We at Sequoia are thrilled to lead the Series C in StarkWare, deepening our preexisting relationship that began in 2018 at the Series A and continued earlier this year with the Series B.

Ethereum has been at the heart of a lot of the innovation in crypto over the past few years — DeFi, NFTs, Web3 and more. But while those use cases offer a glimpse of what’s to come, their mass-market success has been limited by the price and throughput of the Ethereum network. With ~15 transactions per second, tens of dollars in gas fees, and minutes (or hours) to achieve finality, it is difficult to support the global population.

While sharding and proof-of-stake will take Ethereum to thousands of transactions per second (or more), we will need millions per second — or even more — for Ethereum and similar networks to reach their full potential as a global, decentralized computer.

Zero-knowledge (ZK) rollups offer the most tantalizing solutions for scaling blockchains. Today, every single Ethereum node runs every smart contract in its entirety. With ZK rollups, a single node can execute a smart contract and produce a proof that it did so correctly. Other nodes can very quickly trust the transaction by verifying the proof.

StarkWare has been the market leader in ZK rollups since very early on. The founding team includes multiple world-class cryptographers; Eli Ben-Sasson and Alessandro Chiesa are also founding scientists of Zcash, and Michael Riabzev is a co-inventor of STARKs along with Eli. The team also boasts some of the best researchers in the zero-knowledge community.

They have already been iterating at a breakneck pace. Last year, they released StarkEx to help individual applications scale using STARK-based proofs. Those applications are now flourishing — including dYdX, which processed more than $100B in trading volume last month alone. Sorare — the most popular fantasy sports game on the blockchain — has seen the cost of gas drop by more than 200x. Over $200 billion and over 50 million transactions have been settled via StarkWare to date, including over 21 million NFTs that have been minted.

That innovation is accelerating. Cairo, StarkWare’s language for writing scalable blockchain applications, is seeing increased adoption, and third-party Solidity → Cairo transpilers (e.g., Warp) are making it easy for Ethereum developers to get started or to port over their work. Validium and Volition are innovating on what data needs to reside on-chain (useful for both privacy and scalability reasons).

And the most important release is on the near-term horizon: StarkNet. A fully permissionless, decentralized scaling solution for Ethereum, StarkNet will enable any developer to build applications that are orders-of-magnitude more scalable than L1-based apps.

We are grateful to be partnered with Uri Kolodny, Eli Ben-Sasson, Michael Riabzev, Alessandro Chiesa and the entire team.

To learn more and start building, check out https://starkware.co/. If you’re interested in joining the team, go to https://starkware.co/careers/.

StarkWare: Scaling Ethereum through permissionless, decentralized zero-knowledge rollups was originally published in Sequoia Capital Publication on Medium, where people are continuing the conversation by highlighting and responding to this story.

The journey began in two ways. First, I was lucky enough to be classmates with both Parker Conrad (co-founder & CEO) and Matt MacInnis (COO) twenty years ago. Second — and equally important — I spent the early years of my career at Microsoft.

One of Microsoft’s fundamental strategies was to build products that worked seamlessly together. Whether it was Windows, Office or Visual Studio, enterprise buyers had a one-stop shop for software products that worked well alone, but worked even better together.

A critical but under-appreciated linchpin of Microsoft’s technical stack was Active Directory. Active Directory was Microsoft’s enterprise identity system that stored basic information about an organization’s employees and enabled everything from auto-complete of email addresses in Outlook to the sharing of network files and folders.

Netscape CEO James Barksdale once famously observed that tech tends to oscillate between waves of bundling and unbundling. During the early part of a disruption cycle, new companies tend to build narrow, “best of breed” solutions that compete with incumbents. As the market matures, these solutions tend to be bundled back together.

As enterprise software has moved to the cloud, we have seen a massive unbundling of products into ever smaller point solutions. Many startups and small businesses have more SaaS subscriptions than employees.

While each tool might be individually great, this setup introduces a tremendous amount of administrative complexity. Data is fragmented across many different systems, each with bespoke connectors to dozens of other systems. Managing your business and getting a simple, holistic view of how you’re doing becomes incredibly difficult.

Enter Rippling. Rippling’s success is rooted in two core insights. First, most enterprise software is fundamentally about employees doing work — it helps salespeople sell, marketers market, or engineers build. Second, most of that software duplicates a common set of features. Mature SaaS product ends up implementing reports, dashboards, alerts, workflows, user authentication and the like — each with a bespoke, inconsistent approach.

Rippling is underpinned by an employee graph that unifies both IT-style identity management and traditional HRIS/payroll systems. Where Active Directory historically offered a simple list of employees and groups, Rippling provides a rich graph of data about the organization, employees, their devices and apps that can be used across the entire platform.

On top of that, Rippling has built a robust application platform that enables the fast development of both first- and third-party applications. Unlike most companies at this stage, Rippling already has over 10 distinct product lines that are each individually at a $1M+ ARR run-rate.

