Sep 14, 2023 · 1h 4m · in-depth
AI Hot Takes and Unusual Twitter Fundraising Strategies with Dan Siroker (Rewind AI)
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this episode of First Round Capital's In Depth podcast, Rewind AI co-founder and CEO Dan Siroker discusses the architecture behind his memory augmentation platform, frameworks for engineering velocity, practical predictions for generative AI, and his unconventional public fundraising round on Twitter.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Brett holds 0.7% of the talking time here. How this is scored →
Brett as informed peer 4.0 Guest teaching 5.8 Guest disagreement 1.3 Brett pushing back 0.3
speaking balance: gold is Brett, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 48:00 Dan rejects AI doomerism and immediate superintelligence
Dan forcefully pushes back on the common tech industry belief that AGI will instantly trigger runaway superintelligence and societal collapse, calling it a logical fallacy.
Hardest push from Brett ▶ 32:01 Todd presses Dan on narrowing target customer segmentsTodd challenges Dan on maintaining a broad target market, pressing him to specify concentric customer segments beyond broad hardware categories like M1 Mac users.
Biggest teaching moment ▶ 57:25 Dan's unconventional tactic of practicing on VC associatesDan offers tactical, contrarian fundraising advice by breaking down how to exploit junior VC associates at secondary firms to test and refine pitch messaging for lead partners.
Brett holds their own ▶ 43:15 Todd compares AI privacy concerns with early Google/Facebook erasTodd brings his domain expertise as an early executive at Google and Facebook to analyze long-term consumer willingness to trade privacy for utility.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | Brett as informed peer | Guest teaching | Guest disagreement | Brett pushing back | Why |
|---|---|---|---|---|---|---|
| Episode Overview: Introducing Dan Siroker and Rewind | 4 | 5 | 1 | 0 | Todd opens the interview by framing Rewind's unique positioning in the AI ecosystem and invites Dan to share the origin story. Dan explains how his personal experience with hearing loss inspired him to explore technology for human memory augmentation. | |
| How Rewind Works: Personalized AI and Local Indexing | 3 | 5 | 1 | 0 | Todd asks technical questions about Rewind's mechanics and alternative solutions. Dan outlines local indexing on Apple Silicon and reframes human note-taking as an outdated workaround for imperfect memory. | |
| The Predecessor Product: Pivoting from Scribe to Rewind | 4 | 6 | 1 | 1 | Todd probes into the technical complexity and scope management of building the first version. Dan details the pivot from Scribe and explains how ruthless scope cutting was required to maintain imperceptible system performance. | |
| Strategic Decision-Making: Navigating Internal Disagreement and Instinct | 4 | 6 | 2 | 1 | Todd asks how Dan made the strategic call to pivot when the team was split. Dan shares how he favored intuition over consensus despite three out of five team members voting against the idea. | |
| Early Onboarding: Hands-On Testing and Rolling Waitlists | 3 | 6 | 1 | 0 | Todd asks about managing early user demand and unexpected user feedback. Dan explains manual onboarding workflows and admitting an early mistake regarding cloud transcription before switching to local Whisper. | |
| Prioritization Framework: Impact Divided by Effort | 4 | 6 | 1 | 0 | Todd asks how Dan balances user requests versus product vision. Dan shares his prioritization formula of impact divided by effort and how to handle customer problem identification. | |
| Engineering Velocity and Cultural Values around Shipping | 4 | 5 | 1 | 0 | Todd inquires about instilling rapid shipping culture across the engineering team. Dan details specific onboarding practices, such as requiring new engineers to ship code to production within their first 24 hours. | |
