Prompt engineering

Prompt engineering is the practice of writing instructions that get an AI model to produce what you actually want. It covers what you ask, how you phrase it, what examples you include and what format you request the answer in.

Why it matters

The same model can give a vague answer or an excellent one depending on how it’s asked. Early on, the difference was dramatic, and people who learned the tricks got far more out of AI than everyone else. Companies ran training programs and wrote internal prompt playbooks.

Models have since become much better at working out what people mean, and many of the old tricks matter less. What hasn’t changed is that a model can only work with what you give it. A clear goal and the right background still beat clever wording.

What still works

  • Say what done looks like. “A one-page summary for the CFO, with the three biggest risks first” beats “summarize this.”
  • Give the context a colleague would need. Who the audience is, what was already tried, what constraints apply.
  • Show an example of the output you want, when format matters.
  • Ask for the reasoning or the sources when you need to check the answer.

Prompt engineering is not the same as context engineering

Prompt engineering puts the work on the person asking. Context engineering puts it on the system: it makes sure the model already has the right documents, instructions, tools and history before anyone types a word. For one person writing one request, prompting is what matters. For a company that wants everyone to get good results, including people who will never read a prompting guide, context matters far more. Agents make this sharper, since an agent writes most of its own prompts as it works.

Bring your agent. Bring your team.