Why learn this text?<\/h2>\n
one about the best way to construction your prompts to allow your AI agent to carry out magic. There are already a sea of articles that goes into element about what construction to make use of and when so there\u2019s no want for one more.<\/p>\n
As a substitute, this text is one out of a sequence of articles which are about the best way to preserve your self, the coder, related within the trendy AI coding ecosystem.<\/p>\n
\nIt\u2019s about studying the methods that allow you to excel in utilising coding brokers higher than those that blindly hit tab or copy-paste.<\/strong><\/em><\/p>\n<\/blockquote>\n
We are going to go into the ideas from present software program engineering practices that you have to be conscious of, and go into why these ideas are related, notably now.<\/p>\n
\n
- By studying this sequence<\/strong>, you need to have a good suggestion of what widespread pitfalls to search for in auto-generated code, and know the best way to information a coding assistant to create manufacturing grade code that’s maintainable and extensible.<\/li>\n<\/ul>\n
\n
- This text is most related for<\/strong> budding programmers, graduates, and professionals from different technical industries that wish to stage up their coding experience. <\/li>\n<\/ul>\n
What we’ll cowl not solely makes you higher at utilizing coding assistants but in addition higher coders generally.<\/p>\n
The Core Ideas<\/h2>\n
The excessive stage ideas we\u2019ll cowl are the next:<\/p>\n
\n
- Code Smells<\/li>\n
- Abstraction<\/li>\n
- Design Patterns<\/li>\n<\/ul>\n
In essence, there\u2019s nothing new about them. To seasoned builders, they’re second nature, drilled into their brains by years of PR critiques and debugging. You ultimately attain a degree the place you instinctively react to code that \u201cfeels<\/em>\u201d like future ache.<\/p>\n
And now, they’re maybe extra related than ever since coding assistants have grow to be a vital a part of any builders\u2019 expertise, be it juniors to seniors.<\/p>\n
Why?<\/em><\/strong><\/p>\n
As a result of the handbook labor of writing code has been offloaded. The first accountability for any developer has now shifted from writing code to reviewing it. Everybody has successfully grow to be a senior developer guiding a junior (the coding assistant).<\/p>\n
So, it\u2019s grow to be important for even junior software program practitioners to have the ability to \u2018overview<\/em>\u2019 code. However the ones who will thrive in immediately\u2019s trade are those with the foresight of a senior developer.<\/p>\n
This is the reason we can be masking the above ideas in order that within the very very least, you may inform your coding assistant to take them into consideration, even when you your self don\u2019t precisely know what you\u2019re in search of.<\/p>\n
So, introductions at the moment are completed. Let\u2019s get straight into our first subject: Code smells.<\/strong><\/p>\n
Code Smells<\/h2>\n
What’s a code scent? <\/p>\n
I discover it a really aptly named time period \u2013 it\u2019s the equal of bitter smelling milk indicating to you that it\u2019s a nasty thought to drink it.<\/p>\n
For many years, builders have learnt by trial and error what sort of code works long-term. \u201cSmelly\u201d code are brittle, vulnerable to hidden bugs, and troublesome for a human or AI agent to know precisely what\u2019s occurring. <\/p>\n
Thus it’s usually very helpful for builders to learn about code smells and the best way to detect them.<\/p>\n
\nHelpful hyperlinks for studying extra about code smells:<\/em><\/strong><\/p>\n
https:\/\/luzkan.github.io\/smells<\/a><\/p>\n
https:\/\/refactoring.guru\/refactoring\/smells<\/a><\/p>\n<\/blockquote>\n
Now, having used coding brokers to construct every part from skilled ML pipelines for my 9-5 job to total cell apps in languages I\u2019d by no means touched earlier than for my side-projects, I\u2019ve recognized two typical \u201csmells\u201d that emerge while you grow to be over-reliant in your coding assistant:<\/p>\n
\n
- Divergent Change<\/li>\n
- Speculative Generality<\/li>\n<\/ul>\n
Let\u2019s undergo what they’re, the dangers concerned, and an instance of the best way to repair it.<\/p>\n
