Software Plumbing

A quick temperature check of the tech and investing zeitgeist gives away an obsession with developer tools and software infrastructure. The excitement has led to some confusion, as phrases like “platform,” “infrastructure,” “lego blocks,” and “abstraction” are commingled. I’ve meant to write a (separate) post on the nuances of infrastructure and platform companies, and their path to power, but that’s for another time.

I have a working theory that software plumbing consists of families of companies that grow up together, working in lockstep to provide the underpinning for new applications and business models. As an example, the primitives of a user-facing web application are identity, messaging, and payments. See Auth0, Twilio, and Stripe. Several new clusters are coming into the fold, and the (non-exhaustive) list is worth sharing.


Snowflake (warehouse), Fivetran (integration), dbt (transformation)

Running lightning-fast queries. Conducting machine learning experiments. A/B testing and optimizing user journeys. These are all ideals for a data-driven company, but until recently, were quite difficult at scale given how nascent data infrastructure was for a cloud-native world. Now, components like Snowflake for long-term storage & analytics; Fivetran as a data pipe between SaaS applications, event streams, and more; along with modeling and transformation tools like dbt and Dataform set the stage to not only power existing companies with richer insights, but also create an ecosystem of startups with an orientation towards continuous analysis from day one.


Plaid (bank access), Modern Treasury (payment flows)

The American financial system sits at the intersection of legacy systems (did someone say COBOL?) and regulatory entrenchment. Furthermore, the business model of a bank does not lend itself to a destiny as a growth machine. By creating a pathway to “read” bank data, Plaid spurned a generation of neobanks, wealth management tools, and more. On the other end, Modern Treasury can enable insuretechs, lenders, proptech cos, and more to “write” into the banking system by designing abstractions on top of complex payment plows that need to go through wires or ACH. As these two (and others) virtualize the banking functions, I’m intrigued to see what new business models will emerge.


VGS (tokenization), Transcend (privacy workflow)

I’ve written this before, but it feels like being privacy-first as a company has gone from something you take on to save-your-ass to a characteristic that is deeply aligned with customer centricity. VGS maintains compliance and privacy by removing the burden associated with handling sensitive data. Transcend provides the backbone for privacy workflows that entangle end customers and companies. I see a world where these sorts of companies work in tandem to not only “fit” privacy & compliance into current workflows, but also provide a layer on which new privacy-first businesses emerge — think social networks, financial planning, consumer media, etc.

For another time, I’d also like to work through how the cadre of development tools fits into this, whether that be the Jamstack (Netlify, Gatsby, Prisma) or instruments that allow for peace within the confines of a microservices world (Okteto, Terraform, Kong, etc.)

Many of these companies got their start by productizing an oft-repeated process or system across a gamut of companies, building for the markets of the past and present. Yet, the de novo business models that permutations of buildings blocks unleash are the most compelling. Hopefully this list provides the breaching point for exploration.

Two Views on Infrastructure

In the past year or so, we’ve seen significant excitement regarding infrastructure software or technical businesses building beneath the application layer. The stock charts of Datadog ($DDOG) and others, along with impending public offerings (Snowflake), demonstrate investor appetite for this category across late-stage growth and public markets. From my seat in the private markets, the pre-seed and seed rounds of dev tools and infrastructure companies are quite competitive, with demand fueled by a long-term belief in “digital transformation” and runway in the global IT market.

As interest continues to percolate, I find that many people, investors and operators alike, view the category as monolithic (no pun intended). The market rules governing the growth of infrastructure software is the same for all startups, and it seems to be inevitable that secular tailwinds will push all these companies to unicorn status, following a deterministic path. We’ve over-rotated on generalizing specifics and nuanced tactics into abstract concepts about infrastructure software broadly. I touched a bit on this in my piece on open source software.

There are two different ideas about the future of infrastructure software, which I gathered from personal investing, conversations with investors & operators, and observations from the field (or Zoom). One is friendly to the early-stage startup ecosystem, and the other could be perceived as quite unbecoming.

