Jayendra More highlights how Claude’s plain text instruction model reveals common pitfalls in AI task handling, emphasising the importance of clear workflow understanding and stable environment for effective utilisation.
Jayendra More’s account of how he uses Claude begins with a simple but revealing test: he asked the model to draft a PRD for an OTP login flow after already listing seven failure cases himself, only to see it return fourteen. The extra cases, he says, were the kind teams often miss in practice: expired codes, repeated retries, SIM swaps and interrupted sign-ins. The point is not that the model was unusually clever, but that it exposed a pattern many users reverse. They try to teach the tool their way of working before understanding what it can already catch on its own.
That distinction matters because, as More explains, a Claude skill is not a piece of software in the usual sense. It is a plain text instruction file that Claude reads before it begins work, setting out how the user wants tasks handled in ordinary language. In effect, the user is shaping the model’s behaviour through written guidance rather than through code. That makes the process lightweight, but it also makes it easy to approach in the wrong order, by overfitting the instructions before the underlying workflow is clear.
The wider lesson is that Claude is most useful when it is allowed to supplement human judgement rather than imitate a person from the start. In More’s framing, the model can act as a writer, a product manager or a coach only after the user has identified what should stay consistent, what should be delegated and where the model is expected to challenge assumptions. That is especially relevant for structured work such as product documentation, where missing edge cases can create real downstream failures.
It is also a reminder that the practical friction around Claude often sits outside the prompt itself. Anthropic’s Claude Code documentation separates troubleshooting into login problems, OAuth errors, token issues and usage-limit warnings, while other user guides describe browser interference, email verification failures, interrupted conversations and upload problems. Taken together, they suggest a familiar pattern: useful AI workflows depend not only on better instructions, but on stable access, clean authentication and a working environment. For many users, that is the difference between a tool that feels intelligent and one that simply fails at the wrong moment.
Disclaimer: This content is intended for informational purposes only. Readers are advised to exercise their own judgement, conduct due diligence, or consult a qualified expert before acting on any information provided.
