Which path
is right for you?

Learn what OWASP Top 10 for LLMs is. Read the list, understand each risk, and pick one to research deeper. Write a one-page summary of prompt injection in plain English.

Run your first prompt injection test on a public AI chatbot. Document 3 inputs that changed the model's behavior unexpectedly. Write up what you found.

Build a simple threat model for an AI system you use at work. List the trust boundaries, user inputs that reach the model, and top 3 risks.

Draft a one-page AI usage policy for your team. Cover: what tools are approved, what data cannot go in, how decisions get reviewed. This is your first portfolio piece.

Use Claude to summarize 3 recent research papers in your specialty area. Compare the summaries to your own reading. What did it get right? What did it miss?

Draft patient education material for one condition using AI. Review it with the same standard you'd apply to any clinical content — accuracy, tone, reading level.

Map one admin task you do weekly. Identify whether it involves PHI. If not, design an AI-assisted version with a clear human review step.

Create a one-page "AI use checklist" for your unit. What's allowed, what requires review, what's off limits. This is your first real governance artifact.

Pick the most repetitive task you do every week. Write down: what triggers it, what steps you take, what the output is, and who reviews it. This is your automation blueprint.

Use Claude to handle one step of that task. Copy and paste real inputs, review the output, refine the prompt until it's consistently good. Do not automate yet — just test.

Sign up for n8n (free). Connect two tools you already use — for example, Gmail to Google Sheets — and have Claude process something in the middle.

Add a human review step to your automation before any action is taken. Test it five times. Document what it does, how to turn it off, and what to do when it fails.

Read the Anthropic agentic AI docs. Understand what an agent is: goals, tools, planning, memory, output. Sketch a simple agent for a task you'd want to automate.

Build a single-tool agent using the Claude API. Give it one tool (e.g., web search or file read). Test it on 5 different inputs. Document where it fails.

Add a second tool and a planning loop. The agent should decide which tool to use based on the task. Add a step that logs every action it takes.

Add a human approval gate before any irreversible action. Set up basic Sentry monitoring so you can see every run. This is your first production-ready agent pattern.

Learn the Role-Goal-Context-Format template. Apply it to 5 tasks you do every week. Compare the output to what you got before with vague prompts. Notice the difference.

Pick your best prompt from week 1. Test it 10 times with different inputs. Write down every time it gives a bad result and why. Improve until it's consistent.

Build a prompt for a longer task using chain-of-thought: tell the AI to "think step by step" before answering. Compare the results to a prompt without it.

Create a prompt library document with 10 tested prompts. Include: what each prompt is for, the exact text, example inputs and outputs, and known limitations.

Create a free Claude account. Ask it to explain something from your job that you find confusing. Spend 30 minutes just exploring — ask weird questions, see what it can and can't do.

Use the Role-Goal-Context-Format template to write 3 prompts for your actual work. Compare the results to just typing a question. Write down what was better and what was worse.

Find one weekly task that AI could help with. Test it for a full week. Keep track of where it saved time and where it made mistakes. This is data, not failure.

Document your experiment. What was the task, what prompt did you use, what did AI do well, what did you have to fix? This write-up is your first portfolio piece.