Most AI agents are quietly bankrupt — Ilan Zerbib explains why giving them real spending power, not just API keys, is the missing piece of the AI economy.

Ilan Zerbib, founder and CEO of Sapiom, joins Greg Shove to explain why AI agents need their own payment infrastructure — and why most agentic products today are quietly bankrupt.

Ilan built his first company, Earny, running billions of automated bots to recover consumer refunds — a preview of the infrastructure problems agents face today. After Earny's acquisition and five years leading payments engineering at Shopify (Shop Pay, $100B+ in GPV), he founded Sapiom to solve a core problem: the internet's payment and API infrastructure was built to block automation, not enable it.

In this conversation, Ilan and Greg cover:

• Why agents need spending power, not just API keys, to complete real-world tasks
• The difference between the consumer agent economy and the much larger machine-to-machine economy (agents paying agents for APIs and services)
• How Sapiom's governance layer lets businesses give agents autonomy within controlled spending limits
• Why most AI agents today are "bankrupt" — and how cost, model routing, and failure rates determine margin
• Building as a solo founder: raising $51M across three rounds in 12 months, and running a 30-person team without traditional engineering hierarchy
• Why full company automation (80-90%) doesn't mean smaller teams — it means far more ambitious scope
• What it takes to build category-defining, super companies in the AI era

If you're building AI agents, thinking about agentic infrastructure, or trying to figure out how machines will transact in the emerging AI economy, this episode is a blueprint from someone building the rails.

I believe that the company will eventually be eighty to

2

ninety percent fully automated.

3

And then it means we can move at, a

speed much faster than anyone else

4

Greg: You believe at some point in the

future, 80 to 90% of the company, all

5

functions, not just software engineering,

will be automated in some way?

6

Ilan: in a way that

it's kind of AI native?

7

So there's high touch moment

with candidates and with the

8

process, but actually most of the

process could be leveraged by AI.

9

greg-cc272c92-ade1-4182-8e40-1-CFR:

Welcome to Super Companies, my quest

10

to understand how the smartest and

most viable companies in the world

11

are gonna get built in the age of AI.

12

I'm your host, Greg Schove,

seven-time founder, CEO of Section

13

AI, and a super company wannabe.

14

Before we get to our interview, here's

what's on my mind: why small companies

15

have such an advantage in this moment

to build their own super companies.

16

Adoption of AI internally

favors small organizations.

17

Change management is easier.

18

Leadership is closer to the business,

and organizations are usually a

19

little more risk on than risk off.

20

Plus, with a small number of employees,

the costs for inference for AI don't

21

balloon dramatically overnight, so

the CFO doesn't get freaked out and

22

start saying, "Shut it all down."

23

And employees are typically

working really hard.

24

They're small companies, and

th- their to-do lists are long.

25

They're often doing two or three jobs.

26

So they don't, they don't have these

fluff roles that you sometimes see

27

in larger organizations where that

person is worried about getting

28

laid off if AI does their job.

29

In small companies, there's always

more shit to do, and AI can help, and

30

the smart employees figure that out.

31

So if you work in a smaller company, this

is a huge advantage right now, the more

32

swirl there is about AI in the media,

about taking our jobs or killing us, the

33

more your larger competitor will stall

and struggle to handle that anxiety.

34

So press your advantage.

35

Act with conviction.

36

Be David in the land of Goliaths.

37

Take advantage of this moment

38

Greg_Intro: And this week,

I'm talking with Alon Zerbib.

39

He's founder and CEO of Sapiom, which is

an agent infrastructure platform allowing

40

the rest of us to build at scale, secure

and smart agents that work, that actually

41

make money, uh, a-a-and not cost money.

42

Prior to this startup, for which

he's raised $51 million, Alon

43

worked at Shopify, and before that,

he was co-founder of a startup

44

that was later sold successfully.

45

So Alon's an experienced entrepreneur that

this time around is a solo founder and

46

is trying to build one of the companies

that will power the agent economy.

47

Greg: Alain, great to see you.

48

Thanks for joining us.

49

Ilan: Hi, thanks for listening, Greg

50

Greg: Yeah.

51

A- as you know, the podcast is called

"Super Companies" 'cause, and we're

52

all about how to build a super company.

53

I think super companies will

be the most valuable companies

54

in the world in 10 years.

55

We don't know who super companies

are, are yet because to be super,

56

you can't just generate growth.

57

You have to generate earnings,

profits, and consistently.

58

So, you know, today's super companies,

Shopify, Netflix, Amazon, Google,

59

you know, we'll know, in 10 years

who the next super companies are.

60

I think there'll be a different set

of companies that learn how to build

61

their businesses with AI at the core,

both internally and in their products.

62

So their products and services are

different, and, and the way they

63

operate the business is different.

64

I think Anthropic and OpenAI are obviously

candidates to be super companies,

65

and of course, there are hundreds,

thousands of startups, including yours,

66

primarily venture-backed, but not all

venture-backed, that are, are building

67

new firms new potential super companies.

68

And that's why I wanted you, to join

us today to talk about your journey,

69

the opportunity you saw to build a new

company, a super company, and how's it

70

going, and, and then maybe we'll, we'll

end by looking ahead in terms of the

71

future, both for AI and and your firm.

72

But let's start with, Elan,

the opportunity you saw

73

and how did you see it?

74

Ilan: so I'm originating from France.

75

Uh, I'm a software guy.

76

Started coding when I was, like,

twelve years old, so pretty much

77

all my life and, uh, never stopped.

78

computer science degree.

79

Started my career in cybersecurity,

so, I had, quite a different,

80

uh, uh, early career.

81

I was, uh, working for the Department

of Defense in France, uh, but

82

US in two thousand and fourteen.

83

Uh, built the first company

that was actually getting,

84

uh, money back to consumers.

85

Uh, we had an army of bots that

were actually fighting every

86

day to get refunds to people.

87

Uh, so if you buy something on a price

drop, there's price protection benefits

88

provided by retailers on credit card.

89

Uh, and Earny, actually, the company that

I was running at that time, was there

90

to automate all the process end to end.

91

So as a user, you just

buy something on Earny.

92

Eventually, in the first week,

we'll just give you uh, money

93

back on what you purchased.

94

Greg: So your customer was

the consumer basically,

95

Ilan: a consumer

96

Greg: it was a consumer app or product?

97

Yeah.

98

Okay,

99

Ilan: Yeah.