Picture by Greg Jewett<\/a> on Unsplash<\/a><\/figcaption><\/figure>\n Divergent Change<\/h2>\n
Divergent change is when a single module or class is doing too many issues without delay. The aim of the code has \u2018diverged\u2019 into many alternative instructions and so slightly than being centered on being good at one job (Single Duty Precept<\/em>), it’s making an attempt to do every part.<\/p>\n
This leads to a painful scenario the place this code is at all times breaking and thus requires fixing for varied unbiased causes.<\/p>\n
When does it occur with AI?<\/h2>\n
When the developer is just not engaged with the codebase and blindly accepts the Agent output, you’re doubly prone to this.<\/p>\n
Sure, you will have completed all the right issues and made a properly structured immediate that adheres to the most recent is in immediate engineering. <\/p>\n
However generally, when you ask it to \u201cadd performance to deal with X<\/em>,\u201d the agent will normally do precisely as it’s advised and cram code into your present class, particularly when the present codebase is already very sophisticated.<\/p>\n
It’s in the end as much as you to take note of the function, accountability and meant utilization of the code to provide you with a holistic method. In any other case, you\u2019re very more likely to find yourself with smelly code.<\/p>\n
Instance \u2014 ML Engineering<\/h2>\n
Beneath, we’ve a
ModelPipeline<\/code> class from which you will get whiffs of future extensibility points.<\/p>\n\nclass ModelPipeline:\n def __init__(self, data_path):\n self.data_path = data_path\n\n def load_from_s3(self):\n print(f\"Connecting to S3 to get {self.data_path}\")\n return \"raw_data\"\n\n def clean_txn_data(self, information):\n print(\"Cleansing particular transaction JSON format\")\n return \"cleaned_data\"\n\n def train_xgboost(self, information):\n print(\"Working XGBoost coach\")\n return \"mannequin\"<\/code><\/pre>\n\nA fast warning<\/strong>:<\/summary>\n
\nWe are able to\u2019t speak in absolutes and say this code is unhealthy only for the sake of it.<\/em><\/p>\n
It at all times is determined by the broader context of how code is used. For a easy codebase that isn\u2019t anticipated to develop in scope, the beneath is completely high-quality.<\/em><\/p>\n<\/blockquote>\n<\/details>\n
\nAdditionally notice:<\/strong> <\/summary>\n
\nIt\u2019s a contrived and easy instance for example the idea.<\/em>
Don\u2019t trouble giving this to an agent to show it might work out that is smelly with out being advised so. The purpose is for you<\/strong> to recognise the scent earlier than the agent makes it worse.<\/em><\/p>\n<\/blockquote>\n<\/details>\nSo, what are issues that ought to be going by your head while you have a look at this code?<\/p>\n
\n
- Information retrieval:<\/strong> What occurs once we begin having multiple information supply, like Bigquery tables, native databases, or Azure blobs? How probably is that this to occur?<\/li>\n
- Information Engineering:<\/strong> If the upstream information modifications or downstream modelling modifications, this can even want to alter.<\/li>\n
- Modelling:<\/strong> If we use totally different fashions,
LightGBM<\/code> or some Neural Web, the upstream modelling wants to alter.<\/li>\n<\/ul>\nIt’s best to discover that by coupling Platform, Information engineering, and ML engineering considerations right into a single place, we\u2019ve tripled the explanation for this code to be modified \u2013 i.e. code that’s starting to scent like \u2018divergent change<\/em>\u2018.<\/p>\n
Why is that this a potential downside?<\/h2>\n
\n
- Operational threat: <\/strong>Each edit runs the chance of introducing a bug, be it human or AI. By having this class put on three totally different hats, you\u2019ve tripled the chance of this breaking, since there\u2019s thrice as extra causes for this code to alter.<\/li>\n
- AI Agent Context Air pollution:<\/strong> The Agent sees the cleansing and coaching code as a part of the identical downside. For instance, it’s extra more likely to change the coaching and information loading logic to accommodate a change within the information engineering, although it was pointless. Finally, this will increase the \u2018divergent change\u2019 code scent.<\/li>\n