| Second-Time Founder Lessons: Focus, Decisiveness, and Team Size | 4 | 6 | 1 | 0 | Todd invites reflection on the differences between Dan's first company, Optimizely, and Rewind. Dan reflects on past mistakes of chasing shiny objects and highlights the power of lean, highly experienced teams. | |
| Psychological Pressures on Repeat Founders | 4 | 6 | 1 | 0 | Todd explores the psychological pressures of repeat founders. Dan discusses overcoming perfection paralysis and explains how he operationalizes company values and 360 reviews. | |
| Defining Product-Market Fit: Painkillers vs. Vitamins | 4 | 6 | 1 | 0 | Todd asks about defining product-market fit benchmarks. Dan clarifies that true PMF is defined by a passionate core segment viewing the product as a painkiller rather than meeting arbitrary broad survey thresholds. | |
| Operationalizing PMF: The Post-Meeting Email Summary Feature | 5 | 5 | 1 | 2 | Todd pushes Dan on narrowing target customer profiles using concentric circles. Dan candidly admits feeling like an imposter regarding audience focus and reflects on how failure to pick a specific persona previously caused tension at Optimizely. | |
| Metrics for Validating Product-Market Fit | 4 | 6 | 2 | 0 | Todd asks for key metrics and resources that validate PMF. Dan outlines cohort-based retention curves over vanity ARR and critiques hype-driven AI demos that fail to retain users over 8 weeks. | |
| Rewind's Moat: Proprietary Local Data and High Switching Costs | 4 | 6 | 2 | 0 | Todd asks what differentiates thin wrappers from defensible AI products. Dan highlights Rewind's moat of proprietary local data, high switching costs, and strong user lock-in. | |
| The Microprocessor Moment for Vertical AI Applications | 5 | 6 | 1 | 0 | Todd compares current data privacy concerns to consumer sentiment shifts from their early days at Google and Facebook. Dan draws historical parallels with location sharing to show that compelling utility rapidly outweighs privacy fears. | |
| Future AI Outlook: Labor Transformation, AGI, and Superintelligence | 4 | 7 | 3 | 0 | Todd asks about the 10-year outlook for AI and labor. Dan provides contrarian takes, predicting 50% of white-collar jobs will be transformed while rejecting the notion that AGI will immediately cause runaway superintelligence. | |
| Practical AGI Applications: Automated Software Engineering | 4 | 6 | 2 | 1 | Todd shifts the conversation to Dan's unconventional Twitter fundraising campaign. Dan explains why building user trust through radical transparency motivated him to publicly publish his unredacted investor deck. | |
| The Mechanics of Public Fundraising: Video Pitches and Roll-Up Vehicles | 4 | 7 | 2 | 0 | Todd asks for tactical mechanics on running a public fundraise. Dan outlines using Typeform, AngelList RUVs for accredited investors, and cheekily advises using junior VC associates for pitch practice. | |
| Fundraising Retrospective: What Dan Would Do Differently | 4 | 5 | 1 | 0 | Todd asks what Dan would change about the fundraising process and what is next on Rewind's product roadmap. Dan discusses avoiding lengthy auctions and hints at expanding into hardware. | |
| Mentorship Lessons from Paul Graham: Focus on the Customer Problem | 4 | 6 | 1 | 0 | Todd asks about Dan's most influential mentors. Dan reflects on Paul Graham's lesson to solve real user problems over building complex tech for its own sake, contrasting YC's philosophy with Google's. | |
| Recommended Reading: 'The Mom Test' and Episode Conclusion | 3 | 5 | 0 | 0 | Todd asks for book recommendations. Dan recommends 'The Mom Test' for customer discovery before Todd wraps up the episode. |
Statements from this episode (35)
Jackson: Rewind AI raised over $33M from a16z, NEA, and Sam Altman
“You've raised over thirty three million dollars, I believe at this point from great folks, Andreessen Horowitz, NEA, Sam Altman, and of course us here at First Round.”
Siroker: Rewind records user activity and stores all data locally
“Rewind is a truly personalized AI. So it captures everything you see, say, and hear, stores it all locally on your devices.”