100

So we had, like, five million, uh,

users, actually, where we had a full

101

automations of price protection,

which was actually behind the

102

scene, a massive infrastructure.

103

And we are just running

billions of agents a day.

104

at that time, it was called bot, but

it was really automation, software

105

automations that were fully, and

autonomously getting money back to people.

106

Um, we are going to Amazon, for

example, a billion times a day to just

107

check prices on some claims and like,

108

So we had this massive infrastructure

that we built in-house.

109

and actually, the interesting point

with what we are doing today at

110

Sapiom, it's, Sapiom today is here to

remove the barriers of agent builders,

111

really agents to access the economy.

112

at Earny, uh, we had actually that

massive, uh, scale of, uh, bots that

113

actually were facing exactly the same

problems that agents are facing today,

114

meaning autonomous software that

needs to act on behalf of an entity,

115

a business, a person to provide value.

116

But all the infrastructure underneath

that, like internet, uh, all the

117

infrastructure that powers, everything,

payments uh, the web, uh, the commerce, is

118

actually designed to prevent automation.

119

Has been designed for the last

twenty years to prevent bots

120

because bot we have seen as a,

121

Greg: As a bad thing,

122

right?

123

Yeah,

124

Ilan: something we need to prevent,

125

Greg: I just wanna pause there.

126

that's really interesting insight.

127

I haven't heard that before.

128

But the, essentially the internet was

set up to be unfriendly to bots, right?

129

They, they're, because they're

not real, they're not consumers,

130

they're not human, whatever.

131

They're gonna do things

you don't want them doing.

132

and to your point, bots are agents

essentially, and, and in fact we

133

want, we need now a- an agent-friendly

kind of infra basically, right?

134

Ilan: And the infra is like really

from like the web infrastructure to

135

the payment infrastructure, to the

fraud models to understand actually

136

what's fraudulent and what's not.

137

and now we need to reinvent pretty much

everything to open the gate, to agents.

138

so that was kind of the, the,

the first, part of my career

139

where I spent actually six years

140

Greg: and then what was

that business called again?

141

It was called

142

Ilan: Earny.

143

Greg: Earny, right, yeah.

144

And then what happened, what

happened to that company?

145

Ilan: so we got acquired,

uh, in twenty twenty-one.

146

So after like a six years journey,

and then I, joined Shopify.

147

Uh, so then I was, uh, director,

uh, of engineering in our global,

148

uh, uh, payments product called,

uh, Shop Pay, which is a large scale

149

consumer, uh, uh, payment, uh, product.

150

hundred billion GPV, hundred

of, uh, hundred million users.

151

Greg: So six years as an entrepreneur,

was that venture-backed or

152

Ilan: Yeah, we are series

A, we are series A,

153

Greg: Okay, great.

154

And, and was that a good

outcome or an okay outcome or

155

Ilan: It was a kind of

COVID side uh, side outcome.

156

Greg: Okay.

157

Ilan: So we were automating price

protection, uh, for like commerce,

158

uh, many commerce and the COVID uh,

kind of shifted a lot of things.

159

But yeah, it was good

outcome for the company.

160

Greg: Okay, good.

161

A-and, I mean, you can afford

to live in San Francisco, so

162

it mu- it must have been okay.

163

And then, and then you went

to work for a big company.

164

Ilan: Yeah.

165

Greg: So why?

166

I'm just curious why that

switch from like a, you know…

167

W-were you exhausted?

168

Were you like, "Okay, I want,

I want the, I want a…"

169

Or did you want a bigger

platform on which to build?

170

I'm just curious what was

appealing about, Shopify.

171

Ilan: I wanted to remove my

blind spot as a entrepreneur.

172

Um, so, you know, at Earny, like for six

years, like every stage of the company

173

was new because I was kind of still

young in my career, and I never really

174

knew, uh, how to scale engineering

teams, product, uh, infrastructure.

175

So I really wanted to have the next

phase of my career to, uh, seek, and, uh,

176

learn as much as I could to eventually

go build again and move the company

177

to the, the category-defining company.

178

Um, so that, that, that was very

intentional for me to join actually

179

a very tech-forward, like very

advanced company, but like much

180

bigger, to be able to actually arm

mas- myself with, all the things,

181

that I need to know to really build a

category-defining company the next time.

182

Greg: Yeah.

183

And so, and, and I'm curious,

you did that from the Bay Area?

184

Or did you move,

185

Ilan: No, actually at Earny

we eventually moved to LA, so

186

I was based in Los Angeles.

187

Greg: Oh, okay.

188

All

189

Ilan: When we got acquired and then at

Shopify I was based in Los Angeles as well

190

Greg: Okay.

191

And then your team was, w-

was your team distributed?

192

Ilan: at Shopify, yes.

193

Greg: So you're about, you're

there about five years working

194

on this product called ShopPay.

195

As you said, $100 billion in

gross uh, merchandise revenue.

196

Uh, So pretty significant

line of business.

197

and is that where the insight

came in terms of your, your

198

current startup Sapiom?

199

Ilan: Yeah, because at Shopify, you

know, I went, I went very deep into

200

getting expertise into payments

and really like knowing the payment

201

stack, the payment infrastructure,

designing new payments products you

202

know, working with the biggest player

in the world in partnerships and

203

technologies and really advancing

like, uh, the financial infrastructure.

204

and the idea of Sapiom came in like

in twenty twenty-two or maybe early

205

twenty twenty-three, really when

ChatGPT came out for the first time.

206

You know, it was like

everyone was talking about it.

207

I started to play uh, with it

208

Greg: Feb- February the 1st, 2023 is

my, was my ChatGPT moment, you know?

209

So yeah, about three, three

three and a half years ago, yeah

210

Ilan: so that's the time where I say

I, I really realized that first it was

211

extremely powerful, like just asking

any questions and you get any answer.

212

It's kind of mind-blowing when the

previous generation was Google, and

213

if you had one more word, you have

zero, like, very weird results.

214

and for me, it resonate right away

that eventually those kind of LLMs,

215

like the way that we were able to crack

LLM, uh, actually those models would

216

become more and They will add reasoning

capabilities and because at the end

217

of the day, a model is just a piece of

software, so the only interaction we had

218

at that time was like chat interaction.

219

But because it's a piece of software,

it can actually work in the background

220

autonomously, like any scripts.

221

And that's when I resonated right

away with what I was doing at

222

Earny I said, like, those models

that are extremely impressive now,

223

they were just becoming better.

