November 2022
From: Brian and Tobias
Subject: Spawl, AI, & Cloud
Time flies. We’re back with our November edition of our infrastructure newsletter, where we share reflections, trends spotted, opportunities to collaborate, and ways for this group to help us.
We have three core themes to touch on this month: Sprawl, AI, and Cloud. Let’s dive in.
Sprawl:
We have seen a pattern of companies tackling the problem of engineering “sprawl” in one way or another. Sprawl can mean many things, and we’ve seen different flavors of this theme. Repo sprawl, microservice sprawl, model sprawl. The point remains consistent for all: engineering work has become compartmentalized. Establishing, maintaining and monitoring the connections between the compartments has become unmanageable. Amidst amazing capacity and efficiency gains, the sprawl has become untenable.
In the past month or so, we have had the following conversations/experiences:
There are several reasons why now could be the time for this sprawl to begin to get contained. Perhaps it has just become too much to handle, and we’ve reached a breaking point. Alternatively, in a tight macroeconomic environment in which engineering resources are not as plentiful, perhaps activity needs to be organized and consolidated. We think there will be opportunities to elegantly mitigate all this “sprawl” at large engineering organizations.
AI: the good and the bad
AI continues to be an area we explore heavily. In today’s environment, it’s hard not to, as so many founders are capitalizing on the advances in language and foundation models to build new, more intelligent applications. We’ve been thinking about AI from a few different angles.
Skepticism: Like all trends with massive amounts of hype, the AI wave requires a little cold water of skepticism. Namely, we worry about lack of defensibility with many of these applications – the models on which they are based are open source and available to other developers, so they end up not representing any “secret sauce” or moat. Even when we spoke to an investor in a generative AI application that recently raised north of $100M, he flagged a big risk being that the open source models would improve so much over time to eventually remove any sort of product advantage the business has.
Platform shift: With that said, there is reason to be genuinely optimistic and excited about AI as the next big platform shift. Scale Venture Partners recently wrote an article on foundation models being the new public cloud that we liked a lot. The power of these models creates a truly horizontal and public domain capability to build net new things, be them better versions of old software or totally new innovations we haven’t even considered yet. As a result, we think this hype is real. This is not just an interface shift like audio and AR/VR, but rather a step change increase in the power of computers.
Point solutions vs. infrastructure: One of the implications of these dynamics is that at Primary, we’re looking for infrastructure plays as opposed to point solutions when it comes to generative AI. There will likely be point solution, application-layer winners, or at least ones that scale quickly over the short-to-medium term, but we think the longer lasting opportunities will create the piping to properly use different models within the enterprise and get the most possible out of these technologies. This mirrors cloud, which brought forth a bunch of application winners, but also many infrastructure winners helping companies navigate and build on top of a new compute paradigm.
Ethics: One final thread we’ve been pulling is ethics in AI. We think that in the future, AI models will be regulated when they’re used to interface with customers or make decisions on behalf of customers. This is already starting to take place in the EU. Although too early to predict what these laws will look like, we think there could be an opportunity to create workflow tooling for startups to more easily navigate these regulations – perhaps something like “Vanta for AI.” We’ll see if this comes to fruition in a few years.
Problematizing cloud ubiquity:
We recently gathered a group of dev tools and infra founders for a conversation on monetization. It included Paul Dix from InfluxDB, Michel Tricot from Airbyte, Merrill Lutsky from Graphite, Jeremiah Lowin from Prefect, Sam Weaver from Plural, Graham Neray from Oso, and Zach Lloyd from Warp. It was an awesome group and discussion.
One of the key themes that emerged from this conversation was that cloud isn’t the end all, be all for everything; it problematized cloud ubiquity for your data. One phrase that emerged was: on-prem data plane, with a SaaS control plane. This is pretty self-explanatory. You want your data housed on-prem in your infrastructure, with all the workflow and management layer tooling as a SaaS solution.
There are lots of implications of this. One might be that solutions need to emerge for storing and manipulating data in new machines, specifically edge devices. This coheres with DuckDB’s rise and MotherDuck’s big funding announcement. It also implies, for us, that cloud is not necessarily king (see here), but engineering leaders have gotten used to the benefits coming from a managed cloud service. Perhaps there will be opportunities to provide analog services in non-cloud contexts in the future.
As always, please feel free to share feedback, thoughts, and reactions that emerge from this distribution.
Happy early holidays to everyone,
Brian & Tobias