シャドーイング練習: How an Ex-Googler Accidentally Founded a Database Company - 動画で英語スピーキングを学ぶ

練習モードを選ぶ

各モードは全画面の練習ルームで開きます。

532 文

文が短すぎたり長すぎる場合は、Editをタップして調整してください。

There's a war for attention.

If you build something that nobody uses, it might as well have not been built.

This is Jordan Tigani.

With over a decade of experience at Google, and specifically BigQuery, he shares why he started a new company and how he's building Mother Duck.

I'm not sure I'll ever have another idea that's this good, so I might as well go for it.

What is your advice for people that want to get into specifically this niche of engineering?

You just had the partner meeting.

How much do you want?

And I'm like, I don't know, $5 million?

Could you share the mother duck origin story from your perspective?

Sure.

So, I mean, first of all, I was a single store database company, formerly known as MemSQL, and I had just gotten fired.

I was a chief product officer, and I had gotten fired on a Friday morning, and I'm like, you know,

it was like, hit me sort of like a ton of bricks, like, what the hell am I going to do?

And I tried to figure out what I wanted to do.

And I had seen this great open source database called DuckDB.

And I've been thinking, you know, somebody should take that and build a SaaS service out of it.

And I'm like, well, I've built two database SaaS services, you know, first BigQuery and then the SaaS service for a single store.

And I'm like, huh, maybe that should be me.

So I'm like, and I really want to learn Rust.

And so I thought I would use this as an excuse to learn Rust.

I was going to hack on it for a little bit.

And then I asked a friend to introduce me to the creators of DuckDB, Honest and Mark, who were in the Netherlands.

And I had a call with him and he said, hey, it's a cool idea.

I'll introduce you to Honest and Mark, but you should also talk to my friend Tom.

Tom invested in my company and he'll give you some feedback on the idea." I'm like, sure.

Then the next morning I talked to Hannes and Mark and they're like, I was actually asking for a job.

I was like, hey, would you hire me to help you build a SaaS service?

And I said, no, we don't really want to do that.

We want to just focus on the core database.

and that afternoon I talked to Tom who was Tom Tungas.

He was the managing director at Redpoint

and now has his own has his own firm

and I get about 15 minutes in he's like yeah it's a good idea I want to fund it.

Something like that.

Come to the partner meeting next week

but you know it's sort of a foregone conclusion

because he was you know the managing director

and then the next day I was talking to some other VCs about something totally different.

I thought I was wasting their time.

I'm like, hey, I had this idea kicking around.

Do you guys want to hear it?

And they said yes.

And it was Martine Cassato from A16Z and Elad Gil.

And Martine said, you just had the partner meeting.

How much do you want?

And I'm like, I don't know.

Five million dollars?

Like, I had no idea.

I didn't even know who these guys were.

I didn't know if this was a good firm or not.

I didn't know anything.

And, um, you know, they said, sure.

It sounds good.

Anyway, the long, long story short, like I kind of like stumbled into starting a company and,

um, and then, uh, you know, it was about a week later and I'm sort of like, like I sat down, I was like,

is this really what I want to do.

And, and then I just thought like, well, you know, I've like, of the options that I have in front of me,

like, you know, there's, this is the one where, you know, it's a bet on myself.

It's a bet on, I think it's a really good idea.

I'm not sure I'll ever have another idea that's this good.

So I might as well, you know, go for it.

And also I realized that, you know, I've worked at jobs that had great culture and great environment and ones that were horrible and toxic.

And it's like, well, how do you, how do you know that you're going to have a good place to work?

And I'm like, well, you know, if you create it yourself,

then, you know, at least you only have yourself to blame if it doesn't work out.

You know, it's not, you know, you can't, it's not always, you know, sunshine and roses and everybody is wonderful to each other.

But, you know, that's sort of, I figured like, it's the best way to actually build a good work environment

and to win by building a group of people that like each other,

that work hard because they're doing it for their co -workers who they trust

and respect and they want to see them succeed versus everybody out just for a payday.

So that's what we tried to do and that was about four years ago and you know I think we've been having

We're having a lot of fun ever since then.

Yeah, that's amazing.

The relationship of DuckDB and now the company

Mother Duck, can you go a little bit deeper into that?

Because I looked up kind of similar companies

and my mind went to kind of what I know for Vercel and kind of Next .js.

But that's a very different example from the setup you have currently.

Sure.

I think it's a really unique and interesting one.

And, you know, we basically have this commercial company, Mother Duck, and we're building this hosted cloud version of DuckDB.

And then we have the open source company, DuckDB Labs.

They just changed the name to Duck Labs.

And so the Duck Labs, they focus on building a great database, building a great open source database.

They have no pressure to do anything like change the license or do anything, remove things from the open source.

And we're the commercial company and we're trying to build a SaaS service out of it.

We're not selling DuckTB.

We're selling, you know, we run DuckTB for you and with serverless and AI, all this, you know, all this stuff,

all the things you build around it.

