How to start using artificial intelligence at work in 30 days

This practical one-month plan, requiring just 15 to 20 minutes a day, breaks down AI adoption into four stages, using real everyday tasks. The framework is based on academic research and user guides from tech companies.

Why start with rules and testing

A study by Harvard Business School, in collaboration with consulting firm BCG, tracked 758 consultants across 18 realistic tasks. In activities within GPT-4’s capabilities, users who employed the tool completed tasks 25.1% faster. However, in tasks beyond its capabilities, users were 19 percentage points more likely to deliver incorrect solutions.

The authors describe this as an uneven technological frontier: AI performs well in some tasks but poorly in others of similar difficulty. That’s why the plan includes time for testing and verification before expanding its use.

The week-by-week plan

Week 1 (Days 1–7): Rules and first contact

Choose a tool and read the company’s usage policy. Compliance experts warn that personal data entered in prompts may be subject to the LGPD; therefore, customer information, contracts, and internal documents require careful handling. Start with low-risk activities, such as rewriting an email, summarizing a public text, or explaining a technical term.

Week 2 (Days 8–14): Prompts

A prompt is the written instruction given to the AI. OpenAI’s guide recommends clear instructions, examples, reference text, and breaking down complex tasks into steps. Practice with three real tasks, specifying the audience, format, and objective. Example: Write a friendly collection email to a customer, up to 100 words, and propose two payment dates.

For those looking to start learning AI, Saint Paul’s Pre-MBA in Artificial Intelligence offers a practical introduction through four online classes for R$37.

Week 3 (Days 15–21): Map of successes and failures

Test the AI in various routine activities and record the following in a simple table:

• the requested task
• the time spent with and without the tool
• the errors found
• the necessary adjustment

A Microsoft study, conducted in partnership with Carnegie Mellon and involving 319 workers, found that greater trust in AI was linked to less critical thinking, reinforcing the need for verification. Tools like Google’s NotebookLM respond based on user-provided sources and include citations of excerpts to help verify accuracy.

Week 4 (Days 22–30): Routine and measurement

Select two or three recurring applications, such as meeting summaries or report drafts, and save the prompts that worked. Compare the time spent with the traditional method to measure actual gains. If possible, share the best prompts with your team.

Limitations and next steps

This plan is a study guide, not a method validated by research. The Harvard and BCG study used GPT-4 in 2023, and tools change frequently. The LGPD guidelines are based on expert analyses, and the company’s own policy takes precedence over any general rule.

After 30 days, the expected outcome is a record of tasks where AI helped, tasks where it made mistakes, and an initial library of prompts to expand usage more safely.