Agent
How to explain my work well to AI?
The way to create an AI that knows me best begins with explaining my work well.
In the current era, No-Code environments have broken down the technical barriers of coding. However, for creating your own AI agent, the most important task remains.
It is the task of logically translating your work.
In the last post, we covered the core of AI agents and task automation is the ability to explain my work, not coding.
A simply smart AI agent is different from an AI agent that knows you best.
Today, Palette, which aims to be an AI No-code Builder, offers three suggestions on the best way to describe your work.
1. "Assuming they roughly know" doesn't work, unraveling judgment criteria into words.
Most of the tasks we repeat every day are processed by a gut feeling embedded in our bodies, without the need for explanation. Instructions like “with a similar feeling as before” or “moderately clean” work between colleagues working with me, but they become the cause of errors for AI.
The beginning of explaining my work to AI is bringing out what criteria I have been judging by in my subconscious. You must organize the logic of judgment, such as the criteria for this information being important and filtering out that information, the criteria for prioritizing, and which side to decide on when it is ambiguous, into words like “in this case, do this.”
2. Specifically deciding the forms of ingredients and deliverables
If the judgment criteria are organized, now it is time to specifically decide the shapes of the ingredients (Input) that the AI will handle and the deliverables (Output) it will produce. You must not speak vaguely like 'organize this data.' You must accurately decide what data to give and what results to receive.
The ingredients mean clarifying from where the data presented by the AI was brought and up to where it should refer, so that unnecessary information does not squeeze in; the deliverables mean deciding in advance what form it should come out in, such as a table, a list, or a specific format, and the items that must be included. The more specific this criterion is, the more an agent that can be used for work right away is made.
3. Deciding in advance what not to do and exceptional situations
Expertise comes from knowing 'what not to do' rather than 'what to do.' You must set clear prohibition criteria so that the AI does not judge the work arbitrarily and make up wrong answers (hallucination). Those prohibition criteria must be set by me, the field expert.
Preventing it from using specific expressions, or making it answer 'No data' instead of making it up when information is insufficient, are examples of that. When you even decide in advance how the AI should behave in unexpected situations, a business partner that you can finally trust and leave things to is completed.
The process of explaining my work to the AI is, paradoxically, a process of properly looking back into my work again. If you can unravel the work one by one and piece it back together, the No-code Builder will become the most powerful weapon to expand your business.
Experience the Unexperienced
Palette designs the experience itself, of how naturally a practitioner can explain their work and transfer it into an immediately executable agent, rather than how sophisticatedly they implemented AI technology. Designing the AI Agent so that it can easily understand even if you explain the entire process of deconstructing and reassembling work in the easiest language, that is what Palette is currently focusing on.
Do you need answers for customized AI construction and work innovation of a different dimension? Talk with Palette.
📩Contact us : hello@pltt.xyz