And, you know, when we started, we actually gave them a co -founder share of the company.

So like, because I didn't want to just, you know, say, hey, thanks for building this great open source tool.

We're going to take it and, you know, we're going to try to make a bunch of money out of it.

It was like, and we also realized

that the only way to really make it work was if they were bought into it.

And the way to have them bought into it is so that they are incentivized for us to succeed.

And I think that's been great

because it's very hard to build a SaaS service without the ability to make changes to the underlying project.

And so we had a lot of hooks that were put in just for us.

100

We have a lot of features, bug fixes.

101

We also contributed a lot ourselves.

102

I think we have a thousand commits or something into the DuckDB code base.

103

But so far, it's been a really nice partnership.

104

It's been a sort of high trust, strong relationship between us and the DuckDB founders.

105

And kind of the hope is that if this works out, then this can be a model for other open source to use.

106

If it doesn't work out, then it's a cautionary tale.

107

But I think even if like, I mean, the world changes, people change.

108

But, you know, so even if as things change, like we've gotten a really good base and a really good start with this,

109

you know, really key partnership for us.

110

Yeah, I love the word partnership.

111

In essence, that's what it is.

112

All open core companies that I've seen is either people build a company

113

and then they open source part of

114

that kind of to get a community around that

115

or they have the inverse where there are the people that maintain the open source component.

116

They also need to make money and then they say, OK, we're going to build services on top of that.

117

And some functionality is kind of gated.

118

But the core part, that's the open core is still open.

119

Tell me what you have.

120

I don't know of any other company that has kind of this partnership.

121

Because you also get these cases where, you know, people who are building, they build open source.

122

And, you know, you're trying to build a community and you're really kind of incentive to just sort of grow users.

123

And then they're like, well, how are we going to monetize?

124

OK, we're monetized with the SaaS service.

125

And so what they do is they build the like simplest possible SaaS service because it's not what they're focused on.

126

Their focus is on building this great open source project.

127

So the SaaS service is like, okay, we just run whatever in Kubernetes and they call it monetized.

128

Managed.

129

Yeah, they call it managed.

130

And I think there's a real opportunity to innovate in how the service is delivered

131

and innovate around that project that like

132

if your primary focus is building the open source project

133

you just don't you know your all your innovation tokens are being spent on on

134

that versus on on the delivery and

135

so for us the innovation tokens are spent on how do we provide

136

this amazing you know end -to -end service just to

137

so to make analytics delightful for people yeah i love

138

that a lot i'm wondering because your origin story was a lot of, okay, I reached out to these people,

139

I asked for help, and all of a sudden you said, okay, they asked me for a number and I gave them a number.

140

This vision that you have of this partnership

141

and me focusing my innovation tokens on what I think really should be the core of this company,

142

this vision, is that why people invested in the end

143

or helped you in the first place now that you've reflected on it?

144

Or why were people so willing to help?

145

I think there were a couple things.

146

One was Snowflake had just gone public and I think it was the largest IPO at the time.

147

And I think now it seems kind of quaint with all the, you know, you know, Anthropic and, you know, and open, open AI like and.

148

But at the time, like, you know, everybody's like, hey, if they miss out on Snowflake, then then they wanted to find the next thing.

149

I think, you know, DuckDB was an open source project that had a lot of traction and so it was growing really quickly.

150

And so there was a bet on that.

151

You know, I also helped create Google BigQuery and I led engineering and product for a while.

152

And so I had a decent amount of credibility in that.

153

And so it was like, okay, if somebody is going to build something like this, DuckDB is the right thing to build it on.

154

You know, I have a pretty good track record, somebody else to bet on.

155

And I think those things just sort of like sort of came together and in a nice, nice package.

156

Is that also what you looked for

157

when establishing kind of the first core team of people that were going to found this company?

158

So the history plus the enthusiasm of Duck TV?

159

Yes.

160

The founding team was like was people that I'd worked with before, customers that I'd had before.

161

and two of them were like people that I knew as customers.

162

A couple were people that I worked with at Google.

163

And then people who were just DuckDB enthusiasts that were like,

164

hey, I want to be part of this DuckDB thing and this is sort of the way to do it.

165

And we had kind of a big founding team, but that really let us hit the ground running and get a lot of stuff done.

166

And also just like we had some quite senior people

167

And the nice thing about

168

that was it allowed us to like take sort of the obvious

169

but wrong designs and just skip past those and get to like, you know, an actual better design

170

and the right way to do things um you know at least

171

sometimes the senior people they're less like you know the they can be less like sort of like scrappy and like

172

but also um you know you just get more done per kind of per like line of code

173

or per per hour of uh of uh of of building

174

and so that allowed us to have you know i think a quite a good solid foundation

175

that we're you know continuing to build how big was initial team?

176

It was 11 people.

177

Yeah.

178

And when was the first milestone that you've delivered something with that team specifically,

179

or maybe even with a broader team that you were quite proud of that?

