How to hand coding work to AI agents.

To delegate coding to AI agents and get back work you can use, you need three things: a clear brief, a repository the agent is allowed to change, and a way to see what it is doing while it works. Scopeology gives each project its own cloud agent, streams every step to your Agents page and lets you stop or rerun a job from your phone. This guide walks through the brief, the hand-off and the review.

Pick work an agent finishes well

Agents do their best work on jobs with a clear finish line that a test or a quick look can confirm. Good first jobs:

• Fix a bug you can describe in one sentence, such as a signup email that goes out from the wrong address.
• Add a small feature to an existing screen, such as a second payment option on checkout.
• Change copy, links or prices in many files at once.
• Write missing tests for a module, then make them pass.
• Update a dependency and fix what breaks.

Keep the big calls for yourself: a new data model, a pricing change, anything that needs a judgment about your business. Break those into steps and send the steps.

Connect the code first

A cloud agent works only on code you connect. Open the project in Scopeology and pick its repository, from GitHub or Doclish. From then on the agent for that project can read the code, run its tests and propose changes, and nothing outside that repository. See integrations for what connects and security for how each account is kept separate.

Write the brief like a contractor would want it

The agent reads your words as the whole job. A good brief answers four questions:

1. What should change. "The signup email comes from noreply@. Send it from team@ instead."
2. Where it lives, if you know. "The template is in the email folder."
3. What must stay the same. "Keep the design. Do not touch billing."
4. How you will know it worked. "The email tests pass and the preview shows team@."

You can say all of that out loud. Scopeology turns what you say into the written brief the agent receives.

Watch the work as it happens

Every job shows up on your Agents page with a live log: the files it reads, the commands it runs and what it decides along the way. You do not have to watch, but you can, from a laptop or your phone. If the log shows the agent heading the wrong way, stop it.

Then rerun it with a sharper brief:

Because the agents run in the cloud, the job keeps going with your laptop closed. The cloud and laptop agents comparison covers when that matters.

Review before anything ships

When the job finishes, Scopeology tells you, out loud if you are talking to it. Before you accept the change:

• Read the last lines of the log: what changed and whether the tests passed.
• Open the change in your repository and scan the files it touched. A one-line fix that touched twelve files deserves a closer look.
• Try the result yourself on a preview or a test account.

Put a review on your calendar for bigger jobs, so the work does not sit waiting. One sentence does both:

Keep a short list of jobs in flight

Two or three jobs across your projects is plenty to review well. Ask what is running whenever you want the list, and track the follow-ups as tasks with the rest of your week in projects.

Results vary. No specific outcomes are guaranteed.