AI code agents are unlocking software on demand. Already, people with no technical expertise can bring their app ideas to life without ever touching a code editor.
Since January, we've seen standalone prompt-to-app platforms grow at unprecedented rates, and now, SaaS providers are starting to integrate coding agents directly into their products. Dashboards, data visualizations, and workflows that used to be feature requests are being generated on demand. But this is just the beginning.
As autonomous codegen takes off, an unprecedented amount of code will need to be generated and stored. Infrastructure built for humans must give way to systems optimized for the models that use them.
At Relace, we're building the models and infrastructure to power software on demand. That means small, specialized language models on optimized infrastructure that make code agents faster and production ready out-of-the-box.
To accelerate this mission, we've raised a $23M Series A led by Andreessen Horowitz (a16z), with participation from Matrix Partners and Y Combinator.
Infra for Coding Agents
With the current race to dominate the coding market, dozens of teams are producing very similar implementations of coding agents. Each is independently rebuilding the same toolsets, sandboxed execution environments, and source control systems to power their products.
In web development, we no longer see databases, deployments, or authentication as a competitive edge anymore — these are services we outsource. As software on demand moves from developers to fully non-technical users, agent design choices currently considered differentiators will become managed services as well.
This infrastructure requires unique tradeoffs and must be co-optimized alongside models for performance and speed.
At Relace, we're already seeing significant gains for companies like Lovable, Magic Patterns, and Orchids by tackling both. e.g. We can accurately surface the necessary codebase context for agents in ~1-2s with our specialized embedding & reranker models, and we merge file edits at over 10,000 tokens per second with our apply model.
Small Models Unlock New Use Cases
As generating software becomes easier and cheaper, code generation will extend deeper into organizations for all industries. Designers will create mock-ups in code directly, marketers will create graphics with code, and professional service workers will create slideshow presentations by chatting in natural language with a code agent.
Like phone cameras set the stage for Instagram, faster and smaller coding agents will also fundamentally redefine what is possible. We are already seeing social platforms like Gizmo, where users share creative web apps generated with GPT-5/Gemini Flash. If made small enough, these models could run on-device directly and be essentially free to inference.
Moore's law always has unexpected consequences, and we're excited to power the surprises it will bring in code generation.
What's Next
So far, we've shipped three standalone models: embed, rerank, and apply. These models have been called as tools tens of millions of times by coding agents across 40+ prompt-to-app companies.
With the Series A, we're rolling out the public beta of our first infrastructure offering with deep integration to our models — Relace Repos.
Over the next six months, we'll be investing heavily in deploying utility agents that can run on any Relace Repo to perform critical tasks like search, merge conflict resolution, and codebase refactoring.
We're building a scrappy AI research team that will ship models the best product teams rely on. If you're a researcher (Physics, Math, CS, ML), or a talented software engineer, join us!
The future of software is on demand - and we're building the platform to power it.