As with the Microsoft of the past, each product works well on its own, but even better together. The shared data model and the shared platform enables rich integration across all of the applications. This strategy is working — the customers we spoke with love the product, with Rippling boasting a 4.9/5.0 rating on both G2 and Capterra.

Rippling is one of the fastest-growing companies we have seen in the past few years. Today, they have more than 20 different first-party applications on their platform and distribute many more. Over time, we see a world where Rippling becomes the one-stop shop for both SMBs and larger organizations to manage their employees, applications and devices all in one place.

We believe we are at an inflection point on the bundling/unbundling curve — especially for SMBs. We are excited to work with Parker and the team to simplify the lives of millions of entrepreneurs and business owners. To learn more, check out rippling.com or rippling.com/careers.

Rippling: The Rebundling of Enterprise Software was originally published in Sequoia Capital Publication on Medium, where people are continuing the conversation by highlighting and responding to this story.

Today, we are incredibly excited to announce our partnership with Vijaye Raji and the team at Statsig. We believe Statsig will become a foundational platform for the way modern teams ship software.

The Art & Science of Product

Many disciplines start as art and evolve into some blend of art and science over time.

Decades ago, marketing was dominated by brand advertising and Madison Avenue creative on billboards and TV. Today, performance marketers drive the industry with personalized creative and precise measurements of CAC and LTV.

Sales was once defined by schmoozing and steak dinners. The rise of Sales Ops has brought forecasting, measurement and a scientific precision to the profession.

Today, Product is still in the early days of transitioning from art to science. Platforms like Amplitude have made it easy to holistically instrument your product and understand how your product is driving your business goals. But many questions remain hard to answer.

E.g., in most companies, if you wanted to know who the most productive sales rep was in Q2, it would be trivial to find out. If you wanted to know the best marketing channel and campaign in Q2, it would be similarly easy to answer.

What if you wanted to know the most impactful product feature you launched in Q2? How would you find that answer?

For most product teams, answering that question would be hard (if not impossible). The team might have been somewhat scientific — they may have tried a few different variants and be confident that they picked the best colors for their buttons. But it would be nearly impossible to know the absolute impact of that change on the core business metrics (user growth, engagement, retention) and whether that change was the most impactful one.

Statsig

Enter Statsig. Statsig is a platform that helps blend the art and science of product development by letting product teams quickly and continuously measure the impact of product changes as they ship them.

The base of the system is Feature Gates, which enables developers to encapsulate any feature in a basic on/off switch.

Pulse is a tool that then automatically computes the impact of each of these features on every business metric that you care about. This moves you from maximizing button clicks to measuring the impact of each change on user growth, engagement and ultimately revenue.

Experiments+ enables you to experiment with multiple variants while perfecting your new feature. Autotune then helps you figure out the optimal configuration of your experiments for whatever business goal you have. Holdouts enables you to group a set of features together (e.g., the collective output of a given team) to measure the aggregate impact of all of these changes.

Individually, each of these tools is powerful. Collectively, they are transformational to product development.

Why? Because once you can quantify the impact of any product change on every business metric you care about, you empower any designer, engineer or PM to ship product improvements in a massively decentralized way.

If someone has an idea to improve the product, they can just test their idea and — if it is positive with no downsides — they can just ship it.

In the context of investing, my partner Doug Leone likes to say “one hundred smart people are better than one smart person.”

We think the same is true of product development — in an ideal org, any designer or engineer should be able to find their own way to make the product a little bit better each day. By unleashing everyone’s creativity, product velocity and innovation can radically accelerate.

Team

The Statsig team knows this better than almost anyone else. As an engineer extraordinaire at Facebook, Vijaye and his team were the driving force behind many new products, including Mobile App Install Ads, Audience Network, Marketplace, FB Gaming and more. One of their secrets? Speed and constant iteration (largely powered by Statsig’s predecessors).

True to form, the Statsig team has moved incredibly quickly. The company is less than six months old and has already shipped a full product development stack (Feature Gates, Pulse, Experiments+ and more).

We are thrilled to work with the team and look forward to a new era of data-driven product development. To learn more, check out statsig.com.

Statsig and a New Era of Product Development was originally published in Sequoia Capital Publication on Medium, where people are continuing the conversation by highlighting and responding to this story.

We’re excited to be working with Peter Gao, Quinn Johnson and the team at Aquarium! Aquarium is a data management platform for Machine Learning (ML) that enables ML teams to manage and curate their training datasets.

Machine Learning is clearly one of the most important technology trends of our time. After the hype, ML is finally starting to creep into a wide range of industries — manufacturing, agriculture, energy and, of course, automotive.

While ML is increasingly ubiquitous, most people still have a relatively simplistic mental model for how to build an ML model:

• Step 1: Collect and label as much data as possible
• Step 2: Train a model with that data
• Step 3: Fiddle with some knobs (i.e., hyperparameter tuning)
• Step 4: Deploy

If you’re just training a hot dog detector, then this might be all you need, but most tasks are more complicated.