- Threat is magnified by AI:<\/strong> An agent can rewrite a whole lot of strains of code in a second. If these strains symbolize three totally different disciplines, the agent has simply tripled the possibility of introducing a bug that your unit checks may not catch.<\/li>\n<\/ol>\n
How one can repair it?<\/h2>\n
The dangers outlined above ought to provide you with some concepts about the best way to refactor this code.<\/p>\n
One potential method is as beneath:<\/p>\n
class S3DataLoader:\n \"\"\"Handles solely Infrastructure considerations.\"\"\"\n def __init__(self, data_path):\n self.data_path = data_path\n\n def load(self):\n print(f\"Connecting to S3 to get {self.data_path}\")\n return \"raw_data\"\n\nclass TransactionsCleaner:\n \"\"\"Handles solely Information Area\/Schema considerations.\"\"\"\n def clear(self, information):\n print(\"Cleansing particular transaction JSON format\")\n return \"cleaned_data\"\n\nclass XGBoostTrainer:\n \"\"\"Handles solely ML\/Analysis considerations.\"\"\"\n def practice(self, information):\n print(\"Working XGBoost coach\")\n return \"mannequin\"\n\nclass ModelPipeline:\n \"\"\"The Orchestrator: It is aware of 'what' to do, however not 'how' to do it.\"\"\"\n def __init__(self, loader, cleaner, coach):\n self.loader = loader\n self.cleaner = cleaner\n self.coach = coach\n\n def run(self):\n information = self.loader.load()\n cleaned = self.cleaner.clear(information)\n return self.coach.practice(cleaned)<\/code><\/pre>\nPreviously, the mannequin pipeline\u2019s accountability was to deal with all the DS stack.<\/p>\n
Now, its accountability is to orchestrate the totally different modelling levels, while the complexities of every stage is cleanly separated into their very own respective lessons.<\/p>\n
What does this obtain?<\/p>\n
1. Minimised Operational Threat:<\/strong> Now, considerations are decoupled and tasks are stark clear. You may refactor your information loading logic with confidence that the ML coaching code stays untouched. So long as the inputs and outputs (the \u201ccontracts\u201d) keep the identical, the chance of impacting something downstream is lowered.<\/p>\n
2. Testable Code:<\/strong> It’s considerably simpler to put in writing unit checks because the scope of testing is smaller and properly outlined.<\/p>\n
3. Lego-brick Flexibility:<\/strong> The structure is now open for extension. Have to migrate from S3 to Azure? Merely drop in an AzureBlobLoader. Need to experiment with LightGBM? Swap the coach.<\/p>\n
You in the end find yourself with code that’s extra dependable, readable, and maintainable for each you and the AI agent. When you don\u2019t intervene, it\u2019s probably this class grow to be greater, broader, and flakier and find yourself being an operational nightmare.<\/p>\n
Speculative Generality<\/h2>\n
Picture by Greg Jewett<\/a> on Unsplash<\/a><\/figcaption><\/figure>\n While \u2018Divergent Change<\/em>\u2018 happens most frequently in an already giant and complex codebase, \u2018Speculative Generality<\/em>\u2018 appears to happen while you begin out creating a brand new mission.<\/p>\n
This code scent is when the developer tries to future-proof a mission by guessing how issues will pan out, leading to pointless performance that solely will increase complexity.<\/p>\n
We\u2019ve all been there:<\/p>\n
\n\u201cI\u2019ll make this mannequin coaching pipeline help every kind of fashions, cross validation and hyperparameter tuning strategies, and ensure there\u2019s human-in-the-loop suggestions for mannequin choice in order that we are able to use this for all of our coaching sooner or later!\u201d<\/em><\/p>\n<\/blockquote>\n
solely to search out that\u2026<\/p>\n
\n
- It\u2019s a monster of a job, <\/li>\n
- code seems flaky, <\/li>\n
- you spend an excessive amount of time on it <\/li>\n
- while you\u2019ve not been capable of construct out the easy LightGBM classification mannequin that you just wanted within the first place.<\/li>\n<\/ol>\n
When AI Brokers are prone to this scent<\/h2>\n
