It's been a dense week. Between working through a gnarly debugging spiral and rethinking how we run sprint planning, a few patterns kept surfacing — things that actually moved the needle versus things that just felt productive. Here's what landed.
1. Give AI the error and the context, not just the stack trace.
A stack trace alone gets you generic answers. Paste in the surrounding code, your assumptions, and what you already tried. The output quality jumps immediately.
2. Treat your AI prompt like a ticket, not a search query.
Vague input → vague output. Write it like you'd write a Jira ticket: background, constraints, acceptance criteria. It takes 30 extra seconds and saves 10 minutes of back-and-forth.
3. Use AI to stress-test your own thinking before you ship.
Before opening a PR, ask: "What are the three most likely failure modes in this code?" You'll catch things code review misses because reviewers assume you already thought about them.
4. Sprint planning works better when AI does the first draft.
Instead of opening a blank Confluence doc and staring at it, dump your ticket titles and goals into a prompt and let AI scaffold the plan. You edit down, not up. Way faster.
5. Ask for the boring version first.
When generating boilerplate, tests, or docs — ask for the most conventional, straightforward approach. You can always push for creative solutions once the baseline exists.
6. One prompt per problem. Don't chain requests in the same message.
Mixing "refactor this" with "add tests" with "update the docs" in one prompt splits attention and degrades output. One job, one prompt.
7. Save the prompts that work. Seriously, just save them.
Most engineers run a great prompt once and never use it again. A plain text file, a Notion page, anything — it compounds fast.
The debugging and sprint planning themes from earlier this week both trace back to the same root: AI is only as useful as the structure you bring to it. Garbage in, garbage out still applies — it's just faster now.
The full set of patterns I've been building on is in The AI Leverage Playbook: 50 Prompts & Workflows for Engineers — 50 prompts across code review, debugging, refactoring, sprint planning, ADRs, and test design. $19 at https://gumroad.com/l/nhltvo?utm_source=devto&utm_medium=article&utm_campaign=roundup_w2. Stop reinventing the prompt every time.