Bridging the semi-async valley of death
Faster Inference Won't Save You: Part 4
Blog
Engineering deep-dives from the Graphcoder team.
Compaction is a symptom, not a strategy
Faster Inference Won't Save You: Part 2
Faster inference won't save you
A comparison of how coding agents handle inference, infrastructure, and multi-agent scaling — and where the industry is getting it wrong.
Fewer tool calls, not quicker ones
Faster Inference Won't Save You: Part 1
Git is the wrong abstraction for multi-agent scaling
Why we built copy-on-write symlink worktrees with jj-style auto-rebase instead of using git for multi-agent coding.
Plan mode is a crutch
Faster Inference Won't Save You: Part 5
Scaling intelligence without building the model
The next wave of AI scaling is not about bigger models. It is about model alloys, context management, and infrastructure that multiplies any foundation model.
Your agent is fast. Your plumbing isn't.
Faster Inference Won't Save You: Part 3
Your agent shouldn't need to ask
How permission-scoped auto-approval and escalation chains let coding agents work autonomously without sacrificing safety.
Graphcoder: the first step towards hyperscale coding agents
Launch post