180

So we had like our first alpha, you know, kind of showed, you know, this is a hosted DuckDB that, you know, serverless, so you could,

181

we could, you know, assign instances and, you know, in milliseconds.

182

And I think that took us about three, three to four months to have something that,

183

you know, our own storage system and it was, you know, strung together with some bailing wire,

184

but, you know, it was enough to sort of show to some people and start to get some early design partners.

185

It took us about a year to get to beta, to something that we thought this is like, this is good enough that,

186

you know, people will actually, you know, want to use this for real.

187

And then it took about another year to get to the point where we're like, well, okay, this is something that people will pay for.

188

And I think one of the tricky things was just, you know, the state of the art of data warehouses is,

189

you know, is quite, you know, quite far these days.

190

And so just the amount of table stakes that you have to do before people are like,

191

yes, I will actually use this for my analytics were pretty significant.

192

But since then, we sort of have been off to the races on the monetization

193

and actually building a business out of it.

194

Nice.

195

Was there a capability that stands out more so than others?

196

So either developer experience or some of the technical components that needed to be in place for the product?

197

Or what's your view on that?

198

I think some of it is just the reliability and the usability.

199

Like, if you're going to use this in production, it's got to actually work.

200

It can't lose data.

201

It's got to return the right results.

202

You know, I think we had, you know, and I think we were also growing up with DuckDB.

203

So like, you know, I think DuckDB, when we started, it could do a lot of things really well,

204

but it was not fully fleshed out there was a lot of ways you could not get over

205

and queries that went off into the weeds and you know

206

um you know this was i think duck db 1 .3 or 0 .6 so it was uh it was you know, still quite early.

207

And so just hardening DuckDB was also an important part of that.

208

Yeah.

209

What's the latest version, just to get an understanding of how farts come?

210

1 .55, but like there's,

211

you know, they were doing releases kind of every couple of months in the early days.

212

I think they've slowed down a little bit, just as the, you know, the bar get tired for testing and you want to make sure that,

213

you know, everything is solid.

214

It's not a, hey, we just throw it over the wall and like, and, you know, and people will tell us what's broken.

215

Yeah, gotcha.

216

What's the next big problem that you're trying to solve that will really bring value to the users in the end?

217

So, you know, in the last six months or so, it's kind of been clear that AI is changing how people are interacting with data.

218

And we're sort of assembling the adjacent pieces to the data warehouse and adding this functionality to our product.

219

The first one was, after we realized that, hey, Claude is really good at building data visualizations.

220

And so what if we just hosted those visualizations, swapped some things out to make them live and querying live

221

Mother Duck databases, had them shareable in Mother Duck, and then, you know, voila, you have basically BI.

222

It's BI, but it's BI on steroids because there's some things you can do that you can't, the level of interactivity,

223

the sort of the level of creativity

224

you can do kind of is way beyond what you can do in a BI tool.

225

And then if you look at the other sides, the other pieces of the data stack, you know, there's, okay, you have to get data in,

226

you have to data ingestion.

227

which is also pretty vibe -codable.

228

Like if you point Claude at Salesforce and say, hey, pull my data in, it's going to be able to figure that out.

229

And so just like we did with the visualization tool, which we call dives, we built flights,

230

which is a way of allowing people to vibe -code a pipeline that they can get data in.

231

And we did it in a very generic way so that these pipelines called flights are just

232

Python code running on a schedule.

233

But really, you combine that with Claude and you can do some really, really cool things.

234

And so you kind of put that together.

235

You have the ingest, analyze with our core data warehouse offering, and then visualize with our dives.

236

And then you can like build this sort of feedback loop where you can have a visualization

237

that like will cause data to be to be rehydrated or views to be materialized or sort of kick off,

238

you know, right, right backs and kicks off a flight.

239

Or maybe the maybe the flight is actually sending, you know, a, you know, Slack messages to somebody or maybe it's building a PDF.

240

and you know meanwhile you can have flights

241

that are sort of building building dives

242

that are updating you know with with the latest you know days news or you know running inference

243

you kind of put these you know each one of these tools by itself is like is is very interesting

244

and then I think the the feedback loops are also where it gets very interesting underneath

245

we have we call guides which is just a context layer

246

that kind of threads through everything and so in the

247

And the last piece that we're working on is sort of just the agent harness, which is, right now you interact with it through Claude,

248

ChatGPT, Gemini, or whatever, and we're sort of building our own agent interface on it,

249

which will just make it easy to, like when somebody shows up, and you just, instead of like saying,

250

okay, well, I'm going to run these jobs or these queries or to build these things, and just, you have a discussion with

251

our agent interface that you describe what you want to do

252

and we figure out how to do that by combining these pieces together.

253

And I think that's sort of the, I mean, to me, that's sort of the holy grail of sort of AI for data interfaces where,

254

you know, you might not be super, you know, you might not be a data engineer or a data analyst.

