Reason

Software teams spend significant engineering time on tasks like tracking down bugs, writing regression tests, profiling slow jobs, and producing internal reports such as reliability reviews or spend dashboards. Today this work is done manually by engineers across multiple tools: reading code, running benchmarks, writing tests, opening pull requests, and assembling decks and CSVs by hand. The examples on the page show this ranging from a game studio chasing a double-boost bug to an engineering team needing a Q3 reliability deck with raw incident data by Thursday. Doing this by hand means context-switching between the terminal, chat, and reporting tools, and waiting on a human to be available.

Reason is positioned as a coding agent that can be invoked from a terminal, a web app, Slack, or a mobile device, and that performs end-to-end engineering work rather than just suggesting code. In the examples shown, it reproduces bugs, traces root causes, edits multiple files, writes regression tests, runs benchmarks and test suites, and opens pull requests that merge after checks pass. It also spawns sub-agents to parallelize investigation (for example separate subagents for finding callers, profiling, and writing tests), and can go beyond code to query a billing export, build and run a live dashboard, or generate a slide deck and CSVs for a reliability review. The page states it outperforms Codex and is 22% more cost efficient, citing this as its own research.