Siroker: The biggest flaw of note-taking is needing foresight on future needs
“You never know what you're going to need later. That's the biggest downside to taking notes is like you need the foresight to know what is it the thing I might want later.”
Siroker: Rewind stores data locally and Rewind has no access to it
“They just passively capture everything in a privacy sensitive way. It's stored locally. We don't have access to it. Only you have access to it.”
Siroker: Rewind could not exist without Apple Silicon
“And it's so, you know, co occurred with Apple Silicon becoming really powerful and enabling this before Apple Silicon, you know, rewind couldn't exist.”
Siroker: Rewind compresses screen recordings 3,000x for local storage and OCR
“We capture everything your screen is showing. We compress it, you know, 3000 times smaller, so it stores, it can be stored locally. We do optical character recognition on every word that appears on your screen. And now, you know, we're doing transcription as w…”
Siroker: Three Team Members Ranked Rewind as the Worst Strategic Option
“Three out of my team members thought this was the worst of the five strategies to pursue.”
Siroker: Meeting Bots Are Vitamins; Continuous Screen Capture Is a Painkiller
“Scribe was kind of an, and all kind of meeting bots are kind of nice to have. It's a vitamin, it's useful, you know, but like the idea that you could capture anything you've seen and never have to take notes again, that was a painkiller for some people.”
Siroker: Every Rewind engineer ships to production daily on average
“You know, we ship on average, you know, every engineer ships code production every day on average.”
Siroker: Every new Rewind engineer ships to production within 24 hours
“Every engineer, when they start at Rewind, within 24 hours of starting, they ship to production.”
Siroker: Most Regrettable Founder Mistakes Were Deferring to Board or Experienced Hires
“My gut says we should do A, somebody else thinks we should do B, and it turns out poorly. There are plenty of times we said I should do A, and it turns out poorly. I don't remember any of them. The ones I remember are like these deep existential, my intuition …”
Siroker: Rewind's 20-person team executes faster than Optimizely's 450-person team
“Today we have 20 people and rewind we had 450 and I feel like with 20 people we're able to do more and execute faster and be More tightly aligned than and ship than we ever did at Optimizely.”
Siroker: Ex-Apple founders often fail to ship due to unrealistic quality expectations
“I see a lot of companies today that are ex Apple people who've built amazing things and they go off and they spend years and years toiling away and never really ship anything. And, you know, I think they have this sense of like, it needs to be As good as the l…”
Siroker: Google's 'Googliness' Was Poorly Defined and Occasionally Weaponized
“Google has this, Sense of Googliness that you had to assess candidates by, but it always frustrated me that there was no clear definition of what Googliness meant, what it meant to be Googly. And I actually found that it tended to, on occasion, become weaponiz…”
Rewind Evaluates Performance Exclusively Against Defined Cultural Values in Quarterly 360s
“When we do quarterly three sixties, which most startups don't do when we do them, the criteria we evaluate each other and I get evaluated on by my team is our cultural values. There's nothing else. There's no other thing that your performance it is. Do you liv…”
Siroker: Rewind AI exceeds Rahul Vohra's 40% PMF survey benchmark
“Technically by Rahul's framework of, you know, it's greater than 40% of people, very disappointed if they can all use your product. Yeah, we exceed that threshold”
Siroker: Rewind lacks a specific target persona beyond Apple hardware
“This is something I'm probably most embarrassed about in terms of our decisiveness. It's just, we don't have a super specific who you know, we have almost by constraints of the technology M one and M two users and app, you know, Apple or iPhone users.”
Siroker: Optimizely would have been far more successful picking one customer profile
“And had I chosen one or the other, it didn't even matter which one it was, we would have been massively successful one or the other. I mean, there's now huge companies that are, have been built and been successful by doing that. So I know we would have been mu…”
Siroker: Many AI tools are flash-in-the-pan demos with zero retention
“I think a lot of AI tools today are the sort of flash in the pan, exciting, cool demo. Well, cool. Isn't it amazing that technology can do this, but then nobody sticks around.”
