{"id":128746,"title":"Reason: the most efficient agent harness","tagline":"Run hundreds of parallel sessions across cloud VMs using your own keys and subscriptions at 30-80% lower cost","body":"Hey everyone, we’re Adi and Sven, co-founders of Reason Machines 👋\n\n**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.\n\nSince launching 4 weeks ago: **20,000 users**, **14,000 X followers**, and **3M+ impressions**.\n\n<https://www.youtube.com/watch?v=LVRSTgnaexg>\n\n---\n\n## Problem\n\nAI 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.\n\nThat 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.\n\n---\n\n## Solution\n\nWe designed a minimal harness for the most work per token, with the power of a cloud VM and the ease of a personal bot:\n\n* **Cloud VMs that follow you:** Runs across cloud Linux, cloud Mac, local machines, or SSH boxes without losing state.\n* **Swarm-scale parallelism:** Run hundreds of sessions in parallel to ship a feature, or let workflows run for days.\n* **BYO keys &amp; subscriptions:** Plug in your API keys, open weights, and Grok, Copilot, or Kimi subscriptions with auto-failover.\n* **Slack-native:** Mention @Reason in a thread to get back a PR, a deck, or an answer.\n\n---\n\n## The harness\n\nA 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](https://reasonmachines.com/blog/introducing-reason-agent-1-1).\n\n![Coding Agent Index against cost per task, with Reason highest at about $4 per task, beside a comparison of Reason finishing a task in 3 model calls and a call-per-tool harness in 7](/media/?type=post&id=128746&key=user_uploads/3131989/e05783ce-ae24-4e0a-adf3-ec631ee6f965)\n\n---\n\n## Team\n\nAdi 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. \n\n**More AI usage, for less cost.**\n\n![Adi and Sven, cofounders of Reason Machines](/media/?type=post&id=128746&key=user_uploads/3131989/d3d601dc-1b55-4959-9b8b-83f57b071b17)\n\n---\n\n## Our ask\n\n* **Try it:** [start on the web](https://reasonmachines.com/login) or add @Reason to Slack. YC companies get $1,000 in credits via our Deal.\n* **Intros:** We’d love intros to teams hitting AI usage limits or paying heavy bills for Codex, Cursor, or Devin.\n* **Investors:** We’re raising our seed round—reach us at [founders@reasonmachines.com](mailto:founders@reasonmachines.com).","slug":"XUY-reason-the-most-efficient-agent-harness","created_at":"2026-11-13T18:32:28.780Z","updated_at":"2026-11-17T18:50:53.652Z","total_vote_count":30,"url":"https://www.ycombinator.com/launches/XUY-reason-the-most-efficient-agent-harness","share_image_url":"//bookface-static.ycombinator.com/assets/ycdc/yc-og-image-c440a0ad1dacfb86eeeb343717479cc54d256614449b4ef719977a0a451f8bc8.png","company":{"id":31458,"name":"Reason Machines","slug":"reason-machines","url":"https://reasonmachines.com","logo":"https://bookface-images.s3.amazonaws.com/small_logos/ac8d1fbd9622030ddfab1cc9a6fac1907350004f.png","batch":"Fall 2026","industry":"B2B","tags":["Developer Tools","Enterprise Software","Cloud Computing"],"search_path":"https://bookface.ycombinator.com/company/31458"}}