255

You might be.

256

It might just be like, you know, it's, you know, just like a software engineer,

257

you know, it's really nice to have Claude write the code for you.

258

And like, yeah, you could write that, you know, you know, command line, you know, interface code or...

259

or UI code, but it's just, you know, it's, you can do it a lot faster if you have, if you have cloud to do it.

260

And, and so, you know, I think it's going to enable both technical people

261

and non -technical people just to get a lot more stuff done with data.

262

And also to let people, you know, business users who are actually close to like what's actually happening to be able to,

263

you know, get some of this stuff done on your own,

264

like the marketing person that wants to understand what's going on with their campaigns that they're running or the salesperson.

265

A lot of our salespeople are probably some of the biggest users of MotherDuck internally.

266

They use it to find out, hey, what's going on with my customers?

267

What's going on with my book?

268

Which customers should I talk to?

269

How has their usage been changing?

270

And on one hand, everybody's nervous about what happens

271

when this AI goes off and does something weird or does the wrong thing and hallucinates.

272

But I think one of the cool things is, you know, by having the business users do it,

273

they're the ones that can actually spot when it's subtly wrong.

274

Whereas if it's an analyst and kind of this, you know, separate team and you say, you know, somebody files a ticket and says,

275

hey, can you build this dashboard for me?

276

if the numbers like turn out they look kind of wonky um the analyst isn't going to know until like

277

and until like the business users like hey like i know

278

this customer doesn't actually use this feature why does this show up here

279

and like you know and they can they can start poking at it

280

and there's just a long feedback loop whereas i think

281

if you have um you know the business users close to what's actually happening

282

they can sort of spot these these problems and and and fix them

283

And at the end of the day, you know, humans are fallible as well.

284

And so I think, you know, just like self -driving cars, at some point the self -driving cars get, you know, better than, yes, they are.

285

They do make mistakes, but, you know, they're not going to be perfect, but it's, you know, they can certainly be better than a human.

286

Yeah.

287

Having this agent operator kind of in between the tools

288

and capabilities that you offer and the user that actually wants to get these insights,

289

I feel like allows the flexibility to kind of focus on different types of users.

290

But as you say, an analyst has a different kind of reference frame

291

and different judgment or even taste compared to a business owner or a salesperson that knows their process in and out.

292

Are those then more the primary users as you view it or how do you see that?

293

You know, I think that like, it's going to be, it's going to be both.

294

It's going to be, it's going to be sort of like,

295

there's going to be, somebody's going to be responsible for data at a company and it's, you know, it's not going to be the business users

296

and there's going to be a data team

297

and they're going to be responsible for making sure the pipelines are working and debugging things and stuff stays, you know, the weekly business reviews can,

298

can, can proceed with, you know, with, with real data.

299

And when you report data to the board, it's actually uh it's actually you know reliable

300

but i also think it's really important that you know questions

301

that other people have in the in the business don't have

302

to go through like gatekeepers uh uh who can who can figure things out

303

because i think once these folks can start to ask questions about their the things

304

that they're interested in you notice that like the things that they ask are actually different.

305

You know, for instance, the, you know, one of our salespeople, the first time we kind of set him loose on our MCP server,

306

you know, his first question that he asked was like, what customers should I talk to today?

307

That's not the kind of question that like, that an analyst would ask.

308

And, but it's something that Claude actually did a pretty good job.

309

You know, it can look at support tickets.

310

It can look at, you know, usage.

311

It can look at, you know, news about like, hey, this company, you know, just got dinged by the FCC or something like that.

312

There's all sorts of, you know, interesting things that Claude can pull together.

313

Also, our head of sales, first question he asked was, what are the three biggest risks to my business?

314

Oh, yeah.

315

And like, I mean, those are things that Claude actually can give a good answer to,

316

but also not the kind of thing that sort of like a data team is likely going to,

317

you know, they're not going to come up with this report and then tell the, you know, tell the CROs like,

318

hey, here's the risks of your business.

319

Like, what the hell is this?

320

But, you know, I think there's different kinds of questions

321

that you can start to ask when different types of people are doing asking.

322

Yeah, I love that.

323

I was talking to a person from booking .com and he's responsible for developer experience.

324

And they also mentioned that now with agents

325

and them already being quite advanced and starting early

326

and capturing all the data of developer experience on a quantitative level instead of building dashboards

327

and metrics and KPIs because this person said if we do

328

that then those become the targets so then it kind of defeats the purpose

329

They indeed give their users, which could be engineering managers or leadership teams, the capability to kind of debug the data, debug the software development lifecycle,

330

find bottlenecks and then solve those and then ask more questions and find more bottlenecks in that way.

331

And I feel like this interaction of debugging is one of the phrases, but kind of finding gold in your data together with agents,

332

because in that way, you can also have capabilities that you couldn't even imagine, where Claude goes off on the web and pulls that data in

333

or looks at some artifacts that it already has next to your data and then makes a decision based on that.

