00:19:06.480 --> 00:19:15.920
As much as we live in a bubble that you're doing, uh we're working on AI and crypto, whatever it is you're working that's like fully digital, the world still runs on paperwork.
00:19:16.000 --> 00:19:23.599
And if you want your AI agent to be useful to do something for you, at some point they're going to have to interface with paperwork.
00:19:23.680 --> 00:19:27.200
And uh, you know, let Anvil help them do real work.
00:19:27.519 --> 00:19:32.799
So that that that would be the pitch for for humans out there that are building AI agents.
00:19:33.119 --> 00:19:35.920
Well, that's it for today's episode of AI Speed.
00:19:36.079 --> 00:19:50.000
A huge thank you to Mangit for sharing his invaluable insights into how Anvil is helping companies turn documents into data and workflows and for navigating the future of automation and APIs.
00:19:50.240 --> 00:20:01.279
If you're building or leading an AI native company or a service business that uses AI under the hood and you care about revenue adoption and market share, make sure to subscribe to AISpeed.
00:20:01.440 --> 00:20:06.160
Learn how the best AI operators ship faster, sell smarter, and stay ahead.
00:20:06.319 --> 00:20:07.279
Thanks for listening.
00:20:07.440 --> 00:20:12.559
Until next time, keep building, keep selling, and keep moving at AI speed.
