← All speakers

Dax Raad

Conference affiliation: OpenCode · 2025

Dax Raad is a creator of OpenCode, an open-source coding agent that works across model providers. He builds OpenCode at Anomaly, following years of work on developer infrastructure at SST. His design choices address a recurring problem: how to give developers a straightforward starting experience while leaving room to change models, manage their own infrastructure, and configure more demanding workflows.

On this page

From startups to developer infrastructure

Dax began working professionally after high school and started a company with a friend around 2012. An acquihire led to engineering leadership, followed by startup consulting, another venture building healthcare and transportation software, and a director-of-engineering role. He initially avoided cloud-specific technology, then deliberately immersed himself in AWS.

Frustration with its tooling prompted him to experiment with his own tools before finding SST. He became a user, contributor, and investor, then joined its existing founding team around 2021. His career account connects SST’s design to small product teams whose application engineers also have to manage infrastructure.

His serverless database benchmarking explains the practical problems behind that work. In Serverless Relational Showdown, published in October 2021, Dax compared Aurora Serverless, Aurora’s HTTP-based Data API, and PlanetScale under concurrent Lambda workloads. Conventional database connection pools assume relatively long-lived processes; Lambda can generate many short-lived requests competing for connections. He weighed latency and performance spikes against operational effort. All three options were usable in the tested workload, but he preferred PlanetScale; with Aurora, he favored avoiding connection-management difficulties even at the cost of slower requests. His explanation of the Lambda Runtime API similarly made the service easier to understand by tracing the request loop through which a runtime receives work and returns results.

At SST, Dax worked with the team to simplify application development and deployment. The 2023 rewrite introduced TypeScript configuration, a faster command-line interface, and reworked local Lambda development that no longer required a separate debugging stack. Frontend support incorporated OpenNext, which converts Next.js build output into deployable packages, including packages for AWS Lambda. These were collaborative projects, with community contributions, rather than Dax’s individual inventions.

OpenAuth, another project from the SST maintainers, makes a specific separation between authentication and user management. It supplies a self-hosted OAuth server that handles authentication flows and tokens, while application callbacks perform user lookup and creation. Its ready-made interface can be customized or replaced. Applications can therefore adopt the login machinery without adopting a universal model for their users and databases.

A terminal worth spending the day in

OpenCode began with an interruption Dax wanted to remove from his own work. As a Neovim user, he disliked leaving his editor to ask a browser-based chatbot a question, paste code, and carry the answer back. Claude Code demonstrated the value of an agent with direct access to project files. Dax wanted that workflow alongside a richer terminal interface and the freedom to try other models.

His co-founded coffee business, Terminal, had already explored an unusual use of the terminal: customers order whole-bean coffee through an SSH storefront. That experience helped shape his ambitions for OpenCode’s themes and personalization—features for people who spend hours working inside the interface.

For Dax, model-provider independence also gives open source a practical purpose. A small team cannot use every model deeply enough to discover its quirks. Contributors can identify and improve the combinations they rely on daily. His account of OpenCode’s origins connects the freedom to choose models with the community work needed to support those choices.

OpenCode Zen offers a curated starting point: models and providers tested for coding-agent use. It reduces the configuration work involved in choosing among them while leaving developers able to use alternatives.

Deciding what belongs—and when to expose it

Dax’s product arguments become more revealing when he applies them to his own software. In an April 2026 conversation about product sense, he identified a weakness in OpenCode’s terminal interface: its command dialog had collected capabilities without a clear hierarchy or natural route to discovering them. Useful features had accumulated faster than the team had organized the experience.

He described a more deliberate response to requests for Git worktrees, containers, cloud sandboxes, and stronger isolation. The team recognized these as variations of a workspace concept and was developing a shared approach instead of implementing each request separately. Waiting for that understanding was uncomfortable while competitors shipped faster, but it gave the team a way to address related needs together.

Several positions explain the decisions he is trying to make:

  • Give capabilities a path to discovery. New users should reach a meaningful code change quickly; additional features need an intentional place in the experience. Dax acknowledges that OpenCode has repeatedly complicated onboarding and needed to revisit it. Adding a capability to a command dialog does not, by itself, explain when or why someone should use it.
  • Build powerful primitives beneath simple defaults. Dax favors flexible underlying components assembled into an approachable first experience. As users need more configuration, they can work more directly with those components. His AI changes Nothing talk describes the upfront design work this requires: the team has to understand its problem well enough to choose primitives that can serve both an individual hobbyist and an organization with self-hosting or data-control requirements.
  • Earn attention, then remove friction before the first moment of value. Dax treats marketing as creative work developers must learn. A feature announcement can inform existing users without giving a newcomer something worth sharing. Once someone is interested, requests for company details or tours of secondary features can delay the interaction that explains why the product exists. He says AI brainstorming has not supplied the creativity or clarity he needs for these decisions.

Dax closes that talk by acknowledging that users can find places where OpenCode falls short of his advice. The command dialog and repeated onboarding revisions give that admission substance: recognizing a design problem and fixing it consistently are different demands on a team.

Read the topics behind these talks

1 conference talk

Loading more talks and highlights…

References