The Big Get Bigger

The growth of the public cloud providers — AWS, Azure, GCP — over the past decade is no joke. Combine their existing scale with increasingly common investor tropes about technology & information markets (reflexivity, lock-in, increasing returns), and these properties will have impenetrable moats around them.

Often, the rebuttal to a startup pitch that goes as, “What if Google / Amazon / Other BigCo does this?” is usually not a good one. Despite their sophistication, they can’t move fast enough to compete with nimbler startups. Yet, in the case of infrastructure software, AWS / Azure / GCP can literally *see* code being run and new products gaining traction. Their lens into the future is crystal clear, and over time, they will subsume new startups in the market. The new infrastructure player of today is a ticket on the public cloud roadmap of tomorrow.

A Less Sober Portrait

The idea sketched above is not friendly to the believer in the tendencies of entrepreneurial energy. The alternative view is that the public clouds have a utility-like nature — they have tremendous scale, the ability to print cash with a recurring revenue stream, but their margins & growth are ultimately capped. They may even look like the bare metal companies of a different generation — HP, Dell, Cisco, etc., acting as substrates for compute, storage, and networking on commodity hardware.

The opportunity lies in the rich ecosystem of applications and data that developers or operators interact with to provision and manage software, which can run anywhere. And there are many wonderful ideas for this future. Lightning-quick databases for all types of data structures that actually scale. Microservices that don’t break or get tangled, which then paves the way for a progressively modular stack. Access controls and security policies represented as code. Portable development environments that transcend one person’s laptop. An observability platform that finally ties together the three pillars.

As an early-stage venture investor and optimist, I lean towards the latter view, but wrote this piece to recognize and shine light on the nuance.

I recently came across a piece from Matt Klein, the creator of Envoy (a popular open source project), where he notes:

“Providing a full PaaS is synonymous with competing directly with the major cloud providers who are attempting to commoditize “plumbing” like Envoy as quickly as possible. Without providing a full PaaS, ultimately the business boils down to providing services and support (packaging, tooling, observability, etc.) around all of the existing PaaS solutions. Either way, it’s a tough road.”

Matt could have snapped his fingers, raised a series A, and been off to the races with a startup building around Envoy. Nevertheless, he acknowledged the structural disadvantages that would have resulted with the decision. I applaud that decision, and hope more of that style of thinking comes into the ecosystem as the market continues to mature.

Announcing: RFS 100

A new collaboration with Mario Gabriele of Charge Ventures

In 2003, Tesla had nothing to do with Elon Musk.

Sitting in Disneyland's "Blue Bayou" restaurant with his wife, Martin Eberhard suggested a name for the company he'd been working on with his co-founder, Marc Tarpenning. After selling e-reader company Rocket Book in 2000 for $187MM, the two engineers had looked to bring the power of the lithium ion battery, leveraged in the Rocket device to a new vertical. They'd settled on cars.

He'd suggested dozens of names over the past few months without any luck. No matter how hard he tried, each moniker ended up sounding overly eco-minded, wonky. General Motors, automobile royalty, had just frittered $1B on the EV-1 having failed to attract buyers beyond the bohemians and tech geeks. That too weighed on Eberhard's mind.

"What about Tesla Motors?" he asked, finally.

"Perfect!" his wife responded. "Now get to work making your car."

Within a few days was registered and both Eberhard and Tarpenning were off to the races. It would be almost another year until the two would reconnect with the man they'd seen speak at the Mars Society conference a few years earlier. What began as an investing relationshop with Musk would turn into much more. In many ways, Musk was the classic overreaching investor, leading the company's Series A before jumping into the weeds to weigh in on product design and push for carbon-fiber reinforced polymer body for the vehicle. It would take another four years before Musk assumed the CEO role, taking over in 2008 only after three others — Eberhard included — had manned the ship. By that time, Tesla's "founders" had already left. A year later, Eberhard would sue Musk over assuming the "co-founder" honorific, with the trio eventually agreeing that each of them, plus two others, could use the title.

There are many lessons you might take from the story above: the importance of tracking efficiency curves (Eberhard was keenly aware of lithium ion's steady improvement), the megalomania of great operators, the necessity of a good name, the importance of execution. What I choose to take from it is this: the greatest innovator of our era did not come up with the idea for which he is best known. To achieve all that he has, even Elon needed to rely on inspiration from elsewhere.

