THE THREE DISCIPLINES AROUND THE MODEL

HOW THE THREE DISCIPLINES DIFFER AND WORK TOGETHER Here is a compact way to compare them. Prompt engineering asks: what should I say to the model right now? Context engineering asks: what should the model be shown at this step, given everything that has happened? Harness engineering asks: what system should surround the model so that thousands of steps, across many runs, stay reliable, safe, and affordable? They also differ in how they are fixed and tested. A prompt problem is usually fixed by editing words. A context problem is usually fixed by changing what is selected, summarized, or retrieved. A harness problem is usually fixed by changing code, such as adding a retry (trying a failed step again), tightening a permission, or setting a budget. Though different, they cooperate. A well-built harness assembles the context, which includes the carefully engineered prompt, calls the model, checks the output, takes the permitted action, records everything, and loops. If any one is weak, the whole suffers. A brilliant prompt cannot rescue an agent whose harness lets it run forever, and a perfect harness cannot rescue an agent whose instructions are confusing. A skilled team invests in all three, and as a harness engineer you are the person who sees how they fit together.