Second is the world’s most capable AI-powered codebase maintenance platform. Our AI agents automate codebase maintenance work including migrations, upgrades, design system component updates, feature flag management, test generation, and more. Developers connect Second to their repo, select a maintenance module such as AngularJS to React, review the AI agent plan, and run a job. After completion, engineers review the code changes and raise a pull request.
Our primary users are engineers on horizontal application and DevOps teams seeking to streamline routine codebase maintenance.
Engineering leaders, taking a holistic approach, are poised to achieve the most significant gains through substantial improvements in organizational productivity and cost efficiency. Engineering leaders often grapple with balancing large codebase maintenance projects vs. building new customer-facing software. Currently, these maintenance tasks are carried out by human engineers, costing millions of dollars annually. By automating codebase maintenance, engineering teams can reallocate human resources to building new customer-facing software.
Second is completely free for up to 100,000 lines of code changed per year. Try it out here.
We are seeing a lot of market pull for automated codebase maintenance. Last week, we were #1 on Product Hunt, and we’ve had over 3,000 signups to date. While the majority of codebase maintenance challenges are encountered in engineering organizations at large companies, startups and small companies can also benefit from Second's automations on a smaller scale.
We’ve been building codegen AI agents since January 2023. One thing that we learned early on is that it’s absurdly easy to build something that looks like it works, but utterly fails in real life – hence why 2023 was absolutely overflowing with hype videos from startups and hobbyists. Many of these startups went on to raise a lot of capital, only to discover that the distance between hype and real production capability is quite vast.
Like many new startups entering this space, we started with the idea of building a system that could leverage AI inference to create maintenance plans, and then use AI functions to execute against those plans by making a series of codebase edits on disk. We quickly discovered that this approach results in substantial code drift and low quality code changes. Performing complex maintenance tasks like migrations, which can affect hundreds or thousands of files, only exacerbates the problem.
Why is this? Simple. Today, AI is really great at making small decisions with sufficient constraints, and it’s also great at making small tactical code changes within a single file. Unfortunately, in practice, it’s not so great at constructing complex code change plans and then translating this into high quality and reliable code changes.
So what is one to do? Enter STEP, Second’s answer to the current shortcomings of codegen AI agents. STEP stands for “Software Transformation Execution Plan”, which is a layer in between AI agent planning and AI agent execution that yields much higher quality and consistent code changes when compared to other solutions on the market.
We already have five enterprise customers in production, deployed to secure dedicated tenants. Today, our customers use Second for AngularJS to React migrations, Python upgrades, design system component updates, feature flag management, and much more.
I want to give a heart-felt thank you to our amazing investors, including Gustaf Alströmer, David Lieb, JJ Fliegelman, and Diana Hu from Y Combinator, Louis Beryl from Four Cities, Trac, Goodwater, Pioneer Fund, Goodfin, and many other amazing angels and advisors including Matt Mullenweg, Sumon Sadhu, John Bautista, Aria Kashefi, Ian Livingstone, Amir Tarighat, and Adit Parekh
We’ve come so far, but we're just getting started! We’re looking for passionate builders, hackers, and doers who love helping developers, want to make an impact in the codegen AI agent space, and looking to join a world class engineering team to make it happen. Join us!
.jpg)