224

And if I had that technology at the

time of Earny when we had billions

225

of workflows on bots running every

day, it will have been game-changing.

226

And then I said, "Okay, but if you have

non-deterministic software running,

227

you need to reinvent everything."

228

Meaning we need actually to

reinvent the payment infrastructure.

229

Like I understood that those

software were different, shape and

230

very different type of products.

231

And actually the intelligence coming

with that means we actually want

232

to give them autonomy, and they

will not be seen as a bad actor.

233

They will actually be extremely

powerful for everyone.

234

And you will want actually to get those

bots or like agents economic power.

235

So you need to actually give

them access to payments.

236

You will want to

237

have them,

238

Greg: so, so they can

b- so they can buy for

239

Ilan: They can buy things.

240

They

241

Greg: transfer money or whatever,

242

Ilan: Yeah, they can act as an entity

in the background that does good

243

things for consumers or enterprises,

and they will need actually to be

244

able to access and act in the economy.

245

So that's how I got the idea of Sapiom

It's okay, eventually, the software

246

will become mature enough where they

will need to access the economy.

247

And actually, then I was waiting the right

time for, like, the technology to mature.

248

And it took two years.

249

Uh, then the models become better, the

reasoning better, like, become better.

250

And last year, early last

year, that's when we started

251

to see like chain of thoughts.

252

We started to see MCP.

253

So now you have actually those

very smart models that can start

254

communicating between them.

255

And then we started to see emergence

of payment rails last June, and that's

256

when I left Shopify to actually build

Sapiom that really empower those agents,

257

Greg: and, the, the core of this

idea or opportunity is, and, and

258

it, as you said, basically allowing

agents to act in the economy.

259

And really what that means is on

behalf of consumers or businesses

260

buy stuff or, or, or, or, or, or

complete economic transactions

261

Ilan: Y-yeah, or just complete outcome.

262

And most of the time completing

outcome means you need to access

263

other services, other software.

264

You need to buy APIs.

265

So my, uh, my thesis is

really on the machine economy.

266

So there is a market on the consumer

economy, meaning agents buying

267

flight ticket or pair of shoes.

268

So they need like credit cards,

uh, they need access to payment.

269

But this is an economy that's

not going to grow significantly

270

larger because of agents.

271

Uh, the bet at Sapiom it's actually we

are powering agents that needs to pay for

272

APIs on other services, on other agents.

273

So it's really a machine-to-machine

transaction that is kind of micro

274

transactions, and this is as a--

at a scale that we've never seen,

275

uh, because agents paying APIs,

services search generating tokens

276

is a, a massive, uh, new market

277

Greg: So you're not focused

on the consumer application?

278

Ilan: Not, not today.

279

Today we're really focused

on agents getting autonomy

280

Greg: So agents buying APIs, Propose

a real, a real world use case for

281

Ilan: Yeah, the, the way I see it, it's

agent or a piece of software that need

282

to communicate with other pieces, like

of software to do any type of actions.

283

Um, and the way we do that today, it's

actually you give agents like API keys.

284

So you sign up to a, a SMS

provider, you sign up to

285

an email provider, to image

generation, search browser, you

286

know, all the capabilities that

agent needs as a human, and then you

287

seed your agent with those API keys.

288

Actually, if you think about that, that's

extremely limiting for the agent because

289

the agent is a very smart entity that can

do pretty much anything, but at the time

290

of action is limited by a set of tools.

291

Greg: Right, what the API allows,

292

Ilan: yeah.

293

And the way I see it, it's similar to,

I don't know, you, Greg, going to Paris.

294

It's a beautiful f- city.

295

You land in the airport, and when you

land, I give you two gift card, one

296

for Starbucks and one for McDonald.

297

And it's great, you're in a city,

you want to go to museum, you

298

have a beautiful hotel, but you

can't even take Uber because you

299

are limited in what you can do.

300

So the way we see agent, it's

actually, okay, let's give them

301

freedom in a way that is governed.

302

So actually, let's

control what they can do.

303

But we need to give them a payment method,

so they should be able to pick and buy the

304

tools they need for any type of intent.

305

So actually, what Sapiom is, is we

give payment capability for agent.

306

And now, rather than giving

five to ten API keys to an

307

agent, the agent technically

have access to the full economy.

308

And then you provide the governance

layer where you provide like a business

309

credit card where there's some limits.

310

You know, don't spend more than two

hundred bucks on a hotel, on the uh, uh,

311

you can't buy those type of products.

312

So that's kind of Sapiom We

have the governance layer that

313

regulate what the agents can do.

314

The platform is authorizing

every transactions.

315

But actually, the agent is free

to pick any type of systems

316

or services it needs to do the

outcome they're trying to achieve.

317

Greg: what would be an example of

yeah, what, what the agent is doing

318

and how, how they're now enabled

to do something they couldn't do?

319

Ilan: Like a marketing agent like,

you know, there's many use cases,

320

but marketing agent is kind of, okay,

you have an agent that for every uh,

321

user subs- uh, signing up to your

product you eventually want them to

322

sign up to subscribe to a paid plan.

323

Uh, and there is

different ways to do that.

324

And actually you want to have a marketing

agent that or, uh, an upsell agent that

325

go find a list of all the signups and then

try to upsell them to the best product.

326

And maybe you only give gi-gave

to your agents, uh, an API key to

327

send a, an, an email and that's it.

328

Uh, but actually maybe some of

your users they sign up with phone

329

numbers or maybe it's much better

for them to convert with an SMS.

330

Or maybe you have users in

the countries that your email

331

provider is not uh, working.

332

Uh, so actually you want your agents

to be able to decide at the time of

333

execution, okay, for this user, the

best way to upsell them to my product

334

is actually to send him an SMS,

send him a an email maybe sh- send

335

a pu-push notification on his phone.

336

So you actually want to open

the panel, the, the panel of

337

tools that your agent needs.

338

But as a human building the system,

actually you don't know all the kind

339

of the scope of all the possibilities.

340

So actually the way that payment…

341

Greg: you can't, and you

can't predefine them.

342

There's

343

Ilan: And you can't predefine it.

344

Uh, So actually the, the way you do

it is you limit your agents, which was

345

always the case with software because

software was very deterministic.

346

But now that agents can reason, actually

maybe they will find innovative ways

347

to do the goals you are giving them

that you didn't even think about.

348

But in order to do that,

you can't limit him with the

349

tools they can have access to.

350

Greg: So who are your customers then?