Consider any autonomous vehicle (e.g., a car, tractor, drone, etc.). A single camera might record 30 frames/second, capturing about 2.6M images per day. Now imagine multiple cameras on multiple vehicles operating 24/7 — the amount of raw data any vehicle might capture is immense. Practically speaking, it’s impossible to label all this data, let alone train on it.

The simplest solution would be to randomly pick images to label, but this is non-optimal — common things would still be common and and rare things would still be rare. Ideally, you want to build a data set that captures all of the situations you might encounter, not just the common ones.

In reality, the above mental model was too simplistic. Collecting data is not enough. You have to be thoughtful and careful about what data you train on. A better model would be:

• Step 1: Collect as much data as possible
• Step 2: Curate an initial training data set
• Step 3: Refine your model
• — Step 3a: Label any unlabelled data in the training set
• — Step 3b: Train a model
• — Step 3c: Observe failures in that model
• — Step 3d: Update your training data set
• — Step 3e: Repeat until satisfied
• Step 4: Deploy

Enter Aquarium. Aquarium is effectively a platform for managing steps 1–3. It enables ML engineering and operations teams to manage and curate their training data and then iterate on their models until they’re ready for production.

While it’s tempting to fixate on the latest and greatest models coming out of OpenAI or FAIR (and that work is great!), the hard work for most companies is this constant curation of training data. This work is where models go from theory to practice.

Peter and Quinn both know this problem well. They spent years at Cruise Automation where they worked on the ML models for self-driving cars. It was there they realized that the core work of surpassing human-level performance and safety is this constant re-curation of training data. They founded Aquarium to bring those learnings and tools to the broader engineering community. After less than a year since founding, Aquarium is already working with over a dozen customers like Sterblue and AMP Robotics to help them manage their training sets and, ultimately, improve their model performance in production.

As ML technology becomes increasingly accessible, Aquarium’s ultimate goal is to enable non-technical domain experts to build and refine models over time, whether that’s radiologists focusing on early cancer detection or botanists focused on increasing crop yield. We are grateful to be on this journey with Peter, Quinn and team! If you want to learn more, check out https://www.aquariumlearning.com/.

Aquarium: Better Models Through Better Data was originally published in Sequoia Capital Publication on Medium, where people are continuing the conversation by highlighting and responding to this story.

Many of the companies we meet are founded by technologists — either young, dynamic founders with a computer science background or seasoned product managers or executives that see a better, different way of solving a problem.

When we meet them, they often have pitch decks structured using the business plan framework we shared many years ago. For the best pitches, it’s clear the founders have an intuitive understanding of the problem and how their product will be a novel solution loved by users. They’ve studied the competitive landscape and know why now is the right time to start the company. The product demo is great.

And then they bring up the Market slide.

The Market slide is often just a big number, confidently stated (where big varies from $1B to $1T+). If you probe, it’s clear that the number is a result of googling their company’s category plus the words “market size” (e.g., “video software market size”). Critical questions are often deflected.

The market you choose to serve is one of the most important factors for an early-stage startup. And for most technologists, it’s a blind spot. To help, we wanted to share one way of thinking about market.

The Market Formula

Market Size = (# of Customers) × (Revenue/Customer)

The simplest and most important way to think about market size is (a) how many potential customers are there and (b) what might each customer be worth to you.

For many of the companies I meet, I am constantly trying to triangulate these two variables as we talk.

One of my favorite questions for dissecting this comes from my partner Doug Leone. He will often ask:

Two questions, please — what % of the Fortune 2000 will ultimately buy this product? And how much will they be worth to you on average?

The reason this question is great is because for most enterprise-focused companies, the vast majority of their revenue will come from the Fortune 2000 (F2000). This question provides a very quick way to approximate market size.

If your answers are 20% and $500,000, that might seem like a great business (it is hard to get someone to pay $500,000/year for a product!), but 20% × 2000 × $500,000 = $200M. If most of your revenue is going to come from the top 2,000 companies and you are only serving 400 of them, you might be in a niche market.

On the other hand, if you have a broad, horizontal platform that serves the entire F2000 and you expect ACVs of $1M or more, then it’s easy to see your way to a $1B+ market (2000 × $1M = $2B).

If you are going to show a market size number, I would strongly encourage you to have thoughtful answers to these two questions: how many customers? and how much will they pay?

The Market Curve

Thinking about the number of customers and the revenue per customer is a tremendously clarifying way to think about a single company. It also provides a useful framework for unifying seemingly disparate categories:

Here, we have plotted the number of customers on the X-axis and the average revenue per customer on the Y-axis. The green line represents ~$2B in annual revenue (i.e., x × y = $2,000,000,000; my apologies to math purists who will point out that this is not entirely to scale).

