<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Luke AI Dev on Medium]]></title>
        <description><![CDATA[Stories by Luke AI Dev on Medium]]></description>
        <link>https://medium.com/@luczxdev?source=rss-a99bef0c8cc6------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*DeOQHyc4Orca6YRmzoBUIQ.jpeg</url>
            <title>Stories by Luke AI Dev on Medium</title>
            <link>https://medium.com/@luczxdev?source=rss-a99bef0c8cc6------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 19:18:57 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@luczxdev/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Context Is All You Need For AI Performance]]></title>
            <link>https://medium.com/ai-creators/context-is-all-you-need-for-ai-performance-a4d1bb3c4542?source=rss-a99bef0c8cc6------2</link>
            <guid isPermaLink="false">https://medium.com/p/a4d1bb3c4542</guid>
            <category><![CDATA[prompt-engineering]]></category>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[context-engineering]]></category>
            <category><![CDATA[writing-prompts]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Luke AI Dev]]></dc:creator>
            <pubDate>Mon, 01 Sep 2025 20:40:07 GMT</pubDate>
            <atom:updated>2025-09-01T20:40:07.640Z</atom:updated>
            <content:encoded><![CDATA[<p>Since ChatGPT’s release, the AI community has become obsessed with prompt engineering.</p><p>Scroll through any social media, and you’ll find countless threads about “game-changing prompts” and clever techniques for getting better AI outputs.</p><p>But here’s the thing: once you master the basics of prompt engineering, those clever tricks won’t make your output any better.</p><p>Sure, prompts help you control tone, format, and general behavior. They’re essential for guiding the AI and establishing ground rules.</p><p>However, when it comes to generating truly useful, accurate, and relevant outputs, prompt engineering alone falls short.</p><p>And that’s why in this article, I want to introduce you to the real solution: <strong>context engineering</strong>.</p><p>With context engineering, you will be able to squeeze the most out of AI’s intelligence and finally get useful and relevant outputs that you would never be able to do otherwise.</p><p>But before we actually start discussing this new approach, let’s take a minute to discuss the shortcomings of prompt engineering. Just so you know what exactly we are trying to fix…</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/896/1*5ZhC5cNcHFipRGG2jMYclw.jpeg" /><figcaption>Read on to learn how to prompt this guy properly</figcaption></figure><h3>Why Perfecting Your Prompts Is a Waste Of Time</h3><p>You see, the thing with prompt engineering is that it<strong> </strong>has a very narrow scope.</p><p>It’s only goal is to clearly tell the AI what you want it to do and what to avoid. In other words, its only purpose is to precisely guide its actions.</p><p>Of course, doing so is important. You can’t expect AI to do a good job if you don’t instruct it how to do your task.</p><p>The only problem is that people got so obsessed with prompt engineering, that they started adding useless superficial hacks…</p><ul><li>“a cat will die if you fail”</li><li>“I’ll tip you $100 for correct work”</li><li>“My company will file for bankruptcy if you give me wrong information”</li><li>“Make no mistakes”</li></ul><p>Not surprisingly, none of these tricks really work.</p><p>Today’s models (GPT-5 is the best one at the time of this writing) are already intelligent enough to work with just a clear goal and a few basic constraints in the prompt. You don’t need to overthink the prompt.</p><p>What you need instead, is to focus on providing the right information.</p><p>You see, when models lack <strong>sufficient context </strong>for the task you are working on, they will inevitably hallucinate and generate useless slop.</p><p>But when you give them the right information, they produce accurate, relevant results.</p><p>Now, let’s get more specific about what “sufficient context” actually means.</p><p>So please allow me to borrow the definition of sufficient context from this <a href="https://arxiv.org/abs/2411.06037">paper</a>: <strong>“The context is sufficient to answer the query if it contains all the necessary information to provide a definitive answer.”</strong></p><p>In other words, provide all information you think will help the AI generate your answer. That’s it, nothing complicated.</p><p>Your one and only job here is to feed it with the data it needs to generate a useful output.</p><p>And whenever you spot the AI derailing, see it as a sign that you failed to provide it with sufficient context in your prompt.</p><p>This is exactly why in today’s world of advanced language models, you must not think in terms of “prompt engineering”, but in terms of context engineering.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*va_TXBWkNJ6QM7m7REHoqg.png" /><figcaption>Shopify’s Co-founder and CEO take on prompt engineering vs context engineering</figcaption></figure><h3><strong>The Simple Solution to Better AI Outputs</strong></h3><p>Context engineering has one simple goal: provide the AI with all the information it needs to deliver accurate results.