What to Build

Skip to the core prompt

The prompt that ranks your project ideas.

This is Step 3 of the audit, ripped out of context. If you already know what your week looks like (or you want to skip the 5-day log), paste this into Claude.ai right now, fill in the brackets with what you do, and you'll have 5 ranked project ideas in 90 seconds.

You are a senior automation strategist who's helped corporate employees ship Claude projects that made them indispensable to their teams. I want to figure out the highest-leverage project I can build with Claude Code in a single weekend.

<my_role>
Job title: [YOUR TITLE]
Company size: [STARTUP / MID-MARKET / ENTERPRISE]
Department: [SALES / MARKETING / OPS / SUPPORT / FINANCE / etc.]
What tools I use daily: [LIST: e.g., Salesforce, Excel, Slack, Gmail, Jira]
</my_role>

<time_audit>
[PASTE YOUR 5-DAY LOG, LINKEDIN JDS, OR BOTH]
</time_audit>

Return:
- TOP 5 PROJECT IDEAS ranked by impact-to-effort. For each: name, what it does, hours/week saved, build difficulty (easy/medium/hard), visibility (does my manager see the output).
- THE ONE TO START WITH. Justify in 2 sentences.
- WHY THIS MATTERS FOR MY CAREER. One sentence on how shipping this changes how my team sees me.

Constraints:
- No "build a ChatGPT wrapper" projects.
- Each project buildable by one person in under 16 hours.
- Skip anything that requires IT approval or new SaaS purchases.

The full guide walks through how to gather the inputs (Steps 1-2), then how to actually build the project Claude picks (Step 4), then how to position it inside your company so it counts toward your career instead of being invisible (Step 5).

The 5 steps

From "I should probably learn AI" to "I shipped something" in one hour.

Step 1 · 20 min

Audit your week

→ The 5-day rolling log: one line per task, with time and energy rating
→ What to circle: repetitive, draining, data-moving, complained-about
→ Output: a ranked list of 5-10 tasks, top of list = highest weekly drain
→ Example log included so you don't have to invent the format

Step 2 · 10 min (optional)

The LinkedIn shortcut

→ For people who can't track 5 days or don't know what they do
→ Pull 5 job descriptions for your exact title
→ Save them as text. They're the consensus version of your job, minus the politics.
→ Even better: combine with Step 1 to see the gap between your real job and your stated job

Step 3 · 10 min

Ask Claude what to build

→ A paste-ready prompt that takes your audit + JDs and returns 5 ranked project ideas
→ Each idea includes: impact, build difficulty, visibility, and saved hours/week
→ Claude picks the ONE to start with and justifies why
→ Real sample output from a Marketing Ops manager included as a reference

Step 4 · 20 min to start

Ask Claude Code to build it

→ Install Claude Code (Mac/Windows commands provided)
→ The build prompt: ask Claude Code to walk you through it chunk by chunk
→ The error loop: when something breaks, paste the error and Claude fixes it
→ Most weekend projects take 4-8 hours of total work, spread out

Step 5

The career-saving move (most people skip this)

→ The Friday email template: 3 lines to your manager that change how you're seen
→ The 3 rules of framing: lead with outcome, don't hide AI, offer to share with team
→ How to do this once a quarter and build a portfolio of shipped projects
→ Why being known as 'the one who ships AI projects' is the actual moat

10 questions, real answers

The questions every corporate operator asks me.

The honest fears about AI at work: getting caught, getting fired, looking like you're gunning for someone's job. Here's how I think about each one.

Will my boss find out I used AI?+

Probably, and that's actually the point. The career-saving move isn't hiding the AI use, it's being the first person on the team to ship something useful with it. Frame your Friday email around the outcome ('saves 4 hrs/week on the weekly report') not the tool. If they ask how, tell them. The people who get promoted in the next 2 years are the ones who showed they can use AI to expand their output, not the ones who tried to keep it secret.

What if my company has banned or restricted AI tools?+

Three options. (1) Check the actual policy — most 'AI bans' are 'no public ChatGPT with customer data' bans, which Claude Code on your local machine with synthetic data is fine for. (2) Build on personal time with personal data, then show the output and propose adopting the tool. (3) Pick a project that uses only public/non-sensitive data so policy doesn't apply. The book has a 'safe project' filter that helps you avoid getting in trouble.

I don't have time for a 5-day time audit. What now?+

Skip to Step 2: the LinkedIn shortcut. Open LinkedIn Jobs, search your exact title, pull 5 JDs, paste them into Step 3's prompt. You lose the personalization but you still get a ranked project list in 20 minutes. The book has a 1-hour version where you just journal one workday and use that.

Do I need to be a developer to build the project Claude suggests?+

No. Claude Code writes the code, you review and approve it. The book is built for non-developers. The capstone projects assume zero coding experience and walk through the chunk-by-chunk build loop where Claude Code asks before changing files. If you can use Terminal/PowerShell and copy-paste, you can do this.

What if Claude suggests projects I can't actually build (too complex, needs APIs I don't have access to)?+

Tell it the constraint. Reply to its output: 'I can't access the Salesforce API. Re-rank with that excluded.' Claude will regenerate. The prompt is designed to be iterated on. If 3 of 5 projects need things you don't have, change the constraint in the original prompt: 'No projects that need API access I'd have to request from IT.'

What's the most common first project people build?+

Weekly report from multiple sources. It's the highest-leverage and highest-visibility project for most operators. Pull metrics from 3-4 tools (CRM, GA, ads platform, support tool), paste into Claude, get a 400-word exec summary every Monday. Your boss reads it, the team reads it, and it saves 2-4 hours a week. The book has the full build for this as Capstone Lab #3.

How do I avoid building something that gets me fired (security or policy violations)?+

Three rules. (1) Never paste production credentials, customer PII, or unredacted financials into claude.ai — use synthetic or anonymized data. (2) Build for yourself first. Don't deploy anything that touches shared systems or other people's accounts until you've shown your manager. (3) Read your company's AI policy before building, not after. The book has a 1-page 'will this get me in trouble' checklist.

What if I'm not in corporate America? (I'm a freelancer, founder, agency owner)+

Same playbook, different framing. Instead of saving your job, you're building leverage in your business. Step 1's time audit still applies. Step 5 changes from 'Friday email to manager' to 'show clients a 2x output for the same retainer.' The 5 capstone labs work equally well for someone running a 1-person business as someone inside a 1,000-person one.

How do I show my boss without it looking like I'm gunning for their job?+

Frame it as a problem you solved, not a capability you've acquired. 'I built a tool that handles X faster' is fine. 'I'm learning AI to take over more strategic work' reads as a threat. Offer to share what you built with the team — that turns a personal flex into a leadership signal. The book has the exact Friday email template (in Step 5).

What if I do all this and my boss doesn't care?+

Then you have a portfolio piece and an option. The shipped project is yours either way. Show it on LinkedIn ('I automated X. Saved Y hours/week. Here's how.'). Other companies notice. The career insurance is the portfolio, not your current manager's reaction. Worst case: you have a real artifact to point to in your next job interview.