Dr. Neil's Notes
Software > Development > AI in Delivery Systems
Thin-slices at a time
Introduction
Large AI prompts often fail because they are too broad and try to cover too many aspects and features. People chase the one-shot outcome, where a single prompt does all the work they need delivered, as a holy grail. Perhaps the holy grail of laziness. Thin-slices keep risk low and validation fast by delivering one narrow workflow at a time. It also helps to keep token consumption low and the cost of work manageable. Thin-slices are a way to learn quickly and adapt to change.
Pick a thin slice
Choose one task, and one measurable outcome. Keep the scope small enough to complete quickly. This makes the output clearer to see and understand. Between thin-slices, start a new session, do not pull the context across. This keeps the model focused on the current task and avoids confusion from previous work.
Design the workflow
Define success criteria, failure criteria, and a stop condition before starting. It is important to become accountable for interpreting results. The point is not to prove AI is good, it is to discover whether this change delivers value. Each thin-slice should move the needle on a measurable outcome, and you should be able to see that change, and understand why it works or does not work. If the thin-slice fails, revert to the previous state and try a different approach.
Tests First Was Always A Good Idea
The first step in any thin-slice workflow should be to write tests. This ensures that you have clear success criteria and can quickly identify when something breaks. Tests act as a safety net, allowing you to experiment with confidence and iterate rapidly. Create the tests in an agent session and review that they are testing the right thing. Then create a new agent session for the implementation, and run the tests to validate the outcome. If the tests fail, analyse the results, adjust your approach, and try again. This iterative process helps to refine the workflow and ensures that each step delivers tangible value. The tests also continue to serve as validation for future changes, providing a mechanism to ensure that improvements maintain the quality over time. By prioritizing tests first, you establish a culture of quality and accountability, which is essential for successful delivery of any system.
Part of the AI in Delivery Systems series.