@dzhng: I found myself structuring the monorepo into microservices a lot more (even tho technically they are not microservices)…
Summary
The author discusses the challenges of scaling monorepos with AI tools like Codex, noting increased token consumption as codebases grow and suggesting that loosely coupled modules may be needed to handle AI context limits.
View Cached Full Text
Cached at: 08/24/26, 10:03 PM
Kanary’s monorepo is growing by 100,000 lines every month, and it’s already around 450,000 lines now.
Meanwhile, Codex’s token consumption is speeding up in parallel.
In other words, as the number of files the agent explores increases, the token consumption per task also goes up— which, well, is obviously just a natural consequence, if you think about it.
But having watched that growth process in fast-forward has made it hit hard on a gut level.
I wonder what’s the best way to handle this in huge projects.
Even in my own projects, since macOS and iOS share some code, I can’t completely separate the search space within the monorepo.
Given that AI context doesn’t scale up dramatically, it seems like the only option is to make modules more loosely coupled?
Similar Articles
@kenn: Kanary's monorepo is growing by 100,000 lines every month, and it's already around 450,000 lines now. Meanwhile, Codex'…
The author reflects on the rapid growth of Kanary's monorepo and the parallel increase in Codex's token consumption, suggesting that loose coupling of modules might be necessary due to AI context limitations.
@tobi: When I first told @MParakhin that we were planning to move to a monorepo he said wisely: Moving to a Monorepo is an ide…
Shopify's Tobi Lütke shares an anecdote about moving to a monorepo, quoting a discussion noting monorepos are great for AI agents but require thoughtful tooling and effort to mitigate downsides.
Anthropic says coding is no longer the bottleneck. Its monorepo hit 30,000+ merges in a day.
Anthropic claims coding is no longer the bottleneck after its monorepo hit 30,000+ merges in a day, and it migrated off Git to a Claude-built version control system called Scion. The post sparks discussion on where engineering effort shifts when AI handles coding autonomously.
Is monolith AI the future?
A discussion on whether monolithic AI systems are sustainable, drawing parallels to software monoliths and suggesting that specialized micro-service-style agents working together might be the future.
We started with 3 agents. Now I am realizing we accidentally built a tiny distributed system.
The author reflects on the challenges of managing multiple AI agents as they scale, comparing it to infrastructure issues in microservices and mentioning tools like Lyzr for central management.