Enterprise

I tend to think of the top-left corner here as classic enterprise sales businesses. At this part of the curve, you probably have tens or hundreds of customers, but they pay you a lot of money. For instance, Medallia went public last year. According to their Q2 2020 Earnings Deck, they have 839 customers and had $115.5M in quarterly revenue, suggesting ~$550k/customer/year in annualized revenue.

Where you are on the curve informs your sales strategy. If you have hundreds of high-value customers, you can afford to spend a lot on sales, implementation and customer success to make sure those customers remain happy.

As you move down the chart to Zoom, you’ll realize that “Enterprise” is a pretty broad range. According to Zoom’s Q2 Press Release, Zoom now has over 370,000 customers with more than 10 employees and $663.5M in quarterly revenue, suggesting ~$2.65B in ARR and ~$7,100 in annualized revenue/customer in that segment.

Zoom is unusual in that it now basically serves all markets, but it helps show the power of having a viral, bottoms-up product in the enterprise — they have almost three orders of magnitude more customers than a typical enterprise-focused vendor.

SMB

As you move down the curve, you hit the range of small and medium-sized businesses (SMBs). Yelp is a great example here. According to Yelp’s Q2 Earnings, they had 377,000 paying locations and $169M in net revenue, implying ~$1,800 annualized revenue/location.

One of the reasons it’s important to think about this curve is because it changes your sales strategy. If you have customers paying $500k/year, you can afford to have a dedicated account team per customer. If you’re earning $2,000/customer/year, you have to be much more efficient with how you sell to a new customer.

Another tough challenge in the SMB market is attrition. There are relatively few bankruptcies each year in the enterprise. The SMB market tends to have a much higher level of built-in attrition and tends to be more susceptible to macroeconomic shocks. E.g., in Q2 2019, Yelp reported 549,000 advertising locations and ~$247M in quarterly net revenue — their advertising base sadly shrunk by ~31% during the global pandemic.

Prosumer

You wouldn’t think that Peloton and Yelp are comparable businesses, but they looked surprisingly similar at the beginning of the year (they have since started to diverge). When Peloton went public last year, they reported 511,000 subscribers and $915M in FY2019 revenue (vs. Yelp’s ~549,000 advertising locations and ~$990M in annualized net revenue in Q2 2019).

Peloton has seen tremendous growth since the pandemic. In Peloton’s Q4 2020 Shareholder Letter, they reported $607.1M in quarterly revenue and ~1.1M Connected Fitness Subscriptions (suggesting ~$2,200 in annualized revenue/subscriber). These revenue numbers include hardware sales, which can be lumpy. Excluding hardware sales, they reported ~$121M in subscription revenue in the same quarter, suggesting ~$444 in annualized subscription revenue/subscriber.

Peloton and Yelp are obviously very different products with different customers and business models, but thinking about them along this curve helps to highlight some similarities.

Commerce & Marketplaces

As you continue moving further down the curve I tend to think about commerce and marketplace businesses. Typically, these are businesses that sell to millions or tens of millions of people, and they often make tens to hundreds of dollars per year in net revenue.

One public example here would be Grubhub. In Q2, Grubhub announced quarterly revenue of $459M and 27.5M active diners, suggesting quarterly revenue of ~$16.70/diner and annualized revenue of ~$67/diner.

Comparing Grubhub and Peloton, you see that Grubhub has about 30x more consumers and about 30x less revenue/consumer, leaving both with annualized revenue of ~$2B.

Consumer Apps

Lastly, we have mass-scale consumer apps. This is the part of the market where ARPUs are quite low but the number of users is massive.

Snap’s last earnings showed they had around 249 million customers and $679M in quarterly revenue, suggesting about $10.90 in annualized revenue/user. So again as you move from Grubhub to Snap, you have about 10x as many users but about 5-10x less revenue/user.

As you move even further down the curve, you get to the extreme outliers. According to their Q2 earnings, Facebook has 2.7 billion MAUs and their quarterly global average revenue per user (ARPU) is $7.05, which puts them at ~$28 in annualized global ARPU.

Most mass-scale consumer apps are, by necessity, free (and either monetized via ads or, increasingly, optional subscriptions). For these apps, the primary question I ask myself is “how big can this get?”

Some categories seem limited to tens of millions of users. While good, you likely need an ARPU of ~$100 for the market to “big enough.” On the flip side, if you have an app that can credibly reach >1B users if everything goes right, you only need a $1 ARPU to be at $1B in annual revenue.

TL;DR

Having a deep understanding of the market you are serving isn’t just important because potential investors will look for it. It’s critical because a great product is not sufficient for a great business — you need a great market, too.

For your market, the two key things to understand are (a) how many potential customers are there and (b) how much will each of them be worth to you if everything goes right. The better you really understand those details and can go deep on them, the more successful both your pitch and your business will be.

The Market Curve was originally published in Sequoia Capital Publication on Medium, where people are continuing the conversation by highlighting and responding to this story.

