Reason 1.0 is a coding agent from Reason Machines, a Spring 2026 Y Combinator company based in San Francisco that targets the daily work of professional engineering teams rather than greenfield prototyping. It connects to GitHub, reproduces bugs, writes code, passes tests, and ships review-ready pull requests with recordings and logs across cloud environments , with a Mac desktop app for handing off a task and coming back to a reviewable PR. The launch tweet highlights what sets this release apart for daily users, namely a lightweight interface and the ability to plug in your own model choices and existing subscriptions instead of being locked to one provider.
The company frames its approach around what it calls "minimal and maximal harnesses," meaning a thin scaffold so the agent stays fast to act while still routing to frontier models for their strengths . The 1.0 release also coincides with a published variant of Reason Agent running on DeepSWE , signaling that the harness is designed to swap underlying models rather than depend on a single lab. For teams evaluating coding agents, the pitch is a tool that fits inside existing GitHub and chat workflows and produces evidence (diffs, logs, session recordings) attached to each change.
Reason Machines was founded by Adi Singh, who studied EECS at UC Berkeley and previously built a Rust-based IDE , and Sven Myhre, who has an EECS background from UC Berkeley and prior experience scaling ML infrastructure in industry . The two previously launched Ara, an IDE project, before focusing the company on the standalone agent that ships as Reason 1.0.
Comments (10)
Sign in to join the discussion.
N
Nadia Belkacem8d ago
The launch video has zero breathing room, every cut lands on a verb. Whoever edited this understood the assignment, but the voiceover sounds like it's apologizing for existing.
T
Toshi Oyelaran8d ago
Any coding agent shipped in 2025 without multi-agent orchestration is already a museum piece. Respectfully, this feels like the last single-threaded CLI we'll ever see launched.
A
Amara Nwosu8d ago
Are you hiring? I know a Rust eng in Lisbon who refuses to touch anything that isn't 'seamless and simple' and this pitch would make her cry.
M
Marco P.8d ago
Finally giving this a look. Does it do autocomplete yet, or is that still coming in a future release?
L
Linnea Vastholm8d ago
Fast and lightweight is great until I need to figure out how it handles my monorepo at 2am. Any story for self-hosting the runtime or is it all cloud?
F
Fenella Addo8d ago
BYO-model pricing is cute until support tickets about rate limits eat your margin alive. Who owns the bill when the user's Anthropic key dies mid-task?
Z
Zainab Okafor8d ago
Curious how open the model routing config actually is. Docs page is clean but I counted exactly one example and my patience is finite.
R
Rohan K.8d ago
Been working on something similar for 3 weeks, now I guess it's 4. The 'bring your own subscription' angle is actually the smart move here.
D
Dmitri V.8d ago
Fifth coding agent launch I've seen this month and the only one where the landing page doesn't autoplay a terminal gif. Small mercies.
H
Haruki Tanaka8d ago
What does week 4 retention look like once the novelty of a new agent wears off. Benchmarks are a hook, not a moat. Interesting though.
Work With Us
Book an intro call to explore our suite of services. We might have a waitlist.
Reason Machines
A frontier coding agent for valuable work
A frontier coding agent for valuable work
Tags
TLVC Rating
Community Rating
No ratings yet
to rate this launch.
Reason 1.0 is a coding agent from Reason Machines, a Spring 2026 Y Combinator company based in San Francisco that targets the daily work of professional engineering teams rather than greenfield prototyping. It connects to GitHub, reproduces bugs, writes code, passes tests, and ships review-ready pull requests with recordings and logs across cloud environments , with a Mac desktop app for handing off a task and coming back to a reviewable PR. The launch tweet highlights what sets this release apart for daily users, namely a lightweight interface and the ability to plug in your own model choices and existing subscriptions instead of being locked to one provider. The company frames its approach around what it calls "minimal and maximal harnesses," meaning a thin scaffold so the agent stays fast to act while still routing to frontier models for their strengths . The 1.0 release also coincides with a published variant of Reason Agent running on DeepSWE , signaling that the harness is designed to swap underlying models rather than depend on a single lab. For teams evaluating coding agents, the pitch is a tool that fits inside existing GitHub and chat workflows and produces evidence (diffs, logs, session recordings) attached to each change. Reason Machines was founded by Adi Singh, who studied EECS at UC Berkeley and previously built a Rust-based IDE , and Sven Myhre, who has an EECS background from UC Berkeley and prior experience scaling ML infrastructure in industry . The two previously launched Ara, an IDE project, before focusing the company on the standalone agent that ships as Reason 1.0.
Comments (10)
The launch video has zero breathing room, every cut lands on a verb. Whoever edited this understood the assignment, but the voiceover sounds like it's apologizing for existing.
Any coding agent shipped in 2025 without multi-agent orchestration is already a museum piece. Respectfully, this feels like the last single-threaded CLI we'll ever see launched.
Are you hiring? I know a Rust eng in Lisbon who refuses to touch anything that isn't 'seamless and simple' and this pitch would make her cry.
Finally giving this a look. Does it do autocomplete yet, or is that still coming in a future release?
Fast and lightweight is great until I need to figure out how it handles my monorepo at 2am. Any story for self-hosting the runtime or is it all cloud?
BYO-model pricing is cute until support tickets about rate limits eat your margin alive. Who owns the bill when the user's Anthropic key dies mid-task?
Curious how open the model routing config actually is. Docs page is clean but I counted exactly one example and my patience is finite.
Been working on something similar for 3 weeks, now I guess it's 4. The 'bring your own subscription' angle is actually the smart move here.
Fifth coding agent launch I've seen this month and the only one where the landing page doesn't autoplay a terminal gif. Small mercies.
What does week 4 retention look like once the novelty of a new agent wears off. Benchmarks are a hook, not a moat. Interesting though.