</p><p>This could be as simple as:</p><ul><li>Sharing a screenshot example of your desired result</li><li>Uploading a company memo for the report you’re writing</li><li>Including a documentation URL for the system you’re building</li></ul><p>Again, the goal here is to give the AI the necessary information it needs to generate your desired output.</p><p>So, in practice, this approach looks like this:</p><ol><li>Define your task</li><li>Figure out what information AI needs to better deliver on your task</li><li>Provide this information in your prompt</li></ol><p>That’s it. Notice that the point here is that you don’t want the AI to fill in the blanks about your task.</p><p>So instead, you just take a minute to work backwards from your end goal to identify what information the model actually needs and provide it to it.</p><p>Are you asking it to write an article? Debug code using a specific API? Respond to a customer complaint? Once you’re clear on the task, work backwards.</p><p>What background information, specific details, or examples would a human need to do this job well? That’s your context checklist.</p><p>Notice that the amount of context you need scales directly with task complexity and specificity.</p><ul><li>Working on brand strategy? You’ll want to include brand guidelines, target audience descriptions, and current brand assets.</li><li>Just need a casual tweet response? A screenshot of the original tweet might be enough.</li></ul><p>General knowledge tasks require minimal context — if you’re asking for widely known facts (ie. “What’s the capital of France?”), the AI likely has sufficient information already.</p><p>But the moment you venture into company-specific work, personal projects, or niche applications, context becomes everything.</p><p><strong>Think of it as a spectrum:</strong> generic tasks need minimal context, while highly specific or personal tasks demand comprehensive background information.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/864/1*LbWEhVxH0DAwOHpMS_RcwQ.jpeg" /><figcaption>Generic tasks work with minimal context. Specific tasks demand comprehensive background information.</figcaption></figure><p>The key here is recognizing where your task falls on this spectrum and providing context accordingly.</p><p>Also note that it’s not about dumping everything you know into the prompt…</p><p>You only need to provide it with information that is directly relevant to the task.</p><h3>How To Get The Most From Context Engineering</h3><p>Let me illustrate this with a concrete example.</p><p>Imagine you need to write a performance report for your company — something to present to leadership about quarterly results.</p><p>You craft a detailed prompt asking for professional tone, specific structure, and appropriate length.</p><p>Then, the AI delivers exactly what you asked for: a well-structured, professionally written report.</p><p>The only problem is that it’s filled with fabricated metrics, made-up initiatives, and fictional achievements. Your prompt controlled the format perfectly, but without the right context, the <strong>information </strong>of the report is worthless.</p><p>Now, this doesn’t mean your prompt is wrong, it simply means that you failed to provide the AI with sufficient information.</p><p>After all, the AI can’t magically know your company’s actual performance metrics, recent projects, or strategic priorities. And it’s exactly this information that will transform a generic AI generated company report into one that you can present to your superiors with confidence.</p><p>Again, refer back to our context spectrum. The more specific your task, the more context you need to provide.</p><p><strong>What you should have done instead:</strong></p><p>Instead of just asking for a “quarterly performance report,” you should have provided:</p><ul><li>Actual revenue numbers and key metrics from your CRM or analytics dashboard</li><li>A list of completed projects and initiatives from the quarter</li><li>Your team’s specific goals and KPIs</li></ul><p>With this context, the AI would generate a report filled with real data, actual achievements, and genuine insights — something you could confidently present without extensive fact-checking or rewriting.</p><h3>Wrapping Up</h3><p>As you could see, context engineering is more straightforward than it appears. You don’t need to become a prompt wizard or master complex techniques. You just need to give the AI the information it needs to succeed.</p><p>Instead of spending hours crafting the perfect prompt, spend a few minutes gathering the right documents, examples, and background information. Include relevant files, paste in reference materials, provide specific examples of what good looks like. The AI will do the rest.</p><p>Context engineering is ultimately about asking yourself: “What does the AI need to know?”</p><p>Once you make that mental switch, you’ll find that your AI outputs become dramatically more useful, accurate, and relevant.</p><p>Just keep in mind the 3-step approach of context engineering:</p><ol><li>Define your task</li><li>Figure out what information AI needs to better deliver on your task</li><li>Provide this information in your prompt</li></ol><p>And that’s all for today’s article.</p><p>Thanks for reading.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a4d1bb3c4542" width="1" height="1" alt=""><hr><p><a href="https://medium.com/ai-creators/context-is-all-you-need-for-ai-performance-a4d1bb3c4542">Context Is All You Need For AI Performance</a> was originally published in <a href="https://medium.com/ai-creators">AI Creators</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Right Way To Vibe Code?]]></title>
