[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"$fUUSFQBuoGp7ZSsMynyGv1vxtfF_JMS1ocvl8giCr0H4":3},{"item":4},{"id":5,"idKnowledge":6,"idDomain":7,"idCluster":8,"kindOverride":9,"slug":10,"title":11,"description":12,"bodyMarkdown":13,"bodyHtml":14,"author":15,"date":16,"createdAt":17,"topics":18,"image":26,"hasDownload":27,"fileName":28,"youtubeId":29,"domainCrumb":30,"clusterCrumb":33},"331","36E9DF81-22BC-0A41-8E24-7A97BD46360B","8F2761C8-348C-C649-BC16-18822CE2D198","841C795D-B3B8-C74B-B84C-8780C2D44C8A","article","how-to-review-ai-generated-code","How to review AI-generated code","A practical, step-by-step guide to reviewing AI-generated code in FileMaker and other business systems before it reaches production.","Your developer just pasted a script into FileMaker that Klai or Copilot wrote in ten seconds. It runs. It looks clean. Nobody on the team fully understands why it works, and it's about to touch your invoicing data. That moment — approve it, or dig deeper? — is the new daily reality for anyone building business software, and it's the reason \"review AI-generated code\" has quietly become one of the most important skills an in-house dev team can have.\n\nThis article walks through exactly how to review that code so it doesn't quietly break your business six months from now.\n\n## Why can't you just trust code that runs without errors?\n\n\"It runs\" and \"it's correct\" are two very different things, and AI-generated code is especially good at hiding the gap between them.\n\nA FileMaker script generated by an AI assistant to calculate order totals might work perfectly on the five test records your developer tried — and then silently miscalculate the total the moment a record has a null quantity field, a negative discount, or a line item from a deleted product. The script doesn't error out. It just returns a wrong number that ends up on an invoice, and nobody notices until a customer complains three weeks later.\n\nThis is the core risk with tools like Klai (Claris's AI assistant built into FileMaker) or general-purpose AI code assistants: they optimize for \"produces working output on the input I described,\" not \"handles every input this business will actually throw at it.\" The AI has no idea that your warehouse team occasionally leaves a quantity field blank, or that your sales team applies discounts that can push a total below zero. Only someone who knows your business does.\n\n## What should you actually check in an AI-generated script or module?\n\nTreat every AI-generated piece of code — a FileMaker script, a custom function, a web app snippet, an API integration written with FmBetterforms or similar tooling — as a first draft from a very fast, very confident junior developer. Here's the review checklist that catches the problems that matter:\n\n1. **Trace every input path, not just the happy path.** What happens with a null value, an empty string, a zero, a negative number, or a duplicate record ID?\n2. **Check error handling explicitly.** Did the AI wrap risky steps (API calls, file writes, find requests) in error checks, or does it assume everything succeeds?\n3. **Look for hardcoded assumptions.** AI code often hardcodes today's field names, table structures, or business rules ($5% discount) that were true in the example but not in your actual schema.\n4. **Verify it matches your naming and structuring conventions.** Inconsistent naming across a solution is how six-month-old scripts become unmaintainable.\n5. **Check for redundant or dead logic.** AI assistants sometimes generate extra conditional branches or duplicate calculations \"just in case\" — these bloat scripts and confuse the next developer.\n6. **Confirm it performs at scale.** A script tested against 50 records in a demo file can behave very differently against 500,000 records in production — nested loops and unindexed finds are common AI blind spots.\n7. **Check security and data exposure.** Does the generated code log sensitive data, expose an API key in plain text, or skip privilege checks it should have?\n8. **Re-read it for readability.** If a human developer can't explain what a script does in one sentence, rewrite it — even if it \"works.\"\n\n\u003Cimg src=\"\u002Fapi\u002Fknowledge\u002Finline-image\u002F313?w=700&f=webp\" alt=\"checklist next to a code script with a magnifying glass\" loading=\"lazy\" class=\"w-full sm:w-1\u002F3 sm:float-left sm:mr-7 mb-5 rounded-2xl border border-[#E8E8ED] bg-[#F5F5F7]\" \u002F>\n\n## Does the review process change for FileMaker specifically?\n\nYes — a little. FileMaker's low-code layer means AI-generated logic often lives in three places at once: a calculation formula, a script step, and a custom function. Reviewing FileMaker code well means checking all three together, because AI tools frequently generate a calculation that duplicates logic already sitting in a custom function elsewhere in the same file — which means the next schema change has to be made twice, and inevitably only gets made once.\n\nWith Klai specifically, since it's context-aware of your existing schema, it's tempting to trust its suggestions more than a generic AI chatbot's. Don't. Context-awareness reduces some errors (wrong field names, for instance) but doesn't eliminate logic errors, and a confident, schema-correct suggestion is arguably more dangerous than an obviously wrong one, because it looks more trustworthy.