334

I think it's fascinating.

335

And I hope it's kind of a trend that will continue

336

and people are given the tools to kind of solve the problems that they are top of mind of.

337

We've had this internal debate a bit about the future of the dashboard.

338

And I think there's, you know, I wrote an article a couple years ago called Big Data is Dead.

339

And actually, I borrowed the title from that is like, is because people had been writing about dashboards are dead, you know, for a decade.

340

And so, you know, he was like, oh, well, you know, with agents, you know, we don't actually need these dashboards.

341

And, you know, people are, you know, if you have dashboards, then, you know, people are,

342

you know, they're trying to beat that as a KPI versus is actually solving real problems.

343

On the other hand, I'm a big believer in dashboards because if I just think about how I, you know, I use AI with our products,

344

you know, all over and I use, you know, I do, you know, but I build a lot of dashboards, a lot of data visualizations,

345

and I also have a bunch of data visual, you know, dashboards that I look at, I don't know, eight times a day.

346

You know, I'm like, you know, where's the number?

347

What's the usage?

348

Who's using it?

349

How is it changing?

350

I think what the difference is, is that kind of there was this interactive BI movement where you build a dashboard

351

and you can like drill down and click on things.

352

And I think one of the problems is like the type of questions you can answer in

353

that kind of situation are very, very narrow.

354

It's like, okay, well, I can split out this aggregation.

355

I can see where something is actually coming from.

356

But I can't go a step further.

357

I can't ask why, I can't see what actually is happening.

358

I can answer the question maybe that had been prepared.

359

I think the interesting thing with AI is that like you can have that dashboard and you can be like, that number there is weird.

360

Like something is happening today.

361

What's going on?

362

And then you can, you know, you can have, you know, you can ask Claude and Claude will,

363

you know, poke at it and figure it out.

364

And at some point, you know, there'll be an agent running and like, and the agent will tell you there's something weird in the first place

365

but it's also just sort of very nice to have a kind of this just sort of stable thing

366

that you know this is what we report on this is

367

what we care about this is like these are the these

368

are the you know input metrics to our revenue metric and kind of we

369

have a sense of how like we have a mental model

370

or of how the business works through through these metrics uh

371

and that you know when something changes they should like they should change for a reason

372

and we should be able to dig in and understand.

373

I want to pivot slightly and from the perspective of the listener who's typically an individual contributor or an engineer,

374

if they get really excited about this concept of databases on the one hand, infrastructure on the other hand,

375

even the flavor of AI and kind of the user experience that builds on top of that,

376

they might want to join a company that is yours or a company that is similar.

377

What is your advice for people that want to get into specifically this niche of engineering?

378

Well, we're hiring.

379

Yeah, that's good.

380

Okay, check out the website.

381

You know, I think it's one of the just interesting things about a database company like ours

382

and a database SaaS company or database infrastructure company is that there's a whole gamut of engineering jobs from storage,

383

database, infrastructure, SaaS, or like the API level,

384

and then UI, and then AI and agent stuff as well.

385

And it takes all of those things.

386

And so kind of people tend to think about it as like, oh, well, I want to work on the optimizer.

387

I think everybody wants to work on the optimizer.

388

And it's like, first of all, there's not many people in the world that work on database optimizers.

389

And second of all, it's not that interesting of a job.

390

Like it's, you know, kind of you're working in this box

391

and you kind of know where the inputs and you know the outputs.

392

And I mean, for me, it's much more interesting to figure out how do we take kind of these things that are...

393

much more unknown and much more ambiguous

394

and figure out how to like make those all make those all work together

395

and make those work in a in a harmonious way so

396

you know I think the you know the important

397

you know a lot of the important jobs in a you

398

know in a database company are not just the like the core the core database stuff yes

399

that is that's obviously very important and like

400

and at the end of the day you know

401

if you're a database company or an infrastructure company you kind of live

402

or die based on the strength of the you know of the of the the core technology um

403

but if you don't have the great stuff around it then

404

then you know it's you're not going to be successful

405

and um you know i i i think

406

there's a lot of people who um only look at the

407

sort of the the the inner the inner loop uh and not the outer.

408

And I think that's just a, it's also a recipe as a, as an engineer to be sort of, um, you know, have, have blinders about,

409

um, what do users actually want?

410

And like, how do you, how do you build something that's actually, that's actually compelling?

411

Because, you know, in my career, I worked on technology that I thought was like, that I was, it was beautiful.

412

Like I was like, I was so proud of this and it was like, and like, and my heart and my soul went into it

413

and it's like but then it doesn't get used and then

414

or the project gets cancelled or the company

415

that it was built you know behind it goes out of business

416

and like and it's it may as well not have existed and and

417

so that's why I think

418

it's important to um uh to sort of pay attention to the other pieces around it is the the users

419

and you know and the the things

420

that we tend to call product I'm curious what your advice is for people to distinguish themselves

421

because we need people that are both kind of enthusiastic about the technical capabilities

422

that and I also see this trend go a little bit more in width

423

and also think about users usability how do we make this stake

424

and what is important from a user persona perspective there's a lot

425

that goes into it but nowadays with AI potential resumes look the same actually all

426

that they all fulfill your hiring profile basically yet you still need to have people on board

427

that are dedicated to the mission that can execute as you would want them to, how do they distinguish themselves in the first place?

