The work stays close to implementation. AppCubic cares about systems that survive contact with real workflows: LLM orchestration with clear tool boundaries, evaluation loops, data movement, product surfaces people actually use, and the operational choices that make a prototype worth continuing.
What AppCubic builds around
The studio draws on several streams of work: e-commerce and logistics systems, advisory for early teams, LLM decision support, camera-based motion understanding, and research tooling. What ties them together is not a vertical. It is the habit of turning uncertain technical work into something a team can operate.
Open source
Much of the studio's agent tooling is public. App Automaton is the open-source side of AppCubic: portable skills for coding agents, document and web tooling, multi-agent orchestration, and pure-MLX models for Apple Silicon, all MIT licensed on GitHub.
How the work is framed
AppCubic treats AI as infrastructure for decisions and workflows rather than as a decorative feature. Useful systems need retrieval and context design, honest evaluation, clear points for human review, and the judgment to know where automation should stop.
Where to go next
Research systems live at research.appcubic.com. Adjacent product and operations work connects through Futur. Personal background is at Benji's site.