April 12, 2026
If You're Only Running One Claude Code Session, You're Not Going Fast Enough
The real skill in using agentic coding is not coding at all. It's management.
Think like a manager, not an engineer
Imagine you're managing some number of engineers. They're coding. They ask you questions and you tell them what to do, what to write, or what architectural path to take. But you are not writing the code. And you are not even watching them write the code.
This is important because that's what real life is as a manager. You're not watching everybody write the code.
The difference, because these agents are not people, is that they can't test like a person.
Your job: unblock and QA
As their manager, you must unblock them. You have to be their QA. You have to work with them until their feature works. Only after the PR is created, that is when you review the code.
And just like with people, you ask for changes, or you approve it. There's nothing unique about AI-generated code. You must be the gating qualitative factor. And if the AI produced garbage, no big deal. Have it try again and improve the prompt.
There is no point in reviewing the code before the PR. You would never review the code before a PR for someone you manage, unless they asked. (Or you don't trust them, but that's a model and prompting discussion, not the point of this post.)
It becomes manageable with a process
The skill of using AI is how many sessions you can effectively pilot at once. And this becomes manageable with a repeatable process.
Just like scrum, waterfall, and agile before it, AI development processes are emerging. Prompting and planning plugins like Superpowers, GSD, spec-kit, and Compound Engineering can create the foundation for repeatable, reliable output. But those are just frameworks. The key is the prompts and instructions you and your team write.
I'm not sure how to justify working on one thing at a time anymore. And when the cost of building drops, you don't just ship more. You build better. The details you always wanted to get right, you finally can.
Resources
- Superpowers - AI-native development workflow
- GSD (Get Shit Done) - structured prompting for Claude Code
- spec-kit - specification-driven development
- Compound Engineering - multi-agent development plugin
Scape and its creator are not affiliated with or connected to any of these projects.
Built for parallel workflows
Run multiple Claude Code sessions at once.
Terminals, editors, diffs, and notes. All in context, all at once.
Download ScapemacOS only. See pricing