Four years ago, we partnered with Venkat Venkataramani, Dhruba Borthakur and team on a new platform for building data-driven applications. After joining forces with our friends at Greylock on both the Seed and Series A, we are delighted to triple-down on this tremendous team by leading their Series B.

One of the defining trends of the past five years has been the rise of ELT (Extract, Load, Transform). In the beginning of the “big data” era, enterprise data was shuffled from place to place via complex, brittle data pipelines. In the best case, these systems led to hours or days-old data and months of development time. In the worst case, these pipelines would randomly break and your applications would be operating on stale data for days or weeks.

Tools like Fivetran, Snowflake and DBT have ushered in a new model in the data warehouse space. Instead of a complex pipeline, the ELT model connects all of your data directly to your data warehouse (e.g., Snowflake) and then centralizes all of your transformations in a single system like DBT.

This model is transformative for warehouse-style use cases and enables queries that take seconds on data that is hours- or minutes-old.

What do you do if you need something even faster? What if you need application-speed queries (e.g., tens of milliseconds) on always-fresh data that is spread across multiple systems?

That’s where Rockset comes in. Similar to ELT systems, it connects to your data stores or data lakes, indexes them in real-time and then enables fast, scalable, complex SQL queries (with JOINs) across those data sets.

One common use case for Rockset is real-time analytics — aggregating data from multiple sources and enabling real-time metrics and dashboards across that data. Unlike others in this space, Rockset shines by also enabling rich, full-fledged applications that span these data sets. One e-commerce customer is using Rockset to power their personalization algorithm to ensure that their recommendations adapt in real-time as the consumer browses. Another customer is using it to track a complex logistics operation to ensure that they know the status of every asset in real-time.

In our references, the most common thing we heard from prospects was skepticism around whether this was technically possible. The most common thing we heard from active customers was that it was “magic.” For one customer, it took a six-month roadmap item and compressed it into an afternoon.

We think the speed, scale and simplicity of Rockset will be game-changing for the cloud data stack. We are thrilled to strengthen our partnership and excited for the road ahead.

Rockset and the Real-Time Cloud Data Stack was originally published in Sequoia Capital Publication on Medium, where people are continuing the conversation by highlighting and responding to this story.

The past decade has seen a revolution in the ways that computers perceive the world. Computer vision is reshaping industries like retail and transportation. For many tasks, computers are already better than humans at “seeing” and only getting better.

What about hearing?

As it turns out, computers are pretty good at that too (and getting better very quickly). That is the core insight behind Whisper, which is launching the Whisper Hearing System today.

In early 2017, we reconnected with Dwight Crow, who had been an early engineer at Arista Networks, a YC founder and then a PM at Facebook. He was working out of his bedroom on a new project, which was a total mess of computer parts, PCBs and soldering equipment.

After studying recent advances in machine learning, he realized that there was just as much opportunity in the audio domain as there was in the visual. He played us a noisy, unintelligible clip of people talking in a restaurant. He then passed the same clip through an AI-based filter that clarified the voices and removed the background noise. The improvement was amazing.

While the easier path might have been to apply this technology to voice assistants or transcription tools, Dwight instead went after a less-obvious but much more meaningful opportunity — hearing aids.

The sad truth is that most hearing aids are disappointing. My father has owned a hearing aid for a decade (and probably needed one for twice that time), but he almost never wears it. It’s loud and uncomfortable. He constantly faces two terrible choices — keep it in and be fatigued by loud background noise or turn if off and only catch snippets of our conversations. He usually picks the latter.

The core problem is two-fold. First, most hearing aids are like high-end sound systems with a fancy equalizer. An audiologist can help discover where your hearing is impaired and tune the hearing aid accordingly, but that tuning amplifies all the sounds at that frequency. Unfortunately, an air conditioning unit sounds a lot like the “sh” phoneme, meaning that most hearing aids will make your A/C maddeningly loud.

Second, hearing aids are static. Once they are fitted and sold, they remain the same forever. They don’t learn about you and your surroundings. They don’t get better every week like modern software and hardware. They stay the same until you buy a new, static device a few years later.

Whisper takes a completely different approach. Instead of only focusing on frequencies, the Whisper system analyzes sounds (voices, music, background noise, etc.) and amplifies the sounds you want to hear while suppressing others. Using a chip with 1,000 times the power of your typical hearing aid and a connection to your phone, the Whisper Hearing System receives continual software updates to get better and better over time.

Over the past three years Andrew Song, Shlomo Zippel and Dwight have assembled a world-class team from Apple, Facebook, Nest, startups and industry incumbents to rethink the way that hearing aids work. Building a robust, embedded system in a mature industry that runs state-of-the-art machine learning is not an easy task. The team has done an incredible job.

It’s truly rare to find an exceptional team going after a great market with a meaningful mission. We are thrilled to be working with this team. Congratulations on the launch!

