How We Cut Development Time by 90 % with Claude Code: An 8-Step Playbook
Why I Wrote This Guide
a time ago our team needed a new feature in the checkout flow—coupon stacking.
The old way took four weeks from idea to production.
Today the same work ships in three days, and every newcomer runs the process solo by week two.
Nothing in this article is theory.
Every command, timing estimate, and checklist item comes from our logbook.
Feel free to copy-paste and adapt.
Table of Contents
-
One-Page Overview of the 8-Step Flow -
Step-by-Step Walkthrough -
Frequently Asked Questions -
Printable Checklist -
Next Moves
1. One-Page Overview
Step | Tool(s) | Deliverable | Typical Minutes |
---|---|---|---|
1. Plan the feature | ChatGPT | Draft spec | 30 |
2. Break into tasks | ChatGPT / Gemini | Ticket list | 20 |
3. Log tickets | Linear (or Jira/Trello) | Descriptive tickets | 10 |
4. Auto-write tech plan | Claude Code + Linear MCP | Markdown plan per ticket | 5 |
5. Parallel coding | Claude Code + Git worktrees | Independent branches | Concurrent |
6. Auto-create PR | Claude Code | Pull request | 2 |
7. First-pass review | Claude Code | Inline comments | 3 |
8. Manual test & merge | Human | Merged mainline | As needed |
2. Step-by-Step Walkthrough
2.1 Plan the Feature with ChatGPT
Start a plain-language conversation.
Me: “We want coupon stacking at checkout. Give me goals, limits, and acceptance criteria.”
ChatGPT returns something like:
-
Goal: let users apply more than one coupon per order -
Limit: only “money-off” coupons, not shipping coupons -
Acceptance: 5 automated tests + 2 manual walkthroughs
Paste the reply into a file named spec-draft.md
.
That is your living document; update it whenever scope drifts.
2.2 Break the Work into Tickets
Send the draft back to ChatGPT or Gemini:
“Split this into 5–8 tasks that fit inside one working day each.”
Typical output:
-
Add coupon_stack
table -
Checkout UI: list stackable coupons -
Price engine: multi-coupon logic -
Unit tests: edge cases -
End-to-end test: real checkout -
Update API docs
Do not over-polish; clarity beats perfection.
2.3 Log Each Task in Linear
-
Title ≤ 50 characters -
Description field: paste the acceptance criteria from step 2.2 -
Label: coupon-stack
(easy filtering later)
2.4 Generate a Technical Plan with Claude Code
Install the Linear MCP connector once:
npm install -g @linear/mcp
Then ask Claude Code for a plan:
claude-code task-plan --from-linear TICKET-123
Claude Code produces plan.md
inside the ticket directory:
-
Files to touch -
Code snippets for each change -
Rollback notes
Embed the file straight into the ticket’s “Technical Plan” field so reviewers see it at a glance.
2.5 Code in Parallel with Git Worktrees
Create an isolated workspace per ticket:
git worktree add ../feat-coupon-ui TICKET-124
git worktree add ../feat-coupon-engine TICKET-125
Open each folder in its own terminal tab and run:
claude-code dev --plan plan.md
Claude Code writes, tests, and commits inside that worktree without touching others.
When CI passes locally, move to the next step.
2.6 Auto-Create the Pull Request
Inside the finished worktree:
claude-code pr-create --title "feat: support multi-coupon stacking"
Claude Code:
-
creates branch feat/TICKET-124
-
pushes to origin -
opens the browser with a prefilled PR description
2.7 First-Pass Review by Claude Code
Post the PR number:
claude-code review --pr 88
Expect three kinds of comments:
-
Style: naming, formatting -
Logic: null checks, off-by-one errors -
Performance: N+1 queries, duplication
Address anything red; yellow items are optional polish.
2.8 Manual Test & Merge
-
Pull the branch locally -
Run the full test suite -
Perform a real checkout on staging -
If green, click “Merge” in GitHub/GitLab -
Remove the worktree:
git worktree remove ../feat-coupon-ui
3. Frequently Asked Questions
Q1: Do I need a paid ChatGPT plan?
No. The free tier works; you may wait a few extra seconds for answers.
Q2: Is Claude Code free?
It is currently in open beta at no cost, but you must request access. After approval you receive a token; authenticate with claude-code auth <token>
.
Q3: What exactly is Linear MCP?
A small wrapper around the Linear API. Install it once; Claude Code detects it automatically.
Q4: Will Git worktrees corrupt my repo?
No. Worktrees share the same .git
directory. When you delete a worktree, the branch and commits remain unless you explicitly remove them.
Q5: Juniors struggle with AI-generated code. Any tips?
Hand them the plan.md
first. Once they understand the intent, the code itself becomes readable. We schedule a 30-minute walkthrough on day one; after that new hires ship solo.
4. Printable Checklist
Print and tape beside your monitor. Tick each box before moving on.
-
[ ] Draft spec with ChatGPT -
[ ] Break spec into ≤ 8 tickets -
[ ] Tickets live in Linear, each with clear acceptance -
[ ] Claude Code generated plan.md
for every ticket -
[ ] Git worktree created per ticket -
[ ] Local tests pass -
[ ] Claude Code opened the PR -
[ ] No red comments left in Claude Code review -
[ ] Manual end-to-end test on staging passes -
[ ] Branch merged, worktree removed
5. Next Moves
-
Pick a small, real feature from your backlog. -
Walk through the eight steps above and record your own timings. -
Post your results in the comments; we’ll troubleshoot together.
See you in production—three days from now.