428

I think that is a good question.

429

I think understanding product, which means understanding users, how the thing is going to be used,

430

and understanding the business context in which something lives,

431

I think is a superpower for an engineer.

432

Engineers tend to look at down on it, but it's, you know, again, if you wanted to actually turn into something,

433

something that people care about, I think you have to,

434

you have to understand the users and have a lot of respect for them and a lot of like, you know, empathy, empathy with them.

435

And you also have to, you know, it's got, at the end of the day, it's got to be a business that works because,

436

you know, if, if you can't make money off of it, then everybody goes home and,

437

and has to do, do something else unless you're going to, you know, just, um, you know, just work on open source.

438

Uh, but at some point somebody has to pay, somebody has to pay the, somebody has to pay the bills.

439

Um, and I don't think it's, it's not that hard to learn product.

440

You just have to sort of open, you know, open your mind and think about things, um, you know, think about some of these things.

441

I think for me, one of the best ways of doing that was I wrote the first book.

442

I co -wrote, co -authored a book about BigQuery when I was working on BigQuery.

443

And the funny thing was, as I'm describing how to use it, I realized like, hey, that's weird.

444

That's wonky or that's goofy.

445

Like we should change this.

446

So like there were a lot of changes in BigQuery that were the result of, you know, as I'm just as I'm writing this book,

447

like, how do you make it actually so that it makes sense, like the writing makes sense.

448

And there weren't necessarily huge changes, but they were just like, they were, you know,

449

sometimes the command line that you do to do something or the API you do something to do something was like, it just felt goofy.

450

It's like, that's, that's goofy and like and so you you fix it and

451

so like um you know writing things down helps like i think

452

you know whether you know i think

453

that's probably to me in my mind the biggest uh one

454

of the most important skills for for an engineer is to be able to write well um

455

and i was always a terrible writer and i hated writing and i was super slow at it and i

456

but I forced myself to do it.

457

And, you know, and if you force yourself to do it, you get, you get, you get better at it over time.

458

Um, I still hate it.

459

I still like every time I have to write something, I like, I like, I'll, you know, procrastinate.

460

And then I finally start, I write a paragraph and then I get up,

461

you know, and walk around the house and go do something else and eat some gummy bears and then come back.

462

And I like, I'm like, I'm going to write, I'm going to finish this page.

463

It's just like this, that's this, like coding is never like

464

that coding is like this is great

465

i'm gonna just stay in you know work on this for you know for for hours

466

and hours and um

467

but writing like you know it just i i never got into the to

468

that flow state but i do think um

469

if there's like one skill to learn

470

and that's i think it would be it'd be writing

471

and it's also hard in the day in this day

472

and age where people you can like you know send the you know four bullet points to claude

473

and then claude will come up with this you know entirely plausible 30 page document

474

but I think that if you do

475

that then you're doing you're doing yourself a disservice

476

you're also doing the people who would be reading it a disservice because because really like the the amount of

477

actual content in there is just the four bullet points

478

that you have

479

and somehow it blew up to like 30 pages yeah you have to add value on top

480

and sometimes that value is reducing all of the load

481

that is like generated throughout there from what I know from my friends in both open source and open core companies

482

what people contribute towards that open source is for them kind of a hiring funnel.

483

But I don't know if that's the same for you.

484

Have people that contributed to DuckDB in the end also joined Mother Duck?

485

Or how do you see that flywheel?

486

We had a couple of people who were, you know, in the early days who were contributors to DuckDB.

487

But that's actually one of the tricky things with the model that we have is because,

488

like, you know, we really don't want to poach people from DuckDB Labs because of the relationship we have with that company.

489

And so a lot of the DuckDB contributors go to DuckDB Labs.

490

We actually have a decent number of students.

491

So students who have built DuckDB extensions

492

and sort of worked on stuff on DuckDB that we will hire them at Mother Duck.

493

And we have a handful of interns from CWI in the Netherlands,

494

which is like this research institution that spun out DuckDB Labs and spun out and also created

495

Python and some other things as well.

496

But that relationship has been a fruitful one

497

because kind of then we can get people who are just excited about the area, but also have some experience.

498

One of the last thoughts I had is mainly a branding question

499

because the style

500

that you're wearing the product names everything from doc tb already is what it started with mother the doc

501

you mentioned even some of the components it's not pipeline it's flights

502

and dives and guides some of the other product names

503

that are there who came up with that

504

because that's genius um it was a it was a group effort uh

505

but i think like you know if you have a

506

if you have a company that's uh

507

you know based on something with an animal in the name um uh you you have two choices you can either,

508

you know, be like coy and like, yeah, I know it's, I know it's duck in the name, but like, we're too cool for that.

