Agentic coding, from idea to project.

Agentic coding means giving an AI assistant a development task that takes several steps: inspect the project, make a change, use tools and check the result. In JacCode, that work happens alongside your local files and a saved coding conversation.

A question and a project use different amounts of work

“What does this function do?” can often be answered by inspecting a small piece of code. “Fix this bug and check the affected behaviour” may require searching files, reading several modules, editing code and running tests. Each step can add more model work and context.

Agentic work is useful when a task needs that inspect-change-check loop. You still give the assistant a concrete goal and review the resulting files and tool output.

A typical multi-step coding task

  1. Inspect. Find the relevant files and understand the existing behaviour, project setup and commands.
  2. Change. Create or patch the code needed for the requested result.
  3. Check. Run appropriate development commands or tests, then inspect their actual output.
  4. Review. Look at the source changes and the assistant's final explanation. Revisit an earlier step if the evidence shows more work is needed.

In Jac Code, expandable activity collects intermediate updates and tool calls. Permission requests and questions stay visible when input is needed. During a running task, Send becomes Stop.

Choose how the assistant can act

Read only lets the assistant inspect and explain without editing files or executing commands. Ask first requires permission for applicable edits and commands. Full access permits work within the application's workspace policy.

These modes let you match control to the task. Start with inspection for an unfamiliar project, then allow changes when the intended outcome and relevant files are clear. Source viewing, change review and activity timelines keep the work inspectable.

Why usage allowance matters

Input, generated output, reasoning and background model work contribute to usage. Independent work delegated to subagents shares the same account allowance. A task that repeatedly searches, edits and tests can consume substantially more usage than a short question.

Pro 5x is the recommended JacCode plan for agentic coding, at $20 USD per month. Pro 20x is $80 USD per month and has the highest weekly allowance. Both use flagship JCoder 5.6 and have weekly allowances without a separate five-hour limit.

Free offers JCoder 5.6 Flash only with an entry-level allowance. The plan comparison shows the current models, usage windows and prices together.

Give a task the assistant can check

Include the outcome you want, the behaviour that is wrong today, relevant examples and any constraints. For a bug, explain how to reproduce it. For a feature, describe what should happen when someone uses it. Point out the project's build or test instructions if you know them.

For example: “Find why this form loses its draft after a failed save, fix the affected flow, and use the project's checks to verify that the draft stays available.” This gives the work a concrete trigger and an observable result.

Start with the Windows installation guide, or explore JacCode's unrestricted AI coding workspace.