
Hey everyone, we’re Adi and Sven, co-founders of Reason Machines 👋
TL;DR We built the most efficient agent harness, so teams can maximize their AI usage at up to 30-80% lower cost per task than Codex, Cursor, or Devin. Sessions run on cloud VMs that move across your devices, use your existing keys and subscriptions, and work directly in Slack.
Since launching 4 weeks ago: 20,000 users, 14,000 X followers, and 3M+ impressions.
https://www.youtube.com/watch?v=LVRSTgnaexg
AI models are finally good enough for real work. The bottleneck is no longer capability—it’s cost: how much work you get per dollar of inference.
That cost is set by the harness more than the model: the same model can cost up to 5× more depending on the harness. Today, teams are stuck choosing between simple chat bots that can’t touch infrastructure, and closed cloud platforms that lock you into one vendor’s models and markup.
We designed a minimal harness for the most work per token, with the power of a cloud VM and the ease of a personal bot:
A minimal harness (bash + async code execution) beats labs' bloated harnesses on cost per task without sacrificing capability. On the Coding Agent Index, Reason scored 65.1 vs Codex’s 62.5, at 22% lower cost per solved task. Read the full benchmark write-up.
Adi and Sven studied EECS at UC Berkeley. Adi worked in applied ML at Scale AI and previously built a Rust-based IDE; Sven built AI models at a Fortune 100 company.
More AI usage, for less cost.