509

But, um, uh, and, uh, you know, we're too classy or you can embrace it and, and, and really lean in.

510

And so, you know, we decided to, to, to, to lean into it.

511

Um, because at the end of the day as a startup, you know, um, there's a war for attention and,

512

you know, just like, like I mentioned before, like, cause if, if you build something that nobody uses,

513

you might as well, it might as well have not, not been built.

514

And, and the way to like, get attention is, you know, you know, with, with interesting,

515

interesting, interesting branding, and, you know, it takes just, you know, being willing to look a little silly or feel a little silly.

516

But, you know, it's, it ends up being, it ends up being memorable.

517

And, and it's also can be self -sustaining is sort of once you sort of like take that step, then it's like, oh, well,

518

we wear the, you know, we wear the goofy, the goofy stuff.

519

And, you know, the people that join the company are people that are willing to do that and, you know, kind of can,

520

can laugh at themselves or, you know, have enough confidence that they can, you know, you know, they can rock the duck,

521

duck swoveralls or the duck, you know, the weird duck hats and stuff like that.

522

And, And, you know, it's, it, so it's, I wouldn't call it a gimmick because it's,

523

you know, it's, Part of our culture is just being able to like to laugh at ourselves and have fun.

524

You know, it's also a way of just sort of signaling that, you know, hey, we're not this like stodgy enterprise brand that everybody,

525

you know, like databases seem so serious and like, and, you know, the competitors seem like seem really serious and it's sort of like,

526

you know, hey, we can win by, we can win and we can have fun and laugh and, and, you know, and that's all,

527

it can be even the secret to how we, how we can be successful.

528

Yeah, absolutely.

529

Jordan, thanks so much for coming on and sharing.

530

Yeah, thank you.

531

It's been fun.

532

That's it.

アプリをダウンロード

流暢に話すために必要なすべて

AI発音採点文ごとに採点

IPA練習すべての音をマスター

単語学習自分の単語帳を作る

単語ゲーム遊びながら覚える

QRをスキャン

Download on theApp Store Get it onGoogle Play

同じチャンネル

Why Coding Agents Keep Making Your Codebase Worse

Daily ConversationWorking Adults

How Great Software Engineers Are Moving Faster Than Ever

Business EnglishInterview EnglishUniversity Students

FOR YOU

このレッスンの語彙とスピーキングのポイント

このスピーキングレッスンは、動画「How an Ex-Googler Accidentally Founded a Database Company」を教材にしています。この動画には、シャドーイング用の文が355文、単語が7497語あります。音声の長さは40:11です。話す速さは速く、1分あたり約187語です。音のつながりや弱く発音される音が多くなります。単語の87%は英語の頻出3,000語に含まれるため、語彙は分かりやすいです。

この動画の重要語彙

動画に出てくる覚えておきたい単語15語を、発音と意味つきで紹介します。

単語	発音	意味
duck 名詞	/dʌk/	鴨, カモ
database 名詞	/ˈdeɪtəˌbeɪs/	データベース
solve 動詞	/sɒlv/	解決する
partnership 名詞	/ˈpɑɹtnɚʃɪp/	パートナーシップ, 協力関係
hire 動詞	/ˈhaɪɹ/	借りる
capability 名詞	/ˌkeɪ.pəˈbɪl.ɪ.ti/	能力
goofy 形容詞	/ˈɡufi/	すっとこどっこい
infrastructure 名詞	/ˈɪnfɹəˌstɹʌkt͡ʃə/	インフラストラクチャ, インフラ
feedback 名詞	/ˈfiːdˌbæk/	フィードバック, 反応
pipeline 名詞	/ˈpaɪpˌlaɪn/	パイプライン
contribute 動詞	/kənˈtɹɪb.jut/	貢献する
component 名詞	/kəmˈpoʊ.nənt/	部分
innovation 名詞	/ˌɪn.əˈveɪ.ʃən/	イノベーション, 革新
versus	/ˈvɝsəs/	対, バーサス
loop 名詞	/luːp/	ループ

動画に出てくる句動詞

単語	発音	意味
show up 動詞		現れる
slow down 動詞		遅らせる
check out 動詞		試験する, 検査する
come back 動詞		戻る, 帰る
find out 動詞		見出す, 調べる
get up 動詞		起きる
look down on 動詞		見下す, 軽んじる
look up 動詞	/ˌlʊk ˈʌp/	見上げる

繰り返し練習したい文

動画に出てくる短く完結した文です。日常会話でそのまま使えます。

Do you guys want to hear it?
How much do you want?
Can you go a little bit deeper into that?
What's going on with my book?
Which customers should I talk to?