That is a story reprised across the startup world's successes. In order to build The Facebook, Zuckerberg pilfered from the Winklevii. To superpump Uber, Kalanick required the initial insight of Garrett Camp. Or as Kalanick put it, "Garrett is the guy who invented that shit." With similar eloquence Oscar Wilde is believed to have quipped, "Talent borrows, genius steals." When it comes to deciding what to work on, tech's giants seem to ask: why not both?

But for those of us that don't have the benefit of friends like Camp or Eberhard, do not have a single Winklevoss to bait-and-switch, where should we find inspiration? It may very well be TIME TO BUILD, but isn't fair to ask what? Forging "Elon Musk's 'alien dreadnoughts,'" as suggested in the piece, may be beyond some of us. For all we know, Elon may have gotten the idea from someone else.

This is precisely the problem we are hoping to amend with a new initiative, developed with Mario Gabriele of Charge. Of course, we had to steal the name for it. Y Combinator coined the term "Request for Startups" to refer to sectors they'd like to see founders attack. RFS 100 builds on that premise. Over the past couple months, Mario and I have reached out to our favorite investors, founders, and operators to ask what startup they wish existed and why they think it's promising. We've curated the best 100 to share, 10 at a time over the next 10 weeks.

You'll hear which founder thinks there should be an "Affirm for healthcare payments," which designer sees potential in a "Barkbox for Mom & Pop shops," and which investor would like to see an "API for last-mile delivery." In the process, we hope you'll learn about new spaces, see what others in the market find interesting, and maybe even an idea for your next company.

If you have friends who you think might dig the project, shoot me a note.

RFS 100

The Hype Cycle of a Venture Career

Originally, I planned on writing a post on the tradeoffs of thematic investing at the seed stage. As I was making notes, I landed on a working theory about the career arc of a venture investor. It may follow a trajectory similar to the Gartner hype cycle with a peak and trough, followed by eventual stability.

The temptation of a top-down approach coupled with feedback from opportunistic moments factor into a larger story about the investing journey. This frame is based off my personal (and narrow!) conversations and experiences over the past few years — I’m open to feedback.


The initial learning curve has two dimensions — learning the business and learning the game. People entering the venture business aspire to be smart and ambitious, and there’s much to digest about evaluating products & markets, along with developing judgment for talent. As much as I’d like the business (supporting founders, making investments) to be the same as the job; it’s only one part. The other piece is the game — navigating the social mores of Silicon Valley, building enough context for decision making, and finding one’s sources of informational, behavioral, and analytical edge. Uncertainty operates on many levels — time & schedule, tasks, objectives, and the definition of success.

Peak of Inflated Expectations

After time and adjustment, the fog seems to clear. Guided by a peripheral vision of the market, patterns begin to emerge in pitches, investments, entrepreneurs. Due to the velocity of early-stage investing, one has interacted with hundreds (if not thousands) of companies within a couple years. One may taken an overly thematic point of view based on holes and opportunities they see in the market. I’ve noticed that most theses (including many of my own) floating around in the tech ecosystem predominantly stem from funding activity at seed and Series A. Hot deals with multiple term sheets serve as our guide into the future. It’s tempting, but fickle, ground to stand on.

Since real signals are drawn out in the future, people look to immediate proxies of success on which to judge an investor. This could be the calibre of co-investing funds, or even quality of content on Twitter. However, it is unlikely an investor has seen real liquidity early on, and successful investments primarily live on paper gains.

Trough of Disillusionment

One of the first concepts to stomach is that most startups end up not working out. But, the lesson that comes after is even harder to process. One may expect binary outcomes — you invest in a company, and it takes off like a rocketship. Alternatively, you immediately know a company will not succeed right after the investment.