\n\n## How do you review AI-generated code when you're not a developer yourself?\n\nMany business owners and IT managers approving these changes aren't reading the script line by line — and that's fine, as long as the review process compensates for it. Ask your developer or vendor three questions before anything goes live:\n\n- **\"What happens if this input is missing or wrong?\"** A confident answer with specific examples (not \"it should be fine\") is the signal you're looking for.\n- **\"Did you test this against real production-like data, not just a demo record?\"** AI code that only sees clean sample data almost always misses edge cases that live data contains.\n- **\"Can you explain what this does without reading the code?\"** If your developer can't explain the logic in plain language, they haven't actually reviewed it — they've just glanced at it.\n\n## What's a practical review workflow for a team using AI code generation daily?\n\nA lightweight process beats no process. Here's one that works well for small in-house teams and outside development partners alike:\n\n1. AI generates a first draft of the script, calculation, or integration code.\n2. The requesting developer runs it against at least three deliberately \"nasty\" test cases: empty fields, boundary values, and duplicate\u002Fconflicting records.\n3. A second person (peer review, even informal) reads the code and checks it against the checklist above.\n4. Anything touching financial data, customer records, or external APIs gets tested in a staging copy of the file before touching production.\n5. The final version gets a short comment block explaining *why* it works this way — future developers (and future AI assistants) will read that comment before they read the code.\n\nThat last step matters more than it sounds like it should. Six months from now, someone — possibly you — will need to modify this script without remembering the original AI conversation that produced it.\n\n## FAQ: reviewing AI-generated code\n\n**Does using AI to write code make a FileMaker solution less reliable?**\nNot inherently — but it removes a safety net that used to exist by default. When a developer writes every line themselves, they naturally think through edge cases as they go. AI-generated code skips that thinking process, so it has to be added back in deliberately, during review.\n\n**Is it safe to let AI write API integration code?**\nOnly with extra scrutiny. Integration code (connecting FileMaker to an ERP, accounting package, or webshop) fails in ways that are hard to notice — a silently dropped record, a duplicate sync, a timeout that isn't retried. Always test integrations against real transaction volumes before trusting them unattended.\n\n**How much time should code review actually add to a project?**\nFor most FileMaker scripts, a proper review adds 15–30 minutes, not hours — because you're checking a specific list of known failure patterns, not re-deriving the logic from scratch. The time it saves later, when a bug would otherwise reach production, is far larger.\n\n**Should smaller businesses skip formal code review to save time?**\nNo — smaller businesses often have less room to absorb a costly mistake (a wrong invoice, a broken sync with their only warehouse system) than a large company would. A lightweight, consistent review habit matters more here, not less.\n\nAI-assisted development is changing how quickly FileMaker solutions, custom web apps, and system integrations get built — our [broader look at how low-code and AI-assisted development change business software](https:\u002F\u002Floggix.com\u002Fen\u002Fblog\u002Fhow-low-code-and-ai-assisted-development-change-business-software) covers that shift in more depth. But speed only helps if what gets shipped is actually correct, and that still depends on a human who understands the business checking the AI's work before it goes live.\n\nIf your team is generating more code with AI than it can confidently review, that's usually a sign the underlying FileMaker solution, integration, or workflow needs a closer look — not just a faster reviewer. Loggix helps businesses build and audit custom FileMaker systems, connect them to other software through solid API integrations, and bring AI tools like Klai into a workflow in a way that's actually checked, tested, and maintainable — with hands-on consultancy available if you're not yet sure where the real risk in your current setup sits.","\u003Cp>Your developer just pasted a script into FileMaker that Klai or Copilot wrote in ten seconds. It runs. It looks clean. Nobody on the team fully understands why it works, and it's about to touch your invoicing data. That moment — approve it, or dig deeper? — is the new daily reality for anyone building business software, and it's the reason "review AI-generated code" has quietly become one of the most important skills an in-house dev team can have.\u003C\u002Fp>\n\u003Cp>This article walks through exactly how to review that code so it doesn't quietly break your business six months from now.