351

Are they developers or are they

enterprise you know, technology teams?

352

Like who, who is…

353

What are they buying

from you guys and how?

354

And who, and, and, and who is buying?

355

Ilan: yeah, so the, the, the

second product actually is we

356

have actually a developer product.

357

Uh, we

358

help developers and like

builders to build agents.

359

So we have an opinionated

framework, a product called an

360

Agent Studio that actually help

you design, uh, production-ready

361

agents in a multi-agent setup.

362

Like because an agent usually

it's kind of different agents that

363

communicate together to do an outcome.

364

So we actually have an opinionated

way to help builders to

365

actually build those systems.

366

So that's the entry point.

367

And then we have the runtime environment,

which is kind of an agent cloud, where

368

you can actually deploy your agent on

Sapiom, and we provide a scale, meaning

369

like your agents can get to hundreds

of thousands of execution a day.

370

We manage the infrastructure for you.

371

We provide the security, the governance.

372

We make sure that the cost stays

under control, so you have full

373

visibility into the cost of your agent.

374

And we make sure that everything

stays safe is scalable and reliable

375

Greg: So this, this is sort of an

en- was this sort of a Trojan horse

376

idea in terms of this product?

377

Uh, or the, uh, the kind of

the, the entryway into…

378

Ilan: Yeah, it's more like a

platform because there are so

379

many problems actually today what

we've seen with early customers.

380

It's extremely hard to

get agents to production.

381

It's actually easy to build a

demo with Cloud Code or Codex.

382

But actually, when you start getting real

users on scale, everything start breaking.

383

The hardness of the agent is hard.

384

To make sure that it

stays secure, it's hard.

385

To make sure that the cost

is under control is actually

386

extremely ex-extremely complex.

387

Uh, to understand the

metering the tenant isolation.

388

So there's a lot of complexity into

building a real agentic product.

389

And Sapiom is here to eliminate

those barriers to actually make

390

agent builders, uh, give them all

the tools they need to build and

391

ship production, uh, scale agent.

392

So our targets are like developers.

393

Uh, we are as-- We have as well,

uh, customers that are building

394

AI native companies at scale.

395

For example, we're working

with a company called Pulsia.

396

They're automating businesses.

397

So you create an account, and then

you have tons of agents that are just

398

building a business every day for you.

399

All those agents are running on

Sapiom, and we're targeting as well

400

enterprise because we're helping

enterprise deploy their AI strategy

401

and find ways to deploy agents and

get leverage from agentic systems.

402

Greg: how do you serve enterprise?

403

Not developers, but enterprise

404

Ilan: Yeah, the enterprise, it's really

on the governance and observability.

405

So governance of what the agent can

do, authorizing every agent actions,

406

making sure that, okay, when the agent

access a system, we have the authority,

407

like he has the authority to do that.

408

And when he does we have the tracking,

meaning we can audit every agent action.

409

So Sapiom has a, a product that is really

focused on the telemetry observability and

410

like the authorization of agent actions.

411

Uh, So as a business, you can deploy

agents in production in your enterprise in

412

a way that is actually secure because you

can control what the agent can access, but

413

you can as well audit what he did and why

it do-- he did those decisions So that's,

414

uh, uh, enterprise, uh, product that is

really tailored on, uh, observability.

415

And I would say the first bucket is really

companies building agent as a product

416

that they want to sell to their customers.

417

And then what matters as well, it's the

cost, making sure that you can meter

418

the price of every agent run for your

end customers and Sapiom provide the

419

uh, uh, data as part of the platform

420

Greg: I was talking with someone earlier

today, Ilan, that, that used this

421

expression uh, s- I just thought it was

so, um, kind of useful and simple, which

422

is, is your agent bankrupt or solvent?

423

Or, you know, like, just…

424

And I hadn't heard it before.

425

I'm sure others have used it, but I just

thought it was a really good sort of sharp

426

focus on, yeah, there's so much hype and

you can prototype agents and you can pilot

427

them, and you can push some of them into

production, but still you're not done.

428

You have to go and take a look later.

429

Is this solvent or not?

430

Ilan: I will say to them most of

the agents are actually bankrupt

431

Greg: Is that right?

432

Ilan: like it's, uh, it's very, uh,

it's very hard actually to build a, a

433

margin positive agentic products today.

434

We're helping our customers with

that, not just by giving them

435

observability into the cost, but

actually we help really reduce cost.

436

Uh, what we've built in Sapiom

is we have our own router that

437

is really part of the platform.

438

But for every agent terms, every

agent execution, we actually pick the

439

best model and the best tool for the

agents, optimizing latency, quality

440

of the outcome, and as well the cost.

441

Greg: and the cost, right?

442

So, I, I wanna zoom

out just on this point.

443

I just find this obviously fascinating

in terms of this moment that we're in.

444

Do you worry about that?

445

Meaning, if most agents are, are

bankrupt or not, not solvent in terms

446

of, so this AI bubble that we're in,

that there's a lot of, lot, lot of

447

inference being spent and a lot of

effort being spent on, product services,

448

agents, uh, AI applications that just

aren't, you know the ROI is not gonna

449

be there or isn't currently there?

450

Or do you think this is a natural

phase that we have to just go through

451

and m- kind of muscle through it,

where we'll figure out, you know, the

452

best use cases, the best agents and,

and uh, and the best business models?

453

What, what's your, what's your take

on this moment right now that we're in

454

to- you know, towards the end of 2026?

455

Ilan: No, I, I think it's

a, it's a natural phase.

456

I think we like, we already figured

it out with our key customers.

457

We have customers that move from

like minus a few hundred percents,

458

uh, negative margin to actually

very, very, uh, high margin.

459

so they can actually invest

more into marketing, on growing

460

their product, on their team.

461

We really want our customers to win.

462

Like the market is very new and

every customers using Sapiom, we

463

want them to win their market.

464

And actually that comes with, let's

make sure you have margin positive

465

and actually you can reinject more

of your revenue into your business.

466

And, uh, we've been able to do

that actually thanks to, on one

467

end, open weights models that

are extremely powerful today.

468

So we're able to reduce the cost

of the intelligence by ninety

469

percent for some, uh, customers to

the way that we have opinionated

470

frameworks on how agents are built.

471

So we, we reduce the

failure rate of agents.

472

Because the thing if an agent failed,

you already paid for everything.

473

So if the agent failed at the

last step, uh, you paid, uh, for

474

some type of agent a few dollars.