I\u2019ve discovered that the most recent, excessive performing coding brokers are most prone to this scent. Couple a robust agent with a imprecise immediate, and also you shortly find yourself with too many modules and a whole lot of strains of latest code.<\/p>\n
Maybe each line is pure gold and it\u2019s precisely what you want. After I skilled one thing like this not too long ago, the code definitely appeared to make sense to me at first.<\/p>\n
However I ended up rejecting all of it. Why?<\/p>\n
As a result of the agent was making design decisions for a future I hadn\u2019t even mapped out but. It felt like I used to be shedding management of my very own codebase, and that it might grow to be an actual ache to undo sooner or later if the necessity arises.<\/p>\n
The Key Precept: Develop your codebase organically<\/h2>\n
The mantra to recollect when reviewing AI output is \u201cYAGNI<\/strong>\u201d (You ain\u2019t gonna want it<\/em>). It\u2019s a precept in software program growth that implies you need to solely implement the code you want, not the code you foresee.<\/p>\n
\nBegin with the only factor that works. Then, iterate on it.<\/strong><\/em><\/p>\n<\/blockquote>\n
This can be a extra pure, natural manner of rising your codebase that will get issues completed, while additionally being lean, easy, and fewer prone to bugs.<\/p>\n
Revisiting our examples<\/h2>\n
We beforehand checked out refactoring Instance 1 (The \u201cDo-It-All\u201d class<\/em>) into Instance 2 (The Orchestrator<\/em>) to show how the unique
ModelPipeline<\/code> code was smelly.<\/p>\nIt wanted to be refactored as a result of it was topic to too many modifications for too many unbiased causes, and in its present state the code was too brittle to take care of successfully.<\/p>\n
Instance 1<\/h3>\n
class ModelPipeline:\n def __init__(self, data_path):\n self.data_path = data_path\n\n def load_from_s3(self):\n print(f\"Connecting to S3 to get {self.data_path}\")\n return \"raw_data\"\n\n def clean_txn_data(self, information):\n print(\"Cleansing particular transaction JSON format\")\n return \"cleaned_data\"\n\n def train_xgboost(self, information):\n print(\"Working XGBoost coach\")\n return \"mannequin\"<\/code><\/pre>\nInstance 2<\/h3>\n
class S3DataLoader:\n \"\"\"Handles solely Infrastructure considerations.\"\"\"\n def __init__(self, data_path):\n self.data_path = data_path\n\n def load(self):\n print(f\"Connecting to S3 to get {self.data_path}\")\n return \"raw_data\"\n\nclass TransactionsCleaner:\n \"\"\"Handles solely Information Area\/Schema considerations.\"\"\"\n def clear(self, information):\n print(\"Cleansing particular transaction JSON format\")\n return \"cleaned_data\"\n\nclass XGBoostTrainer:\n \"\"\"Handles solely ML\/Analysis considerations.\"\"\"\n def practice(self, information):\n print(\"Working XGBoost coach\")\n return \"mannequin\"\n\nclass ModelPipeline:\n \"\"\"The Orchestrator: It is aware of 'what' to do, however not 'how' to do it.\"\"\"\n def __init__(self, loader, cleaner, coach):\n self.loader = loader\n self.cleaner = cleaner\n self.coach = coach\n\n def run(self):\n information = self.loader.load()\n cleaned = self.cleaner.clear(information)\n return self.coach.practice(cleaned)<\/code><\/pre>\nBeforehand, we implicitly assumed that this was manufacturing grade code that was topic to the varied upkeep modifications\/characteristic additions which are regularly made for such code. In such context, the \u2018Divergent Change\u2019 code scent was related.<\/p>\n
However what if this was code for a brand new product MVP or R&D? Would the identical \u2018Divergent Change\u2019 code-smell apply on this context?<\/p>\n
Picture by Kenny Eliason<\/a> on Unsplash<\/a><\/figcaption><\/figure>\n In such a state of affairs, choosing instance 2 may very well be the smellier selection.<\/p>\n
If the scope of the mission is to contemplate one information supply, or one mannequin, constructing three separate lessons and an orchestrator could rely as \u2018pre-solving\u2019 issues you don\u2019t but have.<\/p>\n
Thus, in MVP\/R&D conditions the place detailed deployment concerns are unknown and there are particular enter information\/output mannequin necessities, instance 1 could possibly be extra acceptable.<\/p>\n
The Overarching Lesson<\/h2>\n