\u003C\u002Fp>\n\u003Ch2>Why can't you just trust code that runs without errors?\u003C\u002Fh2>\n\u003Cp>"It runs" and "it's correct" are two very different things, and AI-generated code is especially good at hiding the gap between them.\u003C\u002Fp>\n\u003Cp>A FileMaker script generated by an AI assistant to calculate order totals might work perfectly on the five test records your developer tried — and then silently miscalculate the total the moment a record has a null quantity field, a negative discount, or a line item from a deleted product. The script doesn't error out. It just returns a wrong number that ends up on an invoice, and nobody notices until a customer complains three weeks later.\u003C\u002Fp>\n\u003Cp>This is the core risk with tools like Klai (Claris's AI assistant built into FileMaker) or general-purpose AI code assistants: they optimize for "produces working output on the input I described," not "handles every input this business will actually throw at it." The AI has no idea that your warehouse team occasionally leaves a quantity field blank, or that your sales team applies discounts that can push a total below zero. Only someone who knows your business does.\u003C\u002Fp>\n\u003Ch2>What should you actually check in an AI-generated script or module?\u003C\u002Fh2>\n\u003Cp>Treat every AI-generated piece of code — a FileMaker script, a custom function, a web app snippet, an API integration written with FmBetterforms or similar tooling — as a first draft from a very fast, very confident junior developer. Here's the review checklist that catches the problems that matter:\u003C\u002Fp>\n\u003Col>\n\u003Cli>\u003Cstrong>Trace every input path, not just the happy path.\u003C\u002Fstrong> What happens with a null value, an empty string, a zero, a negative number, or a duplicate record ID?\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Check error handling explicitly.\u003C\u002Fstrong> Did the AI wrap risky steps (API calls, file writes, find requests) in error checks, or does it assume everything succeeds?\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Look for hardcoded assumptions.\u003C\u002Fstrong> AI code often hardcodes today's field names, table structures, or business rules ($5% discount) that were true in the example but not in your actual schema.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Verify it matches your naming and structuring conventions.\u003C\u002Fstrong> Inconsistent naming across a solution is how six-month-old scripts become unmaintainable.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Check for redundant or dead logic.\u003C\u002Fstrong> AI assistants sometimes generate extra conditional branches or duplicate calculations "just in case" — these bloat scripts and confuse the next developer.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Confirm it performs at scale.\u003C\u002Fstrong> A script tested against 50 records in a demo file can behave very differently against 500,000 records in production — nested loops and unindexed finds are common AI blind spots.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Check security and data exposure.\u003C\u002Fstrong> Does the generated code log sensitive data, expose an API key in plain text, or skip privilege checks it should have?\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Re-read it for readability.\u003C\u002Fstrong> If a human developer can't explain what a script does in one sentence, rewrite it — even if it "works."\u003C\u002Fli>\n\u003C\u002Fol>\n\u003Cimg src=\"\u002Fapi\u002Fknowledge\u002Finline-image\u002F313?w=700&f=webp\" alt=\"checklist next to a code script with a magnifying glass\" loading=\"lazy\" class=\"w-full sm:w-1\u002F3 sm:float-left sm:mr-7 mb-5 rounded-2xl border border-[#E8E8ED] bg-[#F5F5F7]\" \u002F>\n\n\u003Ch2>Does the review process change for FileMaker specifically?\u003C\u002Fh2>\n\u003Cp>Yes — a little. FileMaker's low-code layer means AI-generated logic often lives in three places at once: a calculation formula, a script step, and a custom function. Reviewing FileMaker code well means checking all three together, because AI tools frequently generate a calculation that duplicates logic already sitting in a custom function elsewhere in the same file — which means the next schema change has to be made twice, and inevitably only gets made once.\u003C\u002Fp>\n\u003Cp>With Klai specifically, since it's context-aware of your existing schema, it's tempting to trust its suggestions more than a generic AI chatbot's. Don't. Context-awareness reduces some errors (wrong field names, for instance) but doesn't eliminate logic errors, and a confident, schema-correct suggestion is arguably more dangerous than an obviously wrong one, because it looks more trustworthy.\u003C\u002Fp>\n\u003Ch2>How do you review AI-generated code when you're not a developer yourself?\u003C\u002Fh2>\n\u003Cp>Many business owners and IT managers approving these changes aren't reading the script line by line — and that's fine, as long as the review process compensates for it. Ask your developer or vendor three questions before anything goes live:\u003C\u002Fp>\n\u003Cul>\n\u003Cli>\u003Cstrong>"What happens if this input is missing or wrong?"