The person who gets recognised is not the one who knows the most about AI. It is the one who quietly made a recurring team problem smaller.
How to read this section
Find your role, then apply a filter most people skip. Do not attempt everything listed — score each use case against two questions:
- How often do I do this? Daily beats weekly beats monthly, always.
- Would anyone besides me notice the improvement? Visible wins build credibility; invisible ones build only private convenience.
Pick the one use case scoring highest on both. Start there.
Software Testing and QA
Testing is unusually well-suited to AI assistance because so much of it involves systematic enumeration — thinking of everything that could go wrong. That is precisely what language models do well.
- Generate test scenarios from a requirement or user story
- Identify edge cases a human reviewer would plausibly miss
- Improve bug reports so developers can reproduce without a follow-up conversation
- Analyse requirements for ambiguity and untestable statements
- Create test data ideas including boundary and invalid inputs
- Summarise logs to locate the relevant failure quickly
- Draft test documentation from your existing notes
Edge case generation. Paste a requirement and ask for twenty edge cases, ranked by likelihood of occurring in production and severity if missed. Then add the three you would not have thought of. Do this for a fortnight and your defect-escape rate becomes visibly better — a result your manager can see.
Software Development
The trap here is asking AI to write code you cannot review. The value is in the surrounding work that consumes real hours.
- Explain unfamiliar code in a codebase you have inherited
- Generate boilerplate so you start from structure rather than a blank file
- Review logic for edge cases and error handling you missed
- Create documentation from the code that already exists
- Suggest test cases for functions you have written
- Debug simple issues by explaining the error and the surrounding context
- Compare technical approaches with honest trade-offs for your constraints
Never commit code you could not explain to a colleague in a review. AI-generated code you do not understand is technical debt with your name attached to it.
Project Management
Project management is largely a communication role, and communication is where AI is strongest.
- Create project plans with phases and dependencies as a first draft
- Summarise meetings into decisions, actions, and owners
- Draft status updates tailored to different stakeholder audiences
- Identify risks by asking what typically goes wrong in similar projects
- Create action trackers from unstructured discussion notes
- Improve stakeholder communication, especially difficult messages
- Prepare review documents from scattered artefacts
Status updates written for three audiences from one input — engineering, leadership, client. Most PMs write one version and hope it works for everyone. It never does.
Business Analysis
BA work is fundamentally about precision of language. AI is a good adversarial reader.
- Improve requirements by finding ambiguity and unstated assumptions
- Create user stories from stakeholder conversations
- Draft acceptance criteria that are genuinely testable
- Identify missing information before it becomes a mid-sprint blocker
- Compare solutions against your stated constraints
- Create process documentation from how work actually happens
- Summarise stakeholder inputs and surface where they conflict
"Read this requirement as a developer who will build exactly what it says and nothing more. List every place you would have to guess." Cheaper to find gaps now than in UAT.
Human Resources
HR carries the highest sensitivity risk of any role here. The applications are real, but the guardrails matter more.
- Draft job descriptions that describe the actual role rather than a wish list
- Prepare interview questions targeting specific competencies
- Create onboarding material from existing documentation
- Improve employee communication for clarity on sensitive topics
- Summarise policies into something people will actually read
- Build training plans structured around outcomes
- Analyse feedback themes across anonymised survey responses
Never input individual employee data, performance records, health information, grievance details, or salary specifics. Never let AI make or materially influence a hiring, promotion, or termination decision — beyond the legal exposure, the training data carries bias you cannot audit.
Sales and Marketing
- Research customers and their industry before a first call
- Create follow-up messages that reference the actual conversation
- Analyse objections and prepare honest responses
- Prepare proposals structured around the client's stated problem
- Generate content ideas mapped to buyer stages
- Build customer personas from real patterns rather than assumptions
- Improve campaign messaging and test alternative angles
Objection handling. List your five most common objections and work through honest, specific responses for each. Unlike content generation, this improves a conversation you have several times a week.
Business Owners
For owners, the constraint is rarely ideas — it is that everything undocumented lives in your head, which caps how much you can delegate.
- Create standard operating procedures from how you currently do things
- Improve customer support responses and common templates
- Draft marketing content across channels from one core message
- Analyse customer feedback for recurring themes
- Plan offers and pressure-test the pricing logic
- Create reports from raw operational numbers
- Document workflows so tasks become delegable
SOP creation. Describe a process aloud, have AI structure it into steps, then correct the errors. Every process documented is a task you can hand to someone else — the only real path out of being the bottleneck in your own business.
If your role is not listed
The method transfers. Any role has tasks matching these four patterns, and AI helps with all of them:
| Pattern | What it looks like | AI's contribution |
|---|---|---|
| Enumeration | Listing everything that could apply, go wrong, or be tried | Systematic coverage without fatigue |
| Transformation | Same information, different format or audience | Fast, consistent restructuring |
| Compression | Long input, short usable output | Summarising without losing decisions |
| First drafts | Anything where a blank page is the obstacle | Something to react to and improve |
Common mistakes
Choose yours
🎯 Your use case
If the answer is nobody, choose a different one.
Time taken, quality, error rate, or volume produced.
What comes next
You have a baseline, a direction, a tool, a prompting method, and a specific use case. The next component sequences all of it into thirty days so you actually execute rather than plan.