May 2023
From: Brian and Tobias
Subject: A collapse in the cost to produce software
When we first heard Sam Altman talking about driving down the cost of intelligence in late 2022, it wasn't obvious what this actually meant. After five months of pitches from founders automating writing, coding, designing, creating PRDs, doing customer service, and more, it has become rather obvious what reducing the cost of intelligence means. A recent McKinsey study claims, "Current generative AI and other technologies have the potential to automate work activities that absorb 60 to 70 percent of employees’ time today."
Aside from scientific discovery, i.e. superbug antibiotics being discovered with GenAI, software engineering is the most powerful place where this automation will take place. SK Ventures has talked about “AI as the next collapse dynamic in technology,” meaning that AI will drive a massive cost reduction of code itself, akin to what has happened with the cost of memory, compute, and networking. This has the opportunity to revolutionize the world.
It is unfathomable to comprehend the impact of developers being 2x more efficient writing code – 2x more product, 2x faster, 2x the stuff. From accelerating the evolution of a technology s-curve, to recalibrating build-versus-buy decisions towards building, to reducing the people it takes to build a startup, the compounding effects over time will be enormous. Adoption of AI tools among developers is already pretty breathtaking.
Although this is beginning to unfold, it won’t happen overnight. Today, there are many tools emerging focused on developer productivity that are slowly but surely contributing to this “collapse dynamic” by creating a new modern developer stack for engineers to build software. Code generation is one piece of it, but these tools extend far beyond that to terminals, front-end frameworks, streamlining of various processes and workflows, CI/CD, and more. With this newsletter, we want to highlight some of those tools.
- Copilot: Your AI pair programmer. Copilot now reports that it writes 46% of code for engineers using the tool. Initial skeptics worried that Copilot would only be helpful for remedial engineers, not the best ones. We’ve actually witnessed the opposite. Some sophisticated engineers we’ve spoken to are also using ChatGPT to generate code, either as a replacement for Copilot or as an additional tool for different sorts of problems. Interestingly, we are hearing some engineering leaders express security concerns about using Copilot because the tool could suggest code with security flaws in it. We expect enterprise-grade deployments to become more common, with explicit rules and SLAs around these sorts of risks. There are also other startups emerging like Tabnine competing with Copilot on this exact value proposition.
- Replit: An instant integrated developer environment (IDE). Replit makes it super fast and easy to start coding in a custom developer environment. Replit also serves as a code collaboration platform and cloud service, provisioning the infrastructure you need to get started coding immediately. Replit has developed an AI tool – Ghostwriter – that integrates directly into the IDE for proactive debugging and assistance.
- Warp: A modern terminal. Warp is a Rust-based terminal built from the ground up to be more performant and collaborative than any terminal before. The terminal has never been a cloud-based, collaborative experience, so Warp helps developers move much faster by working together more seamlessly. Warp has also incorporated an AI product into its suite, offering AI embedded into the Warp terminal.
- Vercel: Develop and ship frontends faster than ever before. Vercel is the commercialized version of the next.js framework, a frontend web development framework. Vercel makes configuration and infrastructure integration totally seamless for the front-end developer. In a recent conversation, an engineer described Vercel as “an almost unthinkably easy tool for frontend.”
- Gitpod: The fastest way to spin up dev environments. Gitpod offers a read-to-use cloud developer environment (CDE), preconfigured with all the tools and libraries that a developer needs to get started writing code and building products.
- Graphite: Review code faster and simpler than before. Graphite enables developers to “stack” larger tasks into bite-sized, easily reviewable parts. Doing so enables engineering teams to conduct code reviews faster than before.
- Streamlit: Build and ship data apps with just a data script, no front-end engineering expertise required. Snowflake acquired Streamlit in March for $800M. More recently, Streamlit has developed tooling to help users easily build Generative AI apps, a natural extension from its origins of helping users build data apps. That product has taken off in recent months.
- Buildkite: Easy to build and secure CI/CD workflows. The Buildkite agent runs on your own infrastructure so that your source code never leaves your environment and you can move as quickly as possible. Of course, lots of other CI/CD tools exist, but Buildkite stuck out in conversation with developers.
- Wolverine: Python script leveraging GPT-4 to automatically heal code and fix errors automatically. It has the ability to analyze, diagnose, and heal broken code.
- Refact: A coding assistant for VS Code or JetBrains, with a self-hosted option. Refact does more than just code completion and offers a more comprehensive suite of tools, including existing code analysis and bug detection.
- Cursor: AI-first code editor. Cursor is partnering with OpenAI to create an editor designed for pair programming with an AI counterpart.
- Cloudflare One: Cloudflare has released a new product focused on helping teams build with AI technologies like LLMs while staying secure. Features of the new tooling include measuring usage of models, controlling API access to an AI service, and restricting data access and uploads.
- Plural: Easily deploy and manage open source infrastructure, with all the Kubernetes and IaC work done for you beneath the surface. Plural makes DevOps workflows way easier and removes the headache out of Kubernetes management. Note: Primary is an investor in Plural.
There is a lot more to come that will continue collapsing the cost to produce software. An area of particular interest for us is the intersection of AI agents and developer velocity. We recently saw one pitch enabling companies to create their own agents for specific company jobs, like a recruiter, a PM, or a marketer, depending on the specific customer’s needs and domain context. It is only a matter of time before we see platforms that enable engineers to enlist coding agents with specific expertise – vulnerability testing, DevOps integrations, performance optimization, etc. This AI will elevate, not replace, the power of engineers.
Until next time,
Brian and Tobias