Whisper: Reimagining Hearing was originally published in Sequoia Capital Publication on Medium, where people are continuing the conversation by highlighting and responding to this story.

In 2002, I graduated from Harvard and landed a job at Microsoft as a PM. I ended up at Microsoft largely through the hard work and persuasion of my college recruiter (Colleen McCreary, who is wonderful and is now the Chief People Officer at Credit Karma).

While I’m incredibly grateful for this experience, the story reflects a structural weakness in how college recruiting happens today. I was fortunate enough to attend a top university. Microsoft was large enough to be able to invest in dedicated college recruiters for the top engineering schools.

But most college graduates and companies aren’t so fortunate. Most companies can’t afford to recruit on-campus. If they do, they tend to focus their efforts on a few schools. If you’re not at one of these schools, you have to work that much harder to get access to the same opportunities.

Enter Jumpstart

When we first met Ben, we realized we had found someone with authentic empathy for this problem. Ben grew up in London, dropped out of high school, started in an entry-level role in his teens and then founded and led his own high-end search firm by the time he was in his mid-20s.

Ben specialized in finding great people in non-traditional places. But while he might be able to help dozens of people a year find great opportunities as a recruiting firm, to solve this problem at a much broader scale, Ben would need a different approach, so he founded Jumpstart.

Jumpstart’s mission is to make recruiting fair for all. It enables companies to engage with students across a wide-range of schools via both in-person and virtual events. For students, it provides a community to learn about and engage with a much broader range of companies.

At the core of Jumpstart is an in-depth sign-up experience that lets candidates share much more detail about their background and interests than they would be able to on a traditional resume — what classes they took, what programming languages they know, and what kind of job environment they’re seeking.

When I did on-campus recruiting for Facebook, I was amazed at how primitive the system was — I was emailed a 400-page PDF of every CS student’s resume and left to make sense of it. Jumpstart uses candidate’s deep profiles to build a personalized list of the best matches for your role. It searches over the entire system — not just a single school — and gives you one prioritized list of candidates.

Early results have been incredibly promising. Lyft is an early adopter of the platform. Over a third of their 2020 intern class was hired using Jumpstart — and over 80% of those interns come from underrepresented backgrounds.

The Future

Ultimately, Jumpstart’s goal is to make the entire recruiting world more fair and more efficient. Ben and the team imagine a world where anyone can find their dream job regardless of brand or pedigree and companies can find their ideal teammates regardless of geography or background.

In a little over a year, there are already over 100 companies using Jumpstart to connect with early-career talent, including Airbnb, Doordash, MongoDB and JP Morgan. Today we’re thrilled to announce our partnership with Jumpstart and look forward to working together to make recruiting fairer for all. If you’re a recruiter, a student, or someone who’d like to join us on this mission, please reach out at http://jumpstart.me/.

Jumpstart: A better approach to early-career recruiting was originally published in Sequoia Capital Publication on Medium, where people are continuing the conversation by highlighting and responding to this story.

Group communication in the workplace is broken. People have nightmares about accidentally hitting “Reply All.” Email may work fine for broad status updates, but conventional wisdom is to call a meeting whenever you want to have a real discussion or debate.

Yet meetings have their issues, too. As groups get bigger, conversations generally get worse. “Listening” becomes “waiting to talk.” Many people just give up and let the conversation be dominated by the loudest, most extroverted voices in the room.

One obvious solution is to shrink the meeting to a smaller group. This might save time and make the meeting more efficient, but at the cost of excluding important voices and perspectives from the conversation.

At Sequoia, we believed there had to be a better way. When Rousseau Kazi and team came in to share their early idea for Threads, we loved it — a platform designed to make work more inclusive by empowering teams to discuss and make decisions at scale.

After a year of development, Threads is launching publicly today. To commemorate the launch, we sat down with Rousseau to talk about the past, present and future.

MIKE: So … what’s your story?

ROUSSEAU: My parents immigrated here from Bangladesh. I grew up in southern California. My dad’s a math professor. I was named after the French philosopher Jean Jacques-Rousseau because my mom was a philosophy major and thought it’d be funny.

Have you ever met another Rousseau?

I have not. I do know one person named Locke, but that’s about it.

I went to UC Berkeley and studied computer science. After my second year, I interned at Facebook and loved it. I ended up getting a full-time offer and decided to graduate early and join as a product manager. I was 19 and it was terrifying. I was really insecure at first. I didn’t really know what a product manager did. I was blessed to have good managers and leaders that constantly invested in me.

One of things that stood out to me when you were at Facebook was your energy and the sheer number of extracurriculars you undertook. Of all the side projects you did at Facebook, which are you most proud of?

Oh, Hyperlapse for sure. I originally got involved in that project because I’d been helping manage our Hackathons. A group of engineers showed us these wobbly videos of people using their cameras while on a boat or riding a bike. Then they showed an algorithm that “automagically” stabilized the video even as you sped it up, so you could create these gorgeous time lapses. It was really cool, and the crowd was like, “How did you guys build this?”