Siroker: Private, continuous personal context is Rewind's moat against LLM providers
“The data of the fact that we were introduced by Paul Graham in 2010, the data of the fact that he's an investor in our company and that, you know, he just launched a new feature. All of that is data that no large language model company or provider could offer.…”
Siroker: Rewind AI has massive switching costs due to accumulated personal data
“The switching costs are very high. And we also have everything you've seen that are heard up until you start, since you use our rewind. So even if tomorrow, Some big tech company launches what we do. You know, we've got a lot of users who say, well, why'd I sw…”
Siroker: Thousands of multi-billion-dollar vertical LLM application companies will emerge
“There's whole categories of companies that are going to be cat, you know, Category defining independent successful big public multi-billion dollar companies that are just applications of large language models in a specific domain for a specific user with a gre…”
Siroker: Tech adoption tips when utility greatly outweighs perceived privacy concerns
“And so the, when the utility greatly outweighs the perceived concerns, I think that's when these things tip.”
Siroker: Rewind ensures it passes the 'subpoena test' by retaining zero data
“You know, for us, the test I always use is if we were subpoenaed by the government, would we hand over anything useful or meaningful? No, like we never want to be in a situation where We have data about anything you've done that then we could hand over and tha…”
Siroker predicts 50% of work will be fundamentally changed within 10 years
“I think a lot of jobs will be gone in 10 years, but they'll likely be replaced with new ones.
I think that's probably an obvious
Conclusion.
I think on the order of probably 50% of work will go away.
It would be fundamentally changed.”
Siroker: Highly paid hourly knowledge work like legal tasks will likely disappear
“I do think things like white collar knowledge work that is high paid, lawyers being one of them, is probably a good area. I think anything where today people are paying a lot of money and people are being paid by the hour that a machine could do as better, in …”
Siroker: AGI will eventually perform half the work of human engineering teams
“Eventually, you know, I don't think it will necessarily replace engineers, but it'll certainly augment them. You'll have plenty of, you know, half the work that an engineering team does will be done by AGI, and half will be done by humans, and the human stuff …”
Siroker: Rewind published its unredacted investor pitch deck and video on Twitter
“That's why I announced and shared our investor deck unredacted. I didn't hide any numbers. You know, you can find it on Twitter, you know, seven minute pitch you know, millions of people have seen it.”
Siroker: Public fundraising finds better long-term investor alignment than localized networks
“And that's how I view this relationship with my investor. It's something I'm going to spend the rest of my life with. I'm going to be doing this for decades. I don't want to just meet the person I, you know, find at a bar. I want to cast a wide net and potenti…”
Siroker: Rewind turned down 20 term sheets at a $1 billion valuation
“In fact, we didn't, we had. 20 offers or more of a billion dollar valuation. We turned down.”
Siroker: Pitch junior VCs at non-target firms simply to practice your presentation
“I would hone it and use I hope I just doesn't offend anyone, but use like associates and VPs at firms that you don't want to take money on as practice. Cause you meet these people, their whole job is to meet with you and get deal flow. But like, actually, you …”
Siroker: Investors dislike auction dynamics and prefer negotiating their own price
“The part of the auction dynamic, I think made people feel like they were part of an auction and investors don't like that feeling. They like to think they got a, you know, that they did their job to get a good price at the current price.”
Siroker: Rewind is exploring custom hardware due to off-the-shelf device limitations
“We are hiring for hardware people so it's possible that down the line if we cannot, by the way, I'm very much a big fan of software. I don't know if you're if you've ever done anything hardware related, but software has many, many benefits of which I'm the big…”
Siroker: Google culture prioritizes technical sophistication over solving customer problems
“At Google, you know, they value sophistication of the technology more than they value solving a customer problem. And so that was a cultural thing that started from the early days that served us well, served Google well for certain products like search and pro…”