In reality, randomness largely governs your portfolio. Some companies seem like they’re on a downward spiral for a few years, then miraculously turn it around. Others fly high for a long time before plateauing. It’s rarely cut and dry with exponential growth and sequential, ordered financing rounds. There may be large swaths of your portfolio that are uncertain for years. Fred Wilson has this idea of “The Second Quartile” I find fascinating. He writes,

“And then there is the second quartile that will produce roughly 15% of the returns of the fund. I find that it is this cohort of investments that is the most challenging to manage. The companies in the second quartile are usually very good companies but they lack the explosive value generation characteristics of the top quartile. They tend to have a harder time attracting top talent and financing their businesses at attractive valuations. We often do insider-led rounds for companies in the second quartile as the venture industry is hard wired to invest in the top quartile, particularly the later stage/growth investor community.”

I’m guessing this period a few years in is the make or break it moment. There’s a lot of not-so-glamorous work to be done where the investor’s reputation is at stake. As Fred points out, odds are the high flyers don’t need your help, but the other segments of the portfolio do.

The Final Segments

I’ll say less about the “Slope of Enlightenment” and the “Plateau of Productivity” because I know little about these phases. However, I’m fairly confident in a few things:

  • DPI, which measures the cash-on-cash return on investment, gives you a lot of street cred.

  • Experienced VCs tend to accept the role of chance and accept what is unknown to a greater degree than younger investors.

  • It takes about a decade to get here.

Appendix A: Finding Attachments

Due to the influx of capital into the early-stage ecosystem, it’s much harder to be a generalist venture investor than it was a decade ago. Why would the best founders want to work with you if you have a limited track record and no tangible associations? To combat this, junior investors lean towards specialization, attaching themselves to companies or ideas (in the form of market maps and theses) quickly. The potential cost is attaching oneself to the wrong markets due to the myopia mentioned above. The other danger is in setting and missing expectations. If you’ve branded yourself as a fintech investor, and you miss an iconic fintech deal — you’re caught in a tough position.

Appendix B: The Consequences of Being Right

There’s a “curmudgeon effect” within seed investing I find interesting. It’s a rare byproduct of investing at seed, but present nonetheless. A seed investor could pass on a company due to concerns about market size or belief in the venture viability of the business model. However, that business could grow top line revenue for years, keep consuming capital to grow, and become one of the hottest unicorns in the ecosystem, all with a structurally flawed model. Firms with early dollars in grow their AUM from the markups, but the skeptical investor has been right the whole time. On the subject of long feedback loops, it could take a decade for the business to fold.

It goes without saying that naysayers don’t belong in the venture business. It’s about optimism and faith in talented people, rather than trying to call the future. But, I’m sure there have been more than a handful of seed managers who have felt like Einhorn going against Allied at times.

OSS Preconditions

If physical infrastructure like roads, bridges, and canals enables commerce and the movement of goods and people in the physical world, infrastructure software is the backbone of the digital economy, which continues to eat the real economy. At Haystack, we’ve been lucky to work with defining infrastructure software companies like Hashicorp and continue to be excited about opportunities across enterprise IT domains. Infrastructure software has also emerged as a personal interest area for me, primarily inspired by the founders I’ve met shipping software across commercial, open source, and cloud.

I view infrastructure software with a broad lens — programmatically extensible and hardware-independent virtual resources that support the flow, processing, storage, and analysis of data within an organization. It provides the building blocks for developers and enterprises to launch and support software apps without re-writing the same script over and over again. Open source is a broad bucket within infrastructure that attracts capital and talent, but one can’t peg all these tools or companies the same way. As I’m still early into these explorations, the framework will iterate and adapt over time. If you have any questions or suggestions, please don’t hesitate to reach out.

The 30,000 ft. View

Years ago, investors wouldn’t touch open source software businesses. The conventional wisdom was that one could never make money selling to developers, and there were many skeletons in the graveyard that made the point. The notable exception was Red Hat (and perhaps MySQL). Flash forward to the current moment — OSS infrastructure continues to be red-hot in the private markets, despite the COVID-19 outbreak. In the middle of March, Hashicorp announced a raise at a $5.1B valuation. Confluent executed a $250M raise at a $4.5B valuation in the middle of the downturn. It might be easy to say that we live in a world in which all parts of the IT stack should be open source-first, and but odds are this is hyperbole.