475

And the failure rate is actually

quite high today on agentic products.

476

So we reduce the number of

failures, and actually we-

477

Greg: when you say failure rate, you

mean this agent's not doing what it

478

was supposed to do, or it's not doing

what it's supposed to do reliably

479

and consistently, or what it, what it

ends up doing doesn't create economic

480

value, or maybe all of the above?

481

I don't know, like,

482

Ilan: actually, actually

it's, it's all of the above.

483

And today agents are kind of a black box.

484

Sometimes you ask him to do

something, ask him to do something,

485

and he does something different.

486

Or sometimes because there's a lot

of infrastructure pieces behind the

487

scene, I don't know the last step.

488

Your agent needs to trigger a

sandbox, and there is a failure on

489

the sandbox, and everything fails, so

you have to retry from the beginning.

490

so there's a different type of failure.

491

Sometimes, like you plug API keys into

your agents and actually you run out of

492

fund on one of the API keys, which is

actually a very common patterns we are

493

seeing with builders today because they

sign up to many systems, put their credit

494

card, and eventually their agents will

use those systems and one of the credit

495

card will run out of fund and then the

496

Greg: So they'll, they'll

they'll, they'll blow the limit

497

on the credit card, basically.

498

Ilan: Yeah, yeah.

499

Or like the, the, the

credits that they have

500

Greg: yeah, right.

501

Yeah,

502

Ilan: one of the platform

that you have to sign up.

503

So that's all the, the, the, the failure

points that actually Sapiom is eliminating

504

and where we're helping our customers.

505

And sometimes you're just using too

much intelligence where you don't need.

506

Greg: And is your point of view that,

anyone who's managing agents or mana-

507

ma-managing AI, maybe the, you know, the

head of AI, let's call it, that, that,

508

that individual, that person is, gonna

be always using a mix of models i-in

509

terms of managing their inference cost.

510

Like that, that just, that, that is

gonna be a key skill, if you will, of

511

anyone who's, who's gonna be successful

in this economy so to speak, would be

512

someone that knows how to basically pick

the right models for the right task.

513

Is that, is that sort of we're

not gonna be all, all in on just

514

OpenAI or Anthropic or whatever.

515

We're gonna have to really

manage the portfolio.

516

Is that how you think about it?

517

Ilan: Yeah, definitely we'll not be

all in on one f-frontier model because

518

there is other options that are much

more cost-effective and actually better

519

because they're more specialized or…

520

but I don't think every teams

will be able to do it because it's

521

extremely complex to do evals, to even

understand what, what good look like.

522

what is a great outcome?

523

Because it's very nuanced.

524

i-it's not a zero to one, you

know, if an agent send an email.

525

Okay, is the email, the

writing is good enough?

526

Like, you don't know.

527

And today, the-- what we are seeing, it's

actually every company's building AI,

528

they have to reinvent all those things.

529

So that's why SAPLM is here.

530

We figure it out.

531

Uh, we have enough volume to actually

learn all the right patterns.

532

We know exactly the best models.

533

We abstract that for you, so you as

a builder, you don't have to focus on

534

that piece of infrastructure that is

actually extremely hard to crack and

535

will require a lot of focus for you.

536

So you can actually focus on your

deep domain expertise, which might

537

be building a, a, an accounting agent

or building a, an agent that does

538

marketing, which is really where

you have the, the moat and where you

539

have, like, your highest leverage.

540

Focusing on the orchestration of the

agents, the context you give them, what

541

they're trying to achieve, rather than

rebuilding the infrastructure behind it

542

that just makes those agents reliable

543

Greg: So I, I get that you're mostly

supporting today builders who are

544

building agents, and as you said, you

kind of, you, you, you, you kind of

545

provide that infra layer for agents

so they can focus on the end user

546

essentially, or the end user application.

547

Will you do this for big company-- Will

you do this for companies or big companies

548

who need to build a lot of agents

themselves and also don't wanna deal

549

with everything you just talked about?

550

Presumably that's either on your roadmap

or something that, that you want as

551

customers, people you want as customers.

552

Ilan: Yeah, we are actually doing it.

553

Uh, we have actually part of our team

that is, uh, uh, deployed engineers.

554

So we're actually deploying engineers

in customers', uh, environment.

555

Uh, it could be like AI companies,

but as well a lot of enterprises

556

that actually want to deploy agents.

557

Uh, they don't really know how to do it,

and like actually no, no one really knows

558

today, uh, how to, uh, get leverage from

agents because it's complex and it's new.

559

Because if you think about the

market, we've been talking about

560

agents for a year, but until four

months ago, agents were just chatbot.

561

You know?

562

It was like a human in the loop.

563

Actually, autonomous software is a new

concept for like four or five months ago.

564

OpenCloud was kind of the first

wave of autonomous software.

565

And actually we're helping enterprises

to deploy agents to operationalize

566

some of their workflow to get

leverage for some of their products.

567

And we have like a many engineers

that are working in enterprises

568

for, for that.

569

Greg: So let's talk about your

firm now and super companies.

570

And I, I think this is gonna be…

571

you, you can compare

to your first startup.

572

let's start there, meaning your first

company, which was probably 10 years ago

573

or maybe longer when you started it, then

you sold it, and you built that, and, and

574

now you're building this new business.

575

You got, uh, you got also

venture capital this time around.

576

Tell me about what's different now.

577

First of all, let's talk about you.

578

How are you different now in terms

of building this, this new business,

579

and then how's the business being

built differently, uh, on the inside?

580

Or, or, or is it?

581

Ilan: Yeah.

582

If like it's very different.

583

First on my previous company

I was the technical founder.

584

We are three co-founders, so we had

actually a, a, we are moving much slower.

585

But at that time it was fine because

building businesses took years and like

586

market were much slower than it is today.

587

So I was very focused on my, my lane

to become the CEO of the company.

588

Uh, I'm a solo founder meaning

I'm able to move extremely fast.

589

I know where I'm going, and I'm able to

execute every step at lightning speed, and

590

I'm bringing the team to do that as well.

591

Greg: and did you think

about a co-founder or…

592

'Cause you know the, I think

the data shows that solo founder

593

companies have a lower success rate.

594

I, I don't, I don't know by how much,

but I'm sure there's YC data about this.

595

But was that intentional in, in your part,

meaning you wanted to be the solo guy, you

596

had a vision, you wanted to run at this at

this kind of lightning pace or, or yeah,

597

just talk a little bit more about that.

