AI coding prompts: write, fix and explain code

AI prompts to program They work well when they give the model the context it lacks: language, verbatim error, expected behavior, constraints. A vague prompt produces generic code that it probably does not compile. A precise prompt —even two lines —produces usable code at the first stroke in most cases.

This guide collects ready-made examples to use with ChatGPT, Claude and other models: not only for those who develop every day, but also for those who use AI to automate small tasks, write Python scripts, build formulas or explore code without being a professional developer. To understand how prompts work in general, the starting point is the guide to prompt engineering.

Why generic prompts don't work with code

"Write a function ordering a list" is a prompt that works. "Fix my code" is a time-wasting prompt. The difference lies in the context: an AI model doesn't have access to your environment, your stack, your mistakes — unless you give them to it.

The most common problems with code prompts are three:

• The language or version is missing. "Write a function" in Python 3.10 is different from Python 2.7 or JavaScript. Without this information, the model chooses itself and often misses context.
• There is no verbatim error. If you write "it doesn't work", the model must guess. If you paste TypeError: unsupported operand type(s) for +: 'int' and 'str'He knows exactly where to look.
• It is not clear what the code should do. Expected input and output are the minimum specification. Adding limit cases (what happens if the list is empty? if the value is null?) drastically reduces iterations.

If you want to deepen the basics for building effective prompts, start from the guide on AI prompt and then the one about prompt for Claude.

Table: prompts ready for most common development tasks

Each line shows the task, the prompt pattern and what to expect in output. Fit the part between [ ] to your case.

Practical examples

1. Law Firm — script to rename files automatically

A non-developer lawyer receives hundreds of documents with random names and must rename them following a convention (YYYY-MM-DD_cliente_tipo.pdf). He doesn't know how to program, but he wants to automate it with Python.

Prompt used:

"Write a Python 3 script that reads all .pdf files in a folder, asks the user to enter the name of the customer, and renames them to the AAAA-MM-DD format_customer.original.pdf using today's date. Use only standard libraries. Add a confirmation message before renaming."

Output: script working in about 25 lines, with os, datetime and input(). The lawyer ran it from Terminal without touching the code. Time to get to a usable version: less than 5 minutes.

2. E-commerce team — debugging a slow SQL query

An analyst notes that a query on orders from the last 30 days takes 8 seconds on a Postgres database with 500k lines. It is not a DBA, but it knows how to use Claude.

Prompt used:

"This Postgres query is very slow (~8s on 500k lines). Analyze the execution plan and suggest optimizations — indexes to add, rewrite, etc. Query: [SQL]. EXPLAIN ANALYZE Plan: [output]"

Output: Claude identified a full table scan on created_at (no index), suggested the index and proposed a rewritten version of the query. Time dropped to less than 200ms after creating the index. For an overview of which model to use in these cases, see best AI models for programming.

How to choose between ChatGPT and Claude for the code

Prompt for those who are not developers

You don't need to write code every day to use AI productively. These prompts are designed for those who want to automate small tasks without a technical background.

Simple Python Scripts

"Write a Python 3 script that [precise task]. Don't use external libraries. Add comments explaining each step. At the end it shows an example of how to run from terminal."

Excel Formulas / Google Sheets

"Write an Excel/Sheets formula that [goal]. The data is in column A (header: [name]), column B (header: [name]). The formula goes in column C. It explains what each part does."

Basic SQL Query

"Write an SQL query for [objective] from the table [name] that has these columns: [list columns with type]. Filter by [condition]. Sort by [column]."

No-code automations

"I want to create an automation in Zapier/Make that goes: [trigger], then [action 1], then [action 2]. What steps do I have to configure? Are there any limit cases to watch out for?"

For those who want a reference point on the use of ChatGPT in these scenarios, the guide on ChatGPT for coding goes into detail about Code Interpreter features.

Mistakes to Avoid

• Pasting code without context. "What's wrong here?" without language, error or goal is a prompt that generates generic answers. The minimum context is always: language, what needs to be done, what happens instead.
• Accepting the code without reading it. AI produces the most likely, not the most correct code. On scripts that touch files, databases or APIs, reading before executing is necessary — not optional.
• Using wrong library versions. If you do not specify the version, the model uses the most common one in the training data, which may be outdated. "usa Python 3.11 e pandas 2.x" if relevant.
• Asking for refactoring without testing. Behavior-changing refactoring is not refactoring. If you don't have any tests, first ask the AI to write a base set, then do the refactoring and check that the tests still pass.
• Sending to production without revision. For personal scripts or formulas, the risk is low. For code in production that touches customer data or transactions, a human review is required. AI for code is used as a fast contributor, not as an autonomous system. To build structured AI workflows, our team of Custom software development can help you.

How to apply it in the company

The AI for the code isn't just about developers. In a company of 10-50 people there are dozens of small automations waiting to be built: data import/export scripts, formulas in spreadsheets, queries to extract reports, automations between different software.

• Identify repetitive manual tasks involving structured data: CSV export, massive rename, format transformation.
• Parts from Python scripts or Sheets formulas: short tasks, verifiable output, low risk.
• Document the prompts that work: Create an internal collection of verified prompts for your team's recurring tasks.
• Forms those who are not developers on the basic prompts for SQL and Python — reduces the bottleneck on IT for simple requests.
• Use AI for code review: "Review this code for security or performance issues" is a prompt that also works for a code written by humans.

For teams who want to integrate AI into processes in a structured way —not just in manual prompts — Giallo Studio designs custom-made agents and automations that operate continuously and integrated with existing systems.

Conclusion

The AI prompt to program more effective share a simple structure: context (language, version, environment), precise target (input → output), constraints (libraries, performance, style), and where relevant the verbatim error message. With this basis, most daily tasks —writing a function, fixing a bug, documenting, writing tests. resolve on the first or second attempt.

You don't need to be a developer to take advantage of this: simple Python scripts, formulas, queries, and no-code automations are within reach of anyone who can build a precise prompt. The important thing is to read the output before using it. If you want to deepen how to structure complex prompts, start from prompt engineering and then explores the differences between models in the guide to Best AI for Programming.

If you want to go from testing with ChatGPT to an AI system integrated into business processes, Giallo Studio custom designed agents and automations.