We have come to a point in the hype cycle where we’ve over-rotated on the OSS business model. The product is being given away for free, so re-acquainting and understanding the long-run commercial strategy is crucial. But if it works, it really works.

Three Stages

For a genealogy of commercial OSS, I recommend Mike Volpi’s “How open source took over the world.” The tl;dr:

  • The first-generation of successful companies (a small n) capitalized on delivering almost-ubiquitous projects to the enterprise with services (installation, support, etc.) with the canonical example being Red Hat and Linux. 

  • The second generation (Hortonworks and Cloudera building on Hadoop as examples) were developed within companies and charged customers for licenses to “commercial features.” 

The current evolution of open source businesses have open core or hybrid cloud business models, which combine features of the prior two iterations. They progress to market leadership in three phases (deserving of its own post) — incubation & development, traction from individuals to teams within a company, and eventually, monetization & multi-product growth.

Surface Area

As hinted earlier, not all infrastructure software caters itself to being open source. For example, companies built on web app frameworks for programming languages (think Ruby on Rails, Python / Django) have generally not led to large outcomes. Engine Yard tried, but was unable to build an enduring franchise. 

There needs to be a lot of “surface area” on which the project can grow and adapt. In this conversation with Matt Turck of FirstMark Capital, Benchmark’s Peter Fenton acknowledges that the two core attributes for open core businesses are 1) production value and 2) a big market that would support platform status. Having production value means the tool supports applications that touch end users, that don’t just sit in development or testing environments. Investors could not have anticipated how large some of these markets would become years ago, but millions of people are continuing to go online across the world. Elad Gil comments, “In general, software markets and businesses are 10X bigger than they were 10-15 years ago. This is due to the liquidity provided by the global internet.” Data processing & storage, compute, APM & observability tend to have the core characteristics Fenton alludes to.


An open core company has more credibility if the primary authors of the project run it. They set the direction and priorities of the company in addition to providing stewardship and trust for the original project. Recent successes emerged from two buckets:

  • An internal project of a web-scale Internet company

  • From the hands of an individual product leader and influencer

Jay Kreps, CEO of Confluent, is the primary author of Apache Kafka, a distributed streaming platform. He built the first versions of Kafka inside LinkedIn, trying to optimize the LinkedIn news feed. It was initially released in 2011, but by 2014, Kreps and co-founders Neha Narkhede and Jun Rao left LinkedIn and got Confluent off the ground with the support of Benchmark. By early 2019, Confluent’s bookings had exceeded $100M

In early 2010, a developer named Shay Banon launched an open source search engine into the world called ElasticSearch. By 2012, the project had been downloaded 1.5M times and was growing at a rate of 200,000 downloads per month. By October of 2018, Elastic N.V., the company governing Elasticsearch and its associated properties, went public as $ESTC, climbing to nearly a market capitalization of $5B on its first day of trading. Within eight years, the project Banon had launched into the world from his laptop served as the backbone for a multi-billion dollar Internet franchise.

Demand Impacting Supply

In order to understand the rush of OSS businesses on the supply side, don’t look further than the architectures and IT stacks of customers that constitute demand. As Kevin Kwok writes in “Aligning Business Models to Markets,” 

As the structures of markets change, the optimal business models change with them. Business models are how we align and reconcile the markets needs with the cost and human capital required to provide them. Alignment of markets and the costs to serve them is core. And as either side changes, so to do the business models that are dominant.

The last decade saw the shift from virtual machines and monolithic applications to containers and microservices. This ongoing shift wrestles control away from CIOs, who used to make end all be all vendor choices. Enterprise infrastructure split into a distributed system — modular applications and services working across computing clusters. Individual developers can deploy open source tools on parts of the stack, as some services run like mini-applications with a data layer, networking, and client. Adopting a tool does not mean you have to go all in on day one.

By outlining these preconditions, I hope to provide some nuance and guidance on commercial open source software and its role in enterprise IT. There are a host of ways to dive deeper with OSS — the challenges of monetization (recommend Timescale’s post), providing value to individuals vs. organizations, pricing, the role of community as product managers, and more.

Loading more posts…