            <link>https://luczxdev.medium.com/the-right-way-to-vibe-code-335af93ada6c?source=rss-a99bef0c8cc6------2</link>
            <guid isPermaLink="false">https://medium.com/p/335af93ada6c</guid>
            <category><![CDATA[vibe-coding]]></category>
            <category><![CDATA[coding]]></category>
            <category><![CDATA[claude]]></category>
            <category><![CDATA[prompt]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Luke AI Dev]]></dc:creator>
            <pubDate>Sun, 17 Aug 2025 21:28:15 GMT</pubDate>
            <atom:updated>2025-08-17T21:28:15.346Z</atom:updated>
            <content:encoded><![CDATA[<p>I’ve been experimenting with different approaches to vibe coding for months now…</p><p>Trying everything from completely hands-off prompting to micro-managing every single change.</p><p>And after all of this trial and error, I’ve found that the more I vibe code, the more I keep coming back to one specific system.</p><p>So in this article, I want to show you 4 steps of this system that will transform how you work with AI.</p><p>It’s been the only method that consistently lets me build complex stuff without running into a brick wall sooner or later.</p><p>The only caveat here is that this system might be optimized for how I work and the tools I use (Claude Code and Cursor).</p><p>But if you adapt the principles we are about to discuss to your own particular setup, I am sure you will find this system useful.</p><p>And at the very end, I will share a simple 80-words prompt to implement the steps of this system. So don’t worry, everything we are about to discuss is easier to implement than it seems.</p><p>But before we get into the system itself, we need to take a few moments to pinpoint the root cause of most vibecoding mistakes, just so you can understand exactly what we’re trying to avoid…</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/896/1*Bn4kAznFzKG1k6zHC4DhIg.jpeg" /><figcaption>Me vibe coding right now</figcaption></figure><h3>The Problem With Most Vibe Coding</h3><p>You see, most vibe coders tend to follow the same pattern:</p><ol><li>Open their large code base</li><li>Prompt AI to do XYZ in a very generic way (ie. Add a dashboard for users to track their progress)</li><li>Watch mindlessly as AI do its thing</li></ol><p>Now, if the app you are building is still in its infancy, AI might do the task well. But as the app grows and its features become more complex, problem arises because AI start to “slop”:</p><ul><li>Rewrite existing code unnecessarily</li><li>Maintain god-files that are impossible to change without breaking.</li><li>Make assumptions about your codebase that don’t match your actual implementation</li></ul><p>Before you know it, you will get to a point where you don’t even recognize your app and any new changes you introduce will eventually break it.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_mQjBo2QTmW_dCRodZhcOw.png" /></figure><p>The reason this happens is because of something called that has become known as the Laffer curve of AI…</p><p>If you let AI run freely, like most vibecoders do, it will end up creating a lot of useless stuff. All of that slop might even look useful on the surface, but as soon as you try to change anything, things start breaking apart.</p><p>However, the solution is not to stop using AI. After all, you&#39;d be leaving aside the best productivity-enhancing tech of the decade if you did that.</p><p>So, you need to find a balance on the tasks you delegate to AI.</p><p>Keyword here being <strong>balance</strong>. You should guide AI enough to complete your task, but also step in, adjust, and check its work whenever needed, just in case it&#39;s going off track.</p><p>Of course, each project will have different points where AI should be used and where you should do the work yourself by checking the output.</p><p>Thankfully, the principles and steps we are about to discuss will help you find this balance and make you much more productive with AI.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/955/1*u68cQYeGb9EoErom5oZaEQ.png" /></figure><h3>Change One Incremental Feature At a Time</h3><p>Changing one small feature at a time is really important. Especially if your code or app is already big. And this is because:</p><ul><li>As your app gets more complex, you can&#39;t be sure how features will interact and if anything will break.</li><li>You can’t be sure if the AI will introduce side effects you didn’t anticipate</li><li>AI might not fully understand your prompt and what you are envisioning. (maybe it needs more context, maybe you were too vague, and etc)</li></ul><p>Not surprisingly, the risk of the AI making mistakes or breaking something that was working well is high if you ask for it to change too much at once.</p><p>So, try to keep changes small and simple. The goal here is to keep it simple and be very clear about what you want to change.