AGENCYSCRIPT
CoursesEnterpriseBlog
👑FoundersSign inJoin Waitlist
AGENCYSCRIPT

Governed Certification Framework

The operating system for AI-enabled agency building. Certify judgment under constraint. Standards over scale. Governance over shortcuts.

Stay informed

Governance updates, certification insights, and industry standards.

Products

  • Platform
  • Certification
  • Launch Program
  • Vault
  • The Book

Certification

  • Foundation (AS-F)
  • Operator (AS-O)
  • Architect (AS-A)
  • Principal (AS-P)

Resources

  • Blog
  • Verify Credential
  • Enterprise
  • Partners
  • Pricing

Company

  • About
  • Contact
  • Careers
  • Press
© 2026 Agency Script, Inc.·
Privacy PolicyTerms of ServiceCertification AgreementSecurity

Standards over scale. Judgment over volume. Governance over shortcuts.

On This Page

Mistake One: Withholding Context the Model NeedsWhy It HappensThe Cost and the FixMistake Two: Asking Too VaguelyWhy It HappensThe Cost and the FixMistake Three: Accepting Code Without Reading ItWhy It HappensThe Cost and the FixMistake Four: Trusting Output You Cannot VerifyWhy It HappensThe Cost and the FixMistake Five: Endlessly Patching a Broken ThreadWhy It HappensThe Cost and the FixMistake Six: Over-Constraining Before UnderstandingWhy It HappensThe Cost and the FixMistake Seven: Ignoring Your Own ConventionsWhy It HappensThe Cost and the FixFrequently Asked QuestionsWhich of these mistakes is the most damaging?I do all of these sometimes. How do I stop?Is hallucination a mistake on my part or the model's?Does using a better model eliminate these mistakes?Key Takeaways
Home/Blog/Seven Ways Developers Sabotage Their Own Code Prompts
General

Seven Ways Developers Sabotage Their Own Code Prompts

A

Agency Script Editorial

Editorial Team

·March 31, 2023·8 min read
prompting for code generationprompting for code generation common mistakesprompting for code generation guideprompt engineering

When prompting for code goes wrong, people tend to blame the model. The model is slow, the model is dumb, the model hallucinates. Some of that is real, but in most cases the disappointing output traces back to a handful of avoidable mistakes in how the request was made. These mistakes are predictable, and once you can name them you stop making them.

What follows are seven failure modes that show up again and again, from beginners and experienced developers alike. For each one, we cover why it happens, what it actually costs you in time and quality, and the concrete corrective practice that replaces it. None of these fixes are exotic. They are small habit changes that compound.

If you read only one section, read the first. Missing context is the root cause behind more bad output than the other six combined.

Mistake One: Withholding Context the Model Needs

The most common mistake is treating the model as if it can see your project. It cannot. It knows only what you type.

Why It Happens

In your head, the whole project is present—the file structure, the existing functions, the conventions. It is easy to forget the model has none of that. You write "add a method to the user service" and the model invents a user service that has nothing to do with yours.

The Cost and the Fix

The cost is code that does not fit, which you then rewrite, often more slowly than if you had written it yourself. The fix is to paste the relevant existing code and state the environment every time it matters. The step-by-step process makes context-gathering an explicit step so you cannot skip it.

Mistake Two: Asking Too Vaguely

A close cousin of missing context is the vague request that leaves the model guessing about basic decisions.

Why It Happens

Vagueness feels efficient. "Write a function to parse the file" is fast to type. But it hides a dozen unstated decisions: which file format, what to do on malformed input, what shape to return.

The Cost and the Fix

The model fills every gap with a guess, and its guesses rarely match your intent, so you iterate repeatedly to claw back to what you meant. The fix is to specify inputs, outputs, edge cases, and format up front. Specificity is not extra work—it is work you would have done during the failed iterations anyway.

Mistake Three: Accepting Code Without Reading It

The most dangerous mistake is copying generated code straight into your project without reading it.

Why It Happens

The code looks right. It is syntactically clean, well-formatted, and confident. That polish creates false trust, and the temptation to just run it is strong, especially when it appears to work.

The Cost and the Fix

Models produce plausible code that is subtly wrong: a nonexistent function, a security hole in input handling, an edge case silently mishandled. These slip into production when nobody reads them. The fix is non-negotiable: read every line before running it. This single habit catches the majority of serious errors. The best practices guide treats this as the foundational discipline.

Mistake Four: Trusting Output You Cannot Verify

Related to reading but distinct: assuming code is correct because it ran without errors.

Why It Happens

Running without errors feels like passing. But "it executed" and "it is correct" are different claims. Code can run perfectly while computing the wrong answer.

