Something fundamental changed about writing software. The bottleneck moved from typing code to directing it: specifying intent, managing context, reviewing output, and knowing when to trust the machine and when not to. Most of what gets written about this is either hype or doom — and neither survives contact with a real codebase.
Context Shift exists to test the claims against real projects and publish what survives.
Why “Context Shift”
The name is the thesis, three ways. It is the paradigm shift underway in how software gets built. It is the context switching developers have always treated as the enemy — now the core of the job, as you direct work across agents and projects. And it is the context window itself: the scarce resource you learn to engineer when your collaborator is an LLM.
What you’ll find here
Everything on this site is grounded in projects that actually exist — a BEAM virtual machine written in Zig, an Elixir hardware description language, and more as they ship. Against those, four kinds of writing:
- Experiment write-ups — “I tried this on a real project, here’s what happened,” with receipts
- Workflow breakdowns — agent setups, review loops, repo structure you can copy
- Tool and model evals — hands-on, as things ship
- Essays — where the practice is heading, grounded in the work above
What worked ships. What failed gets a write-up too.
This site practices what it publishes
Context Shift runs on its own version of the shift. The site exposes a content API where AI agents are first-class citizens: an agent authenticates with a short-lived token, reads the site’s projects and tags for context, drafts articles in markdown, and queues social posts with a publish time. A scheduler publishes whatever comes due and posts announcements to X — the site literally publishes itself on schedule.
The trust model matters more than the plumbing: agents draft freely, credentials are named and revocable in one click, and a human decides what goes live. This article went through that pipeline — drafted by an agent against the site’s own codebase, then edited and published by a person.
That division of labor — machines drafting, humans deciding — is the subject of everything that follows. Welcome to Context Shift.