この動画の文法

話し手がよく使っている文型を、動画の実際の表現とともに紹介します。

文型	動画での表現
現在完了進行形 have/has been + -ing — 以前に始まり今も続いている動作	I've been thinking · we've been having
現在完了形 have/has + 過去分詞 — 過去の出来事が今も関係している	have not been built · I've built · I've worked
受動態 be + 過去分詞 — 誰がするかより、何が起きるかに焦点を当てる	have not been built · were bought · are incentivized
関係詞節 who / which + 節 — 人や物について情報を加える	people who are · people who were · guides which is

注意したい発音

話し手はI'm, we're, you'reなど、短縮形や弱形を123回使っています。聞こえたとおりの短い形で発音しましょう。

「th」の音: empathy /ˈɛmpəθi/, thread /ˈθɹɛd/, underneath /ˌʌndɚˈniθ/, enthusiasm /ɪnˈθuziæzəm/, width /ˈwɪtθ/
「sh」と「zh」の音: partnership /ˈpɑɹtnɚʃɪp/, innovation /ˌɪn.əˈveɪ.ʃən/, distinguish /dɪˈstɪŋ.ɡwɪʃ/, establish /ɪˈstæb.lɪʃ/, essentially /ɪˈsɛnʃəli/
長い単語（アクセントの位置に注意）: capability /ˌkeɪ.pəˈbɪl.ɪ.ti/, infrastructure /ˈɪnfɹəˌstɹʌkt͡ʃə/, innovation /ˌɪn.əˈveɪ.ʃən/, developer /dɪˈvɛləpɚ/, contributor /kənˈtɹɪb.jə.tɚ/

日本語話者が苦手な音：

/r/ と /l/ の区別 — /l/ は舌先を歯茎につけ、/r/ はどこにもつけない: necessarily /ˌnɛs.əˈsɛɹ.ə.li/, horrible /ˈhɔɹ.ɪ.bəl/, reflect /ɹɪˈflɛkt/, reliable /ɹɪˈlaɪəbl̩/, formerly /ˈfɔɹmɚli/
/v/ — /b/ にならないように、上の歯を下唇に当てる: solve /sɒlv/, innovation /ˌɪn.əˈveɪ.ʃən/, versus /ˈvɝsəs/, developer /dɪˈvɛləpɚ/, dive /daɪv/
/f/ — 「フ」ではなく、上の歯と下唇で出す: goofy /ˈɡufi/, infrastructure /ˈɪnfɹəˌstɹʌkt͡ʃə/, feedback /ˈfiːdˌbæk/, interface /ˈɪn(t)ɚˌfeɪs/, founder /ˈfaʊ̯n.dəː/

この動画での練習方法

まず声を出さずに動画を最後まで聞き、知らない単語をメモします。
まず0.75倍速で一文ずつシャドーイングし、慣れてきたら通常の速度に戻します。
自分の声を録音して元の音声と比べます。duck, database, solveなどの単語に特に注意しましょう。

シャドーイングとは？英語上達に効果的な理由

シャドーイング（Shadowing）は、もともとプロの通訳者養成プログラムで開発された言語学習法で、多言語習得者として知られるDr. Alexander Arguelles によって広く普及されました。方法はシンプルですが非常に効果的：ネイティブスピーカーの英語を聞きながら、1〜2秒の遅延で声に出してすぐに繰り返す——まるで「影（shadow）」のように話者を追いかけます。文法ドリルや受動的なリスニングと異なり、シャドーイングは脳と口の筋肉が同時にリアルタイムで英語を処理・再現することを強制します。研究により、発音精度、抑揚、リズム、連音、リスニング力、そして会話の流暢さが大幅に向上することが確認されています。IELTSスピーキング対策や自然な英語コミュニケーションを目指す方に特におすすめです。

シャドーイングのやり方: ステップ別の完全ガイドを読む →

シャドーイング練習: How an Ex-Googler Accidentally Founded a Database Company - 動画で英語スピーキングを学ぶ

練習モードを選ぶ

アプリをダウンロード

同じチャンネル

Why Coding Agents Keep Making Your Codebase Worse

How Great Software Engineers Are Moving Faster Than Ever

おすすめ動画

The Only SaaS Moats Left in 2026

PostgreSQL in 100 Seconds

How I deploy serverless containers for free

If you’re an aspiring product designer, watch this (roadmap included)

Did Google just dethrone OpenAI and Anthropic?

Go’s Big Mistake

System Design Was HARD - Until You Knew the Trade-Offs

Mark Zuckerberg: “Alex, are you sure this is going to work?”

Jensen Huang: The Mindset That Built NVIDIA

How To Land Your First Junior Dev Role

Something Weird Happens When You See How Connected Everything Is

My 4.9-Star iOS App Is Actually a Website: Here’s How I Built It

このレッスンの語彙とスピーキングのポイント

この動画の重要語彙

動画に出てくる句動詞

繰り返し練習したい文

この動画の文法

注意したい発音

この動画での練習方法

シャドーイングとは？英語上達に効果的な理由