598

Ilan: Yeah, it was my decision.

599

Like, I had opportunities to

bring co-founders very early.

600

But you know, it's a

very technical product.

601

I spent, like, twenty-five years

very deep in engineering, both as

602

a IC, you know, really building

systems, architecting systems,

603

scaling them, and as well as a leader.

604

I was managing a pretty

large teams at Shopify.

605

And I have the v-- So this

is a technical platform.

606

The vision of Sapiom is to

power the next trillion agents.

607

We'll see trillions of autonomous

software operating in the

608

economy in the next few years.

609

We'll be the platform

that powers all of them.

610

Um, so I have a very ambitious vision.

611

I know to get there it's not an easy path.

612

Like, you know, there's a--

every startup is a rollercoaster.

613

Uh, but I know to get there, I know to

raise capital, I know to build a team,

614

I know how to hire very good engineers.

615

So actually, there is no need

for me to have co-founders.

616

And on the areas where I want

actually someone that is more

617

specialized, I will hire someone.

618

Uh, so that was kind of

my thesis, to be able

619

Greg: don't you want someone to talk

to at 10 o'clock at night when you're

620

m- when you're, when you're unhappy?

621

Or is that your, is that

your spouse or partner?

622

Like, who, who, do you, who do you,

623

lean on?

624

Ilan: I, I have my spouse.

625

I have my spouse.

626

She's very supportive.

627

I have as well very, very

strong in- investors.

628

So actually my cap table, I build

my cap table m- mainly based on

629

people rather than investors.

630

I really wanted like a very supportive

investors that are very strong in

631

their domain, but that are very open

as well to discuss strategy and help me

632

because they can do pattern matching.

633

So I have-- I'm involving actually some

of my investors quite a lot you know,

634

weekly sync and getting them up to speed.

635

They're kind of my I can bounce ideas

with them and just my network of founders

636

Greg: I'm just curious, do you feel

there's a risk there, Alon, in terms

637

of letting your investors sort of

that close to the company in terms of

638

they can see how the sausage is being

made or not being made, so to speak?

639

I'm just curious how you thought

about that balance of they're

640

investors but they're also sort

of, operating advisors, I guess.

641

Just h- did you think about sort of

the risk of that or, or, or you're

642

just w- way more upside than risk?

643

Ilan: I definitely thought about it.

644

I think there's much more upside because

actually I handpick my investors.

645

We did three funding rounds

over the last twelve months.

646

Each one we were kind of preempted.

647

So I always had the opportunity

to actually pick where I

648

want it in the cap table.

649

Every round we're highly competitive.

650

And I'm a people person, so I build

relationship with my investors.

651

I believe like the bene-- anywhere,

like I want the company to be

652

transparent with investors.

653

Like it's our market.

654

Everything is new.

655

Everyone understand that the

opportunity is infinite because

656

AI is going to change everything.

657

Uh, and I have high trust and they

have high conviction in what we do.

658

So,

659

Greg: Cool.

660

How much money have you guys raised?

661

Is that public?

662

And h- how

663

Ilan: Yeah, we raised, uh, 51 to date.

664

Greg: Total, total.

665

Okay.

666

I know the first round was 15, I think.

667

So you've done sub-

668

Ilan: No.

669

Yeah, we raised actually a

proceed of 4 million in August.

670

Uh, then we raised the 12 million in, uh,

January from Accel, uh, and then we closed

671

our series L as, uh, two months ago,

672

Greg: Great.

673

Awesome.

674

And then how, how many people

675

Ilan: So we are a team of 30 people now.

676

Greg: So small,

677

Ilan: yeah, we have like 22 engineers

and then other people for go-to-market on

678

Greg: so what's different about this

company f- when compared to your

679

last one and/or Shopify in terms

of how, how you're building it?

680

You know, what, what do you observe

the top two or three things that

681

in terms of the either the pace

or the scope, the ambition, the,

682

the reliance on AI internally?

683

Just t- talk about the business on

684

Ilan: Yeah.

685

first ambition we didn't really have

that grand Dios vision previously.

686

Like we had, you know, we

are very ambitious, but it

687

was a very different market.

688

The way I see it is

it's quite interesting.

689

It's-- Actually, all my life, I kind of

regret not being born ten years before

690

because I missed the internet era.

691

Uh, me, like the last fifteen

years, twenty years were quite

692

boring building startups.

693

Like, well, fifteen years.

694

The most, the, the, the latest,

like amazing companies that we

695

have built were kind of Shopify,

Uber and Airbnb, you know.

696

Then we had good businesses that

were built, but it was just a bit

697

better than the previous generation.

698

But really category,

category-defining businesses,

699

it's

700

Greg: Trans- transformative businesses.

701

Yeah.

702

Ilan: Yeah, we didn't see

s- new businesses like that

703

for like ten, fifteen years.

704

Uh, And I think now AI is

kind of resetting everything.

705

I think it's kind of the internet

era, but multiplied by one

706

hundred in terms of opportunity.

707

So it was a no-brainer for me to go build

because there is an unlimited market, uh,

708

opportunities everywhere, and there is

really opportunity as a strong founder

709

to build category-defining companies

that will last gen-generation, decades.

710

so first, the vision and the

ambition is extremely different.

711

The market is extremely different,

where the speed matters a lot.

712

As I said, at my previous company,

you know, building a company for a few

713

years and, you know, scaling slowly

and things were fast at that time, but

714

it was still slow compared to today.

715

In a year, like we move

at lightning speed here.

716

We are three founding around.

717

Uh, we are building a

very s- impressive team.

718

Uh, we are locking customers, like the

product took a shape extremely, uh, uh,

719

fast thanks to AI, because actually AI

helps you to move faster and just the

720

opportunity is so much bigger everywhere.

721

Um, so it's really

different way of building,

722

Greg: what's that mean for

you and the team though?

723

Does that mean like we can go so fast so

we can, we can try more stuff, meaning

724

the f- like the cost of failure is lower?

725

Or does it…

726

Like, like what's, what's it really…

727

I mean, like, you know, speed

is a moat and all this bullshit.

728

I mean, I get it, but like

what's it, what's it really mean?

729

Does it mean that you can build

out the roadmap faster and

730

therefore get to an answer faster

in terms of this vision you have?

731

Or is it we can try things and again,

as I said, the cost of failure is lower?

732

Like what, h- how do you interpret

or get value from the speed?