\u003C\u002Fstrong> A confident answer with specific examples (not "it should be fine") is the signal you're looking for.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>"Did you test this against real production-like data, not just a demo record?"\u003C\u002Fstrong> AI code that only sees clean sample data almost always misses edge cases that live data contains.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>"Can you explain what this does without reading the code?"\u003C\u002Fstrong> If your developer can't explain the logic in plain language, they haven't actually reviewed it — they've just glanced at it.\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch2>What's a practical review workflow for a team using AI code generation daily?\u003C\u002Fh2>\n\u003Cp>A lightweight process beats no process. Here's one that works well for small in-house teams and outside development partners alike:\u003C\u002Fp>\n\u003Col>\n\u003Cli>AI generates a first draft of the script, calculation, or integration code.\u003C\u002Fli>\n\u003Cli>The requesting developer runs it against at least three deliberately "nasty" test cases: empty fields, boundary values, and duplicate\u002Fconflicting records.\u003C\u002Fli>\n\u003Cli>A second person (peer review, even informal) reads the code and checks it against the checklist above.\u003C\u002Fli>\n\u003Cli>Anything touching financial data, customer records, or external APIs gets tested in a staging copy of the file before touching production.\u003C\u002Fli>\n\u003Cli>The final version gets a short comment block explaining \u003Cem>why\u003C\u002Fem> it works this way — future developers (and future AI assistants) will read that comment before they read the code.\u003C\u002Fli>\n\u003C\u002Fol>\n\u003Cp>That last step matters more than it sounds like it should. Six months from now, someone — possibly you — will need to modify this script without remembering the original AI conversation that produced it.\u003C\u002Fp>\n\u003Ch2>FAQ: reviewing AI-generated code\u003C\u002Fh2>\n\u003Cp>\u003Cstrong>Does using AI to write code make a FileMaker solution less reliable?\u003C\u002Fstrong>\nNot inherently — but it removes a safety net that used to exist by default. When a developer writes every line themselves, they naturally think through edge cases as they go. AI-generated code skips that thinking process, so it has to be added back in deliberately, during review.\u003C\u002Fp>\n\u003Cp>\u003Cstrong>Is it safe to let AI write API integration code?\u003C\u002Fstrong>\nOnly with extra scrutiny. Integration code (connecting FileMaker to an ERP, accounting package, or webshop) fails in ways that are hard to notice — a silently dropped record, a duplicate sync, a timeout that isn't retried. Always test integrations against real transaction volumes before trusting them unattended.\u003C\u002Fp>\n\u003Cp>\u003Cstrong>How much time should code review actually add to a project?\u003C\u002Fstrong>\nFor most FileMaker scripts, a proper review adds 15–30 minutes, not hours — because you're checking a specific list of known failure patterns, not re-deriving the logic from scratch. The time it saves later, when a bug would otherwise reach production, is far larger.\u003C\u002Fp>\n\u003Cp>\u003Cstrong>Should smaller businesses skip formal code review to save time?\u003C\u002Fstrong>\nNo — smaller businesses often have less room to absorb a costly mistake (a wrong invoice, a broken sync with their only warehouse system) than a large company would. A lightweight, consistent review habit matters more here, not less.\u003C\u002Fp>\n\u003Cp>AI-assisted development is changing how quickly FileMaker solutions, custom web apps, and system integrations get built — our \u003Ca href=\"https:\u002F\u002Floggix.com\u002Fen\u002Fblog\u002Fhow-low-code-and-ai-assisted-development-change-business-software\">broader look at how low-code and AI-assisted development change business software\u003C\u002Fa> covers that shift in more depth. But speed only helps if what gets shipped is actually correct, and that still depends on a human who understands the business checking the AI's work before it goes live.\u003C\u002Fp>\n\u003Cp>If your team is generating more code with AI than it can confidently review, that's usually a sign the underlying FileMaker solution, integration, or workflow needs a closer look — not just a faster reviewer. Loggix helps businesses build and audit custom FileMaker systems, connect them to other software through solid API integrations, and bring AI tools like Klai into a workflow in a way that's actually checked, tested, and maintainable — with hands-on consultancy available if you're not yet sure where the real risk in your current setup sits.\u003C\u002Fp>\n","Jeroen","2026-07-24",1784901674000,[19,20,21,22,23,24,25],"AI-generated code","code review","FileMaker development","Klai","software quality","low-code development","custom software","\u002Fapi\u002Fknowledge\u002Fimage\u002F331\u002F?v=bde9e378e01b",false,"",null,{"title":31,"slug":32},"Modern Software Development","modern-software-development",{"title":34,"slug":35},"How low-code and AI-assisted development change business software","how-low-code-and-ai-assisted-development-change-business-software"]