At twenty-five, I can ask AI to draw the sound of a note. When the pattern appears on screen, I want to try ideas I would have given up on until yesterday. Would I notice if, in making it clearer, the filter had erased the very start of the note?
What are we looking at?
The move from a god complex to impostor syndrome can happen in the time it takes to ask that question. A moment ago, I felt I could build anything. Now I risk giving the machine credit for the idea too. Yet the experiment has a value I can test with someone who understands the problem. A junior engineer with enough inventiveness can enter a new field and build a product that stands up to those made by people who have worked there for years. It is the possibility Felix Rieseberg describes in Hunger beats tenure.
An agent can suggest different equations for combining two sound waves. We can start with a predictable result. At the same point and instant, two equal and opposite contributions sum to zero. If the program represents that sum and shows a maximum, something is wrong. A filter calls for another judgment. A steadier pattern can be easier to read and less sensitive to changes in sound. We have to decide what we want to show.
Algorithms and system design help us reason about alternatives even when changing languages. Object-oriented programming (OOP) principles can highlight a problem when changing the pattern’s colour requires touching the microphone code. That’s a reason to rethink how responsibilities are divided. These are tools for exercising critical judgment in an unfamiliar domain and justifying decisions.
Who decides the trade-off?
In an agricultural field, retrying a transmission can get a reading through and use more battery. The energy cost can be estimated. The value of receiving the data an hour earlier needs the person working in the field. What decision would they make if they already had that reading? AI helps compare options, while priorities need to be agreed with the system’s users. A creative solution might vary transmissions according to need instead of looking for the best fixed setting.
Decisions can become reusable experience. An up-to-date copy of a data store also reproduces deletions, while a backup preserves an earlier state. That distinction suggests a test for the next project, recovering the data on a test copy. If included in an agent’s instructions, it becomes part of its harness, the environment of instructions and tools built around the model. The reason behind the rule remains available and needs reconsidering when the context changes. This work is discussed by Anthropic and OpenAI.
I can feel unprepared and still choose to build. Foundations help identify where to test a solution and what to ask people who know the field. AI makes it possible to bring something that already works into that conversation instead of leaving the idea on paper.