We took it to Zuck, and everyone thought it was special. But there was a lot we had to get done at Facebook, so it wasn’t anyone’s top priority. The team was still passionate about the project, though. I decided to work with them and we got it to the point where it was focused enough to launch. On the first day, an intern used it to make a 15-second White House tour that was shared from Obama’s Facebook page. Then Jimmy Fallon started using it. Someone made a music video. It was just one of those things that was simple to use but also powerful, magical technology. It’s great to go on Instagram and see all the #hyperlapse videos getting created every day.

Why leave Facebook and set out on your own?

I have always had an itch to start my own thing. I remember for career day when I was in third grade, I dressed up as Thomas Edison because I wanted to be an inventor. I brought a briefcase full of Legos and my sister powdered my hair. I’ve always loved the idea of innovating and making people’s lives easier.

I met all three of my co-founders while working on side projects. Suman Venkataswamy and I were in a math class at Berkeley, and we’ve been roommates ever since. He’s my best friend. Jon McCord is the first engineer I ever worked with. We did our first hackathon together at Facebook. And Mark Rich was an incredible analyst I worked with at Facebook. I had this short list of people I wanted to found a company with and all three of them were on it. They were always ahead of me on something, and they made me want to be better every time I was around them. So when I was deciding to leave and they were all willing and ready to try to solve a problem together, it felt like the right time.

How did you land on Threads?

We started out exploring the education space. We noticed that as we were trying out different ideas, things were getting lost. I’d write out something in a messaging client and it would disappear into the ether. We moved to email, but people wouldn’t read them or the discussion would get kind of crazy. So we ended up having many more in-person meetings, which felt like an unnecessary use of time. Nothing was being documented. Even with so many available tools, this problem still existed. Like so many people, we thought: “Our team is growing. We can’t have everyone in the same room because nothing ever gets done. But if we exclude people, we’ll have death by a thousand paper cuts with all these follow-ups.” Meetings are great for emotional discussions about things like growth — areas of improvement or things you’re doing well. But if someone’s just presenting a deck, I think there are better things you can do with that time.

Meetings also tend to optimize for certain types of people — extroverts and the most senior people in the room. But if you want to innovate to solve a problem, it’s a multiplayer sport. You want to be able to have an environment where everyone feels comfortable engaging. Some people might want more time to think about their ideas before they share them broadly, so they might not be chiming in on the conversation. With meetings, you’re not going to get everyone to speak up.

We started talking to a bunch of friends who worked at other companies, big and small, and they were experiencing the same thing. A lot of people weren’t being heard. We asked ourselves, “Is there a way we can solve this?” And that’s how we came across Threads.

The benefits of this platform are: one, there’s no scheduling required. Instead of spending a bunch of time at work finding time to do work, you can start a discussion on Threads and everyone can engage with it. Two, so much valuable information doesn’t get shared because people are afraid of over-notifying and annoying others. With Threads, you can opt in and out of the notifications you receive. And three, no one has to take meeting notes. The discussion becomes the document, and when you’re done, everyone has this awesome log of all the decisions the company has made. All of this opens up the ability for everyone to have a seat at the table, so you can make sure you’re getting the best ideas from your team. To me, that is the future of work.

How has the idea evolved since you started?

It actually hasn’t changed much since we started building it out, but I think we’ve gotten clearer about what parts we liked and didn’t like. We were inspired by looking at how the internet talks. Online communication is broken up into two categories — synchronous, real-time chat, which is really good for transactional, urgent communication; and asynchronous discussions, where people have more critical and long-lasting conversations. We asked ourselves, “Where are large groups of people having efficient discussions?” And we realized, “Oh, all over the internet.” It’s not atypical to see tens or hundreds of people on a product like Reddit having a focused discussion and moving a topic forward. We realized all of these different platforms effectively do the same thing, which is to scale elegantly as more people enter. So we took that, stripped it down, rebuilt it in a way that’s really focused and fine-tuned to solve this exact use case.

For example, you can mark one of the replies in a Threads discussion as the decision, so even people who weren’t initially involved can quickly and easily get on the same page. You can imagine what a difference that makes when you’re coming back from vacation — instead of going through hundreds of emails you can skip to the five decisions made in the past week.

We also realized there’s an asymmetry between how long it takes you to skim a discussion versus how long it might take you to engage with it. So we allow you to quickly skim threads and then mark them for follow-up. It creates a natural to-do list.

What’s exciting to me now is that we have very different industries and types of companies using our product — from a cricket farm in Ghana, to a company in Austin that sells ketogenic supplements, to a cryptocurrency index fund here in San Francisco. One company said they’ve been able to reduce meetings by 80 percent, and they’re sharing internal information in a way they didn’t think was possible.

