Assembling @droid, the frontier software development agent. Available to anyone, with any model, in any IDE/terminal.
curl -L app.factory.ai/cli |sh47K followers
TLVC Rating
Hook
Editing / Creativity
Copy
Sentiment of launch
Distribution strategy
Community Rating
★★★★★
No ratings yet
Your rating
Sign in to rate this launch.
About
Factory Router is a model routing layer that sits inside Factory's Droid coding agent, automatically picking which underlying model handles each session instead of leaving that choice to engineers. It targets enterprise engineering teams running Droid across coding, refactors, bug fixes, documentation, and investigation work, where defaulting every task to the most expensive frontier model has driven AI bills up without a matching lift in output. The launch matters now because token spend on coding agents has become a real line item for engineering orgs, and Factory is making the case that routing can hold quality steady while trimming roughly a quarter of that cost.
According to Factory's own benchmarks, Factory Router cuts cost by 20-25% while pass rate holds at 99% of Opus on Terminal-Bench 2 and 96% on Legacy-Bench . It automatically selects the right model for each task and routes across providers if an endpoint degrades , with failover, reserved enterprise throughput, and US-hosted open-source options for teams that want a cheaper or more controlled path. Admins can also shape the routing policy with guidance about which workflows are routine and which need deeper reasoning, so the automatic selection reflects how a given org actually ships code. The product is shipping in private research preview.
Factory was founded by Matan Grinberg and Eno Reyes , and the Sequoia-backed company has been building an enterprise-focused autonomous software engineering platform since 2023 , with Droid available across IDEs, the terminal, and CLI. Router slots in as the cost and reliability layer beneath that agent, useful for buyers evaluating how to scale agent usage across a large engineering team without writing a blank check to a single model provider.
Tags
Product launch500K-1MB2BGlobalSeries BAI-generatedUSVertical AI
Comments (14)
Sign in to join the discussion.
Priya Mahadevan21h ago
25% cost cut is a nice headline but what's the retention curve look like once teams realize they can DIY a router with two if-statements? Curious how sticky this gets past month three.
Tomas Bregenz21h ago
"Your software Factory powered by Droid" reads like a Scooby Doo episode. Try: "One agent. Every model. Right one, every time."
Priya Mahadevan20h ago
Also worth asking: does the 25% include the markup Factory takes on top, or is that gross savings before your cut?
keiko_l21h ago
Third router launch in this category this month and the only one that didn't bury the savings number in a footnote. Small win, take it.
Bjorn Aaltonen20h ago
25% cost cut on inference is cute but who eats the routing overhead, you or me? Margin math gets weird when the cheapest model is also the chattiest.
Hank Ozawa20h ago
"Frontier performance while cutting costs" is the AI-era version of "have your cake and eat it." Show me the eval suite or it didn't happen.
Chidi Okonkwo20h ago
Is the routing logic open or is this a black box I have to trust on vibes? Asking because my staff eng is already drafting the "why not just use a switch statement" doc.
Yusra Halabi20h ago
Routers are the last manually-configured layer before the agents start picking their own models mid-thought. Enjoy this UI while it lasts.
Marta Voss20h ago
The router diagram in the video has the arrows misaligned by like 3px and now I can't unsee it. Otherwise the palette slaps.
Aïcha Belkacem20h ago
Love the launch cadence but the thumbnail is just the word Router on a gradient. You spent how long on the agent and 4 minutes on the cover?
Ravi Subramanian20h ago
Can I override the router per-request via header, or do I have to trust it on the hard tasks? Also rate limits per underlying model or aggregated?
Hana Park20h ago
Tweet's pulling decent engagement but the video drop-off at the 8 second mark suggests the "automatically" promise should've been frame one. Hook is buried.
sage_ofelia20h ago
Every abstraction layer is a bet that the thing below it won't change. Bold bet in November 2025.
Dmitri Larsson20h ago
Is the demo using cached responses or live routing? Latency numbers are conspicuously absent and that's the only metric I actually care about.