733

Ilan: Yeah, the experiment is

kind of dropped to almost zero,

734

like experimenting something.

735

We are actually building like prototype of

like things we are trying to build, like

736

very early, like real prototype that we

can test even before shipping any product.

737

So the cost of like

testing is extremely low.

738

First-- second, the, the speed of

building is actually much faster.

739

It's-- we are not as fast as most

of the companies out there because

740

actually we, we don't write code.

741

We are building a kind of deep

tech infrastructure platform.

742

It requires, um, ninety-nine

point ninety-nine percent SLA.

743

It requires like a pla- a platform

that sustain load, that can scale, and

744

actually you can't write code that today.

745

Like, it's not just prompting on asking

codecs to build an infrastructure.

746

It requires like very deep

engineering skills, architecture.

747

So we are spending actually much more

time on building than other companies

748

that are sitting at the application

layer, but we provide the reliability.

749

Greg: But you're still mo- even at that

750

layer, you're moving, you're, you're

moving faster than what-- than you've ever

751

moved becau- because of coding agents,

because every engineer on your team is, is

752

Ilan: Yeah.

753

Yeah.

754

We automated a lot of internal

workflows across every

755

department, across every function.

756

We are using, we are

using agents everywhere.

757

And as well, the way I see it, it's

even the way I build engineering

758

team today, it's very different.

759

Previously, even at Shopify, we

used to have like team of, you know,

760

seven, ten engineers working on one

problem with one product manager, one

761

engineering manager, one designer.

762

So it was kind of a big team to

work on a kind of subset of the,

763

the, the stack or the problem.

764

Where at Sapiom, we are very different.

765

It's every engineer, they own one part

of the stack, one KPI, one function,

766

and then they figure out with AI

like leading indicators, success

767

metrics, what do we need to build?

768

How do How do we solve those problems?

769

So actually it's much leaner.

770

And actually every engineer, every

individual, they have much more

771

ownership and agency to do things.

772

So we act-- can move horizontally

rather than vertically.

773

Rather than building one team that does

one thing, actually have every individual

774

that act as a team, as a part of one,

and we can move much faster as a company.

775

Greg: that is working, this idea of an

engineer owning his or her kind of roadmap

776

Ilan: we still need to layer on top

of coordination because we tried

777

earlier when the team was smaller

and it worked, but as you scale,

778

you still need like coordination on

making sure that the vision kind of

779

goes down to the team on what we do.

780

But yeah, we are able to

operationalize that and scale it

781

Greg: is why how, how you faster get

the breadth of roadmap basically, right?

782

In terms of…

783

and do you think about it, and I'm, I'm

jumping around here, but do you think

784

about it as, as you're willing to spend

so much on inference per engineer?

785

I'm just curious how you think

about you know, how AI augmented

786

your engineers are and what…

787

Like, if you're spending I

don't know, three, 400 grand an

788

engineer, what would you spend on

inference for the same engineer?

789

Ilan: yeah, we don't limit token.

790

Sometimes we try to optimize a

bit, but we don't limit it for

791

engineers because that's the highest

leverage or for any individual.

792

But we have some engineers that

like the cost of their salary is

793

similar of the cost of inference

794

Greg: inference.

795

And, and in your c- and in your

mind, that's a, that's a no-brainer.

796

Like

797

the, the e- yeah, absolutely right.

798

The economics of that makes, make

total sense, at least at this moment.

799

Yeah.

800

Do you think that happens in the

rest of knowledge work if you

801

had to and maybe even ins- inside

of your company in terms of…

802

They, they, they won't probably

spend that much 'cause they don't

803

n- they don't need that many tokens.

804

But when you think about marketers or

salespeople or the, all the business

805

functions, i- is your sense that

that inference will be a significant

806

spend for them as well over time?

807

Ilan: It should and I

wish it will be the case.

808

Because if people use tokens,

it means they are kind of

809

getting, getting leverage.

810

Not always.

811

We need to make sure that there

is the right training as well

812

on how we can leverage by AI.

813

It's

814

not just,

815

you know, getting a prompt on tokens.

816

there is some structure

and some best practices.

817

But yeah, we-- I believe that the

company will eventually be eighty

818

to ninety percent fully automated.

819

And then it means we can move at, you

know, a speed much faster than anyone else

820

Greg: You believe at some point in the

future, 80 to 90% of the company, all

821

functions, not just software engineering,

will be automated in some way?

822

Ilan: Yeah.

823

And that's why I-- even when we,

you know, my head of talent, that's

824

kind of one of our key topic.

825

It's okay, how do we automate talent?

826

How do we think about talent

function from the ground up in a

827

way that it's kind of AI native?

828

So there's high touch moment

with candidates and with the

829

process, but actually most of the

process could be leveraged by AI.

830

So it's really trying to figure out where

is the border between what do we want

831

to keep very human, and that's where we

invest a lot of people and resources on

832

making sure that we are the best at that.

833

But everything else, actually

let's find ways to automate it.

834

Greg: do you think about one of the end

results of that, Alon, one of the key

835

metrics is basically revenue per employee,

and that the revenue per employee at,

836

at your firm will be better than your

previous firms and/or your competitors?

837

Like, do you think that's a key m-

do you think that's a key metric

838

basically for investors and, and,

and the rest of us to evaluate?

839

Are you a super company?

840

you know, Anthropic's probably at $10

million revenue per employee right

841

now, some number like that, right?

842

When, when Google was at two,

two and a half million, that

843

was, that was great for years.

844

how do you think about that metric?

845

Ilan: Yeah, I think that

metric will keep growing.

846

Um, it's not a metric we are looking

at, and I don't-- I never see that, that

847

metric like being asked by investors.

848

But eventually, yes, that will

get you a competitive advantage

849

Greg: so in your definition of a super

company, it sounds like that a super

850

company would be, would be a, will be a

company that is, automated a lot of work.

851

You said 80 to 90%, I think.

852

That, is that, is that what, in your

mind, is that the kinda core attribute

853

or key attribute of a super company?

854

Ilan: Yeah, but it doesn't mean

the company will be smaller.

855

And that's the way I see it, and that's

the thing I'm trying to crack here.

856

It's actually if you build it in the

right way, it doesn't mean you build, uh,

857

rather than having a one hundred people

company, you have a ten people company.

858

It means you still have a one

hundred people company, but actually

859

you do ten times more things.

860

And that's how you figure out

what's the best structure of teams

861

to actually give more ownership

on agency to every individual, so

862

actually you can create more lanes.