You joined Facebook and rose very quickly through the ranks. Now you’re leading Threads. What’s the most helpful career advice you’ve received?

I remember once at Facebook, I’d written out all these ideas I thought we should do. But I was afraid if I shared them with the larger team, they’d take it as prescriptive versus letting them be creative. I shared the doc with the VP I reported to at the time, and he typed out my concerns at the top, in red lettering: “Hey everyone, I took some time to write up all these thoughts. I’m a little concerned you might view this as being prescriptive. That’s not my intention. Just want to help. Hope this works.” And then he hit send. I was so worried! But I got so many positive responses. People were like, “Oh, thanks so much for putting this together. This is so helpful.” That taught me that over-communicating your fears is the best way to avoid misinterpretation.

Another lesson was during a one-on-one with a different manager. I remember I was upset because I felt like I wasn’t growing fast enough, and I asked how I could get better, faster. He said, “Look, if you want to grow really quickly, the best thing you can do is get really good at recruiting. Nothing has a higher ROI. Even if you spend hundreds of hours finding someone for a key position, they’ll pay that back in a few weeks.”

It’s good advice. We were just talking about this last week at Sequoia, and Omar Hamoui said that when he was at AdMob, he tried to spend 60 to 90 percent of his time every week recruiting.

Yeah. Recruiting has been the most leveraged thing I’ve done at Threads.

Threads is a company that thinks about how companies should work. How does that inform your hiring? What kind of company are you trying to build?

Our mission is to make work more inclusive. We say that our team needs to embody that mission as much as our product facilitates it. That means making sure we have a really inclusive culture where anyone we hire, agnostic of level or seniority or anything else, feels comfortable bringing their best self to work. Share ideas, critique ideas. Move the discussion forward. It also means we need to have both advisors and team members with different perspectives. And I’m proud to say 40 percent of our team is from an underrepresented background. I think bringing diverse viewpoints together is already paying off tremendously. When you’re all working toward a shared vision, giving everyone a voice has incredible impact.

Start your own thread at threads.com.

A Thread With Threads was originally published in Sequoia Capital Publication on Medium, where people are continuing the conversation by highlighting and responding to this story.

One of the most important technology themes of the past 10 years has simply been data.

At the beginning of the decade, new data techniques helped drive an increased focus on analytics and A/B testing. In the past few years, the focus has shifted to machine learning. Both technologies are rooted in the same trend — the exponential increase in data.

The CEOs we meet know that their businesses need to become data-driven to compete and survive. But not everyone knows how.

Getting started is easy — you instrument your product or business and start collecting data. But turning that data into valuable, actionable insights is much more challenging.

One of the biggest pain points is just physically moving the data around. The data pipelines and ETL jobs that move data from its initial raw form to the final product are often slow, fragile, and hard to maintain.

There have been many efforts to simplify these pipelines, but the state of the art is still Rube-Goldberg-esque. It can take weeks or months for new types of data to make their way into production applications — if they ever make it at all.

Rockset

At Sequoia, we love partnering with founders that challenge conventional thinking.

When we first met Venkat and Dhruba, they wondered — why do we need data pipelines at all? In this new era of cloud-native infrastructure, why can’t we use web-scale search techniques to build data-driven applications directly on top of the raw data? Why make pipelines 10% better when we can eliminate them entirely?

Their answer is Rockset. A cloud-native service that helps developers run production-ready SQL directly on top of raw data. It’s super simple to get started:

1. Create an account
2. Point Rockset at your data source (e.g., Apache Kafka, Amazon S3)
3. Start writing SQL

Once you’ve configured your data source, Rockset immediately starts indexing your data, enabling you to explore your data and write application-ready SQL within seconds. At Sequoia, we’ve used Rockset to move many cumbersome nightly jobs to real-time dashboards. Not only are our internal tools faster, but our data science team now spends less time babysitting pipelines and more time helping our companies grow.

Rockset is the result of a wonderful, interdisciplinary data team coming out of Facebook. CEO Venkat Venkataramani was the founder of TAO (Facebook’s online graph database) and a database engineer at Oracle. CTO Dhruba Borthakur was one of the key architects of Facebook’s data warehouse and the co-creator of both RocksDB and HDFS. Tudor Bosman was a co-creator of Unicorn (Facebook’s internal search backend) and the Gmail backend. Shruti Bhat held senior product roles at both VMware and Oracle.

We first partnered with Rockset at the seed stage in 2016, along with our friends at Greylock. At the time, Rockset was a rough idea inspired by the idea of converging a traditional SQL database with a cloud-native search engine. Today we’re proud to deepen that partnership with Rockset’s Series A financing.

It’s been a fantastic journey with the team so far — and it’s just getting started. We’re incredibly excited to see what you build with Rockset.

Rockset and the Future of Data-Driven Apps was originally published in Sequoia Capital Publication on Medium, where people are continuing the conversation by highlighting and responding to this story.