The Cost and the Fix

Logic errors that pass execution are the hardest bugs to find later because nothing flags them. The fix is to test against real inputs, including edge cases, and to read the logic rather than trusting the absence of a crash. Ask the model for tests too, but review them—a test that asserts the wrong behavior is worse than no test.

Mistake Five: Endlessly Patching a Broken Thread

When the model gets something wrong, people often keep correcting it in the same conversation long past the point of usefulness.

Why It Happens

Each correction feels like it should be the last one. Sunk cost keeps you in the thread, adding patch after patch.

The Cost and the Fix

Once a thread accumulates a wrong assumption, the model tends to keep reproducing it, and your corrections fight against its own earlier output. You burn time going in circles. The fix is to recognize the pattern—the same error twice—and restart with a fresh prompt that includes what you learned. A clean start usually beats the tenth patch.

Mistake Six: Over-Constraining Before Understanding

Some developers swing the other way, packing the prompt with implementation demands before they understand the problem.

Why It Happens

It feels rigorous to specify everything: use this algorithm, this data structure, this pattern. But premature constraints can box the model into a worse solution than it would have found on its own.

The Cost and the Fix

You get exactly the suboptimal approach you specified, missing a cleaner one the model knew. The fix is to constrain behavior and quality, but leave implementation open unless you have a real reason to dictate it. State what the code must do and what standards it must meet, then let the model propose the how.

Mistake Seven: Ignoring Your Own Conventions

Generated code that works but does not match your project's style creates a slow, accumulating cost.

Why It Happens

The model defaults to common patterns from its training, not your team's patterns, unless you tell it otherwise. Naming, error handling, file organization—all drift toward generic norms.

The Cost and the Fix

A codebase full of subtly inconsistent code is harder to read, review, and maintain, and the inconsistency compounds with every generated contribution. The fix is to show the model an example of your conventions and ask it to match them. The examples article demonstrates how a single reference function aligns the output.

Frequently Asked Questions

Which of these mistakes is the most damaging?

Accepting code without reading it, because it lets every other error reach production undetected. Missing context produces the most bad output, but the read-before-running habit is the last line of defense that catches everything else.

I do all of these sometimes. How do I stop?

Pick one mistake and fix only that for a week. Most people start with reading every line, because it has the highest payoff and requires no extra prompting skill. Once it is automatic, add the next habit. Trying to fix all seven at once rarely sticks.

Is hallucination a mistake on my part or the model's?

Both. The model produces hallucinations, but supplying real context and reading the output are your responsibility, and together they neutralize most of them. You cannot stop the model from sometimes guessing, but you can stop the guesses from reaching your code.

Does using a better model eliminate these mistakes?

No. Better models reduce raw error rates but do not fix vague prompts, missing context, or skipped review. The mistakes here are about process, not model quality, which is why they persist across every generation of tools.

Key Takeaways

  • Missing context is the root cause of most bad output—paste relevant code and state the environment every time.
  • Vague requests force the model to guess; specify inputs, outputs, edge cases, and format up front.
  • Read every line before running it; this single habit catches the majority of serious errors.
  • Code running without errors is not proof it is correct—test against real and edge-case inputs.
  • Stop patching a confused thread after the same error twice; restart with a cleaner, better-informed prompt.
  • Constrain behavior and quality but leave implementation open, and show the model your conventions so the output fits.

Search Articles

Categories

OperationsSalesDeliveryGovernance

Popular Tags

prompt engineeringai fundamentalsai toolsthe difference between AIMLagency operationsagency growthenterprise sales

Share Article

A

Agency Script Editorial

Editorial Team

The Agency Script editorial team delivers operational insights on AI delivery, certification, and governance for modern agency operators.

Related Articles

General

Rolling Out AI Hallucinations Across a Team

Most teams discover AI hallucinations the hard way — a confident-sounding wrong answer makes it into a client deliverable, a legal brief, or a published report. The damage isn't just to the output; it

A
Agency Script Editorial
June 1, 2026·11 min read
General

Case Study: Large Language Models in Practice

Most teams that fail with large language models don't fail because the technology doesn't work. They fail because they treat deployment as a one-time event rather than a discipline — pick a model, wri

A
Agency Script Editorial
June 1, 2026·11 min read
General

Thirty-Second Wins Breed False Confidence With LLMs

Working with large language models is deceptively easy to start and surprisingly hard to do well. You can get a useful output in thirty seconds, which creates a false confidence that compounds over ti

A
Agency Script Editorial
June 1, 2026·10 min read

Ready to certify your AI capability?

Join the professionals building governed, repeatable AI delivery systems.

Explore Certification