863

Because there's opportuni-- there's

massive opportunities in the market.

864

There's a lot of problem to solve.

865

We, for example, just launched

two months ago a lab, so we have a

866

researcher that just spend days because

we're already working at the frontier

867

of what's the RNS versus model.

868

How do we leverage like,

agents to do more things?

869

When do we understand how they fail?

870

How do we make the smart, the mo-

the roo- the smart router better?

871

So there's a lot of frontier things

that there is no zero or one answers.

872

So we started to build a lab.

873

We have a few researchers

that are working on the lab.

874

And those are the initiatives that

actually you can have many parts like

875

that and have another thousand ideas that

I would love to have people to work on.

876

So I don't think

companies will be smaller.

877

It's just we will be able to do much more.

878

Greg: H-Alon, how do you

manage your own ambition then?

879

'cause the scope is possible.

880

H-how are you as an entrepreneur or

CEO, you know, not, doing too much?

881

Or not…

882

Or, or, you know, executing well enough?

883

'Cause the, the risk, of course, of

of ambition and, and scope is that

884

quality suffers, or you're distracted,

or you're in too many markets, or

885

you don't have an ICP, or, you know,

all, all, all the risks of, of doing

886

too much, which I feel some days.

887

So h-how do you manage that?

888

Ilan: Yeah, it's interesting.

889

So that's definitely a

top of mind, uh, for me.

890

It's being very grounded

as well in customers.

891

First, on one end, there

is engineering excellence.

892

That's my background.

893

So actually, I have very, very high bar

for talent on what is great in building

894

products and building, uh, software.

895

Uh, so making sure that we meet

the bar, and it's easy to move

896

towards like AI slope on some areas.

897

So catching it very quickly

on uh, resetting the bar.

898

So very high bar for talent on outputs.

899

And on the second end, it's actually

going very deep with customers.

900

And actually, we spent a few months

like very, very deep with one of our

901

customers that was growing extremely fast.

902

All the company was focused on that

to make sure that they can win and

903

to make sure that we actually absorb

the most of the learnings of what

904

agents are doing, what they need when

they fail, why they fail, how we can

905

scale them, how we can secure them.

906

And I think it's breadth as a company,

but actually very, very depth with

907

customers to understand every single

pain points, give them We have engineers

908

that are deployed with our customer--

in our customer's code base, and

909

it's really learning and accelerating

the learning, being very deep with

910

customers because at the end, it's kind

of deep tech, and we have to get those

911

learnings to build the best product.

912

Greg: Yeah, no, I love that idea

though, that framework, the simple

913

fra- you know, sort of br- breadth

of roadmap or breadth of capability,

914

but deep with key customers.

915

And you can probably go deeper

faster, obviously, in this moment

916

than, than you could, you know,

10 years ago or five years ago.

917

let's look to the future, as

it relates to super companies.

918

Uh, revenue per employee,

yeah, should, should be great.

919

Uh, an automated company, but

doesn't necessarily mean smaller.

920

It probably means you know, more

ambitious and being able to tackle, you

921

know, more markets, more opportunities.

922

What, what else do you think

will define super companies?

923

And w- and, and your own firm.

924

When you think about a year or two,

and we're talking again how will

925

you be operating or what will be

different about the business when

926

compared to your previous companies?

927

Ilan: yeah, so definitely like, the way

we build with the team, the ownership,

928

like really the, the, the culture.

929

But I think on the other hand

as well, it's the quality of the

930

product and the things you build.

931

Because with AI actually, the cost of

software is kind of getting close to

932

zero, so anyone can build anything.

933

So it's really how do we build

great software that will just work,

934

and not works in prototype, but

really work on scale and deliver

935

massive value to the end users.

936

And this still require craft.

937

It requires extreme uh,

938

Greg: care, right?

939

Judgment.

940

Ilan: Yeah, extreme care,

941

re- really, right, paying

942

Yeah.

943

yeah, Attention to detail.

944

So those are really the, the, the things

that we focus on as a team and me as

945

a leader, making sure that every new

employees, every new team members, they

946

care about what we do, they care about

customers, they care about the details

947

and leveraging AI to make sure that we can

move faster but without lowering the bar.

948

And that's actually not easy.

949

And On the other side, I

think it's being right.

950

It's super hard today to understand

where the market is going, and it's

951

making bets that are hard to make, but

having a, a good sense of where the

952

market is going, what the technology

is going to deliver to be able to

953

make those bets early and be able to

leverage them when the time is right.

954

And that's that's more an art

955

Greg: Yeah, it's more an art and it

seems really hard these, these days in

956

that the, the pace of everything and

the, and you know, companies moving

957

up and down the stack you know, l-

like model companies deciding to focus

958

on applications and, and, and so on.

959

Just how do you feel about that?

960

you seem excited about it, exhilarated

about it, but also, like, do, do you find

961

at moments it's like, "Oh shit, this, it's

hard to figure this out," and or the rate

962

of change is just k- kind of stunning?

963

Ilan: yeah, I'm quite excited by that.

964

Um, the-- we are not an

application lay-layer.

965

We're really like an

infrastructure platform.

966

So we power really the application.

967

I don't think OpenAI and Anthropic

have a right to wins in every market.

968

I think the deep domain expertise

on, you know, being very

969

specialized and really caring.

970

Because the size of OpenAI, Anthropic is

quite large now, and most of the things

971

they are doing are kind of side quests.

972

And at Shopify, it was really the

I don't know if, uh, you know Tobi

973

well, but he's the Shopify CEO.

974

He's always, like, building a platform

that powers the commerce but still

975

very focused and making sure that we

are always focused on the main quest.

976

And when you do side quests, usually

they are not well-funded on those labs.

977

A side quest would never win a founder

with a team who really care that

978

want and have the appetite to win.

979

So I think there's always room

for, like, any type of products

980

to emerge and win their market.

981

And there's just so much things to do.

982

Like, everything is going

to change with AI, so

983

Greg: All right.

984

Thanks for joining.

985

I loved hearing your story and just this

idea that we'll, we'll automate s- 80-plus

986

percent of our companies, not to reduce

headcount necessarily, but to expand

987

scope and to accelerate our opportunities.

988

Just, uh, it's an amazing

moment for entrepreneurs

989

Ilan: Il se va être maman

990

Greg: great to meet you again see you

again, and let's check back in a year.

991

Ilan: Let's do it.

992

Greg: All right.

993

Thanks, Ilan.