What these two code smells reveal is that software program engineering is never about \u201cappropriate<\/strong><\/em>\u201d code. It’s about context<\/em><\/strong>.<\/p>\n
A coding agent can write good Python in each operate and syntax, however it doesn\u2019t know your total enterprise context. It doesn\u2019t know if the script it\u2019s writing is a throwaway experiment or the spine of a multi-million greenback manufacturing pipeline revamp.<\/p>\n
Effectivity tradeoffs<\/h3>\n
You could possibly argue that we are able to merely feed the AI each little element of enterprise context, from the conferences you\u2019ve needed to the tea-break chats you had with a fellow colleague. However in apply, that isn\u2019t scalable.<\/p>\n
If you must spend half and hour writing a \u201ccontext memo\u201d simply to get a clear 50-line operate, have you ever actually gained effectivity? Or have you ever simply remodeled the handbook labor of writing code into that of writing prompts?<\/p>\n
What makes you stand out from the remaining<\/h3>\n
Within the age of AI, your worth as an information scientist has basically modified. The handbook labour of writing code has now been eliminated. Brokers will deal with the boilerplating, the formatting, and unit testing.<\/p>\n
So, to make your self stand out from the opposite information scientists who’re blindly copy pasting code, you could have the structural instinct to information a coding agent in a path that’s related on your distinctive scenario. This leads to higher reliability, efficiency, and outcomes which are mirrored on you, making you stand out.<\/p>\n
However to attain this, you could construct this instinct that comes years of expertise by figuring out the code smells we\u2019ve mentioned, and the opposite two ideas (design patterns, abstraction<\/em>) that we are going to delve into in subsequent articles.<\/p>\n
And in the end, with the ability to do that successfully offers you extra headspace to deal with the issue fixing and architecting an answer an issue \u2013 i.e. the actual \u2018enjoyable\u2019 of knowledge science.<\/p>\n
Associated Articles<\/h2>\n
When you favored this text, see my Software program Engineering Ideas for Information Scientists<\/strong> sequence, the place we broaden on the ideas most related for Information Scientists<\/p>\n<\/div>\n\n","protected":false},"excerpt":{"rendered":"
Why learn this text? one about the best way to construction your prompts to allow your AI agent to carry out magic. There are already a sea of articles that goes into element about what construction to make use of and when so there\u2019s no want for one more. As a substitute, this text is […]<\/p>\n","protected":false},"author":2,"featured_media":11982,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[55],"tags":[544,617,1256,1893,7921,157,3712,4704,1101],"class_list":["post-11980","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-machine-learning","tag-age","tag-agents","tag-coding","tag-concepts","tag-curriculum","tag-data","tag-essential","tag-missing","tag-scientists"],"_links":{"self":[{"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=\/wp\/v2\/posts\/11980","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=11980"}],"version-history":[{"count":1,"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=\/wp\/v2\/posts\/11980\/revisions"}],"predecessor-version":[{"id":11981,"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=\/wp\/v2\/posts\/11980\/revisions\/11981"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=\/wp\/v2\/media\/11982"}],"wp:attachment":[{"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=11980"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=11980"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/techtrendfeed.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=11980"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}
![\"\"](\"https:\/\/contributor.insightmediagroup.io\/wp-content\/uploads\/2026\/02\/greg-jewett-r-1XRSe-DYo-unsplash-683x1024.jpg\")
![\"\"](\"https:\/\/contributor.insightmediagroup.io\/wp-content\/uploads\/2026\/02\/content-pixie-UXuT3GEn4TA-unsplash-1024x683.jpg\")
![\"\"](\"https:\/\/contributor.insightmediagroup.io\/wp-content\/uploads\/2026\/02\/kenny-eliason-2RRq1BHPq4E-unsplash.jpg\")