Low, High, Max, Ultra... how hard should my AI think?
Effort is how long the model thinks before it talks. Low blurts, high plans first, Ultra splits the job. Match the dial to the work, and stop babysitting it every time.
· 3 min read
Writing from the Orbit team.
Effort is how long the model thinks before it talks. Low blurts, high plans first, Ultra splits the job. Match the dial to the work, and stop babysitting it every time.
· 3 min read
Haiku, Sonnet, Opus, Fable... Luna, Sol, Astra. They're sizes from two companies on one ladder. How to pick the right model for the job, and why the biggest isn't always the move.
· 4 min read
You text your friend localhost:3000 and it won't open. Why localhost only means this computer, and how hosting, deploys and DNS get your app a real address.
· 6 min read
Compare 13 ways to run parallel coding agents: platforms, isolation, review, remote machines, prices and licenses, with dated first-party sources.
· 13 min read
Set up GitHub required reviews and status checks first, then use Orbit's queue approval to review work before an agent starts.
· 4 min read
A manual checklist for verifying AI coding work, then how Orbit's inspector, Git evidence and completion ledger help track what shipped.
· 4 min read
Start with native Claude Code options, then use worktrees and tmux, add Codex, and decide when a shared queue helps.
· 4 min read
Use native instruction files first, keep shared rules in one source, and understand Orbit's ORBIT.md and context-handoff layer.
· 4 min read
Create separate agent checkouts, plan dependencies and shared services, review branch conflicts, and clean up worktrees safely.
· 4 min read
When Claude Code Remote Control is enough, and when Orbit's phone interface helps you manage a queue across agents and projects.
· 3 min read
Use Orbit Boards to keep durable plans, findings and running notes alongside the queue that executes agent work.
· 2 min read
Start with SSH, separate clones or CI runners, then use Orbit multi-node when you need one queue across machines.
· 3 min read
A practical overnight checklist for sandbox boundaries, permissions, approvals, spending limits, secrets and morning review.
· 5 min read
Your agent says done, but the page is blank. Why that happens, and how linters, type checkers and GitHub Actions catch it before you ever open the app.
· 5 min read
Most of my biggest project was written by agents. Here are the rules I'm using to keep it from falling over, and why each one exists.
· 8 min read
Everyone argues about models, but the harness around the model matters just as much. What a harness is and how to pick one by how you like to work.
· 7 min read
You ask your agent to add some AI to your app, it works on the first try, you're thrilled...
· 5 min read
Yesterday I published my article on Git, and within minutes the replies started rolling in...
· 6 min read
Back in February I didn't know what Git was...
· 6 min read
Boris Cherny, the creator and head of Claude Code at Anthropic, posted something recently that got a lot of people talking...
· 7 min read