</p><p>Let me show you what I mean, here&#39;s <a href="https://github.com/luc-pimentel/YT_crawler">one of my projects</a> (<a href="https://github.com/luc-pimentel/YT_crawler">a YouTube data scraper I built recently</a>).</p><p>And for the sake of our example, I am creating a feature to scrape YouTube playlists.</p><p>Considering it’s a real time YouTube data scraper, it relies on YouTube’s unpredictable HTML/CSS structure to work well.</p><p>This adds a big layer of complexity, and if I were to prompt AI to <strong><em>“Create a Youtube playlist scraper. Make no mistakes”</em></strong> chances are that it would fail.</p><p>In fact, the code might even look correct at first glance, but would fail when actually trying to run it against YouTube’s real web responses.</p><p>But if I mentally break the task down into steps like:</p><ol><li>Making the web request;</li><li>Parse the results;</li><li>Wrap it all into a single function;</li></ol><p>And prompt the AI one incremental step at a time, the chance of getting working code is much higher.</p><p>Which is exactly what I did below:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Yvb1jCQEaqdMDDI42qohDA.png" /><figcaption>Instead of asking AI to build the entire scraper, I started with two manageable pieces</figcaption></figure><p><strong>PS:</strong> Don’t worry if you don’t know how to code in python, the point of these steps and examples is that they are simple enough for you to grasp them</p><p>So as you can see, instead of prompting AI directly to create a scraper for me, I broke the task into easier and smaller steps.</p><p>The only problem is that this code does not work (yet).</p><p>The first function (easy and straightforward web request) is working, but the second one (parse the results) does not.</p><p>And that’s because Youtube’s data structure is not that simple, it actually is a 800K+ characters giant HTML response.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*CoGooQUz6sqy1GL4g58svw.png" /><figcaption>Small snippet of Youtube’s web response. No AI can make sense of this mess yet</figcaption></figure><p>Most vibe coders would stop here and assume vibe coding does not work, but that’s because they don’t yet know how to prompt their way out of this complexity.</p><p>So head on to the next step…</p><h3>Get a Decent Spec</h3><p>Now, prompting AI to do one task at a time is not enough… So before having AI to write the code, have a clear spec.</p><p>Ideally, your spec should have:</p><ol><li>Clear context of exactly what functionality you want to add</li><li>What changes (that is, additions, modifications or deletions) you want to make in the code</li><li>Constraints and edge cases to consider</li></ol><p>The bigger and more complex the change, the more detailed your spec needs to be.</p><p>This step is important because it ties back to what I’ve found to be the one overarching principle of vibe coding properly: one change at a time, described clearly.</p><p>Even if you forget all other steps, remembering this principle will help you.</p><p>Now, I realize it might seem like overkill to write detailed specs for every small change.</p><p>But when you give AI a clear spec of what you need, it will do exactly what you both agreed on. Not more, not less.</p><p>Again, at the risk of sounding like a broken record, the key is to steer the model very precisely. So, take time to create a clear plan.</p><p><strong>PS: </strong>Later in the article, I will show you a prompt that deals with all the steps we’ve discussed so far, making it easier to implement than you might think.</p><p>Now, back to our <a href="https://github.com/luc-pimentel/YT_crawler">YouTube data scraper</a> example.</p><p>After reading through the HTML response, I realize the data we wanted was nested in the following way:</p><ol><li>There was a HTML tag called &#39;script.&#39;</li><li>Inside this &#39;script&#39; tag, there was a dictionary named &#39;playlistVideoListRenderer&#39;.</li><li>Within this dictionary, there was a list with the playlist video details we where looking for.</li></ol><p>So, now armed with this info, I created a new AI prompt to:</p><ol><li>Create a function to get the ‘script’ tags data;</li><li>Create another function to get the ’playlistVideoListRenderer’ dict data.</li></ol><p>Notice that I didn’t send the 800K+ characters web response to AI and told him to figure out on his own (as most vibe coders would do here).</p><p>Also notice (in the screenshot below) that the <a href="https://github.com/luc-pimentel/YT_crawler/blob/55c5ed468ddcd8f823cff9358339596c2b95af1d/yt_crawler/utils.py#L162">second function AI created</a> is a complex recursive function to read through all the scripts and grab the dictionary data by its key.</p><p>In all honesty, I have no idea how I would recreate this function on my own.</p><p>But that doesn&#39;t matter, because the AI made it for me, I tested it a bit, and it works every time I tried it, and that’s the point of vibecoding at the end of the day.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*1lI_DfvHtEkvEliL3oOwkg.png" /><figcaption>Snippet of the functions AI created for me. The second function is actually much bigger</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*7D0uEIsKPz1UI1v_dzyPng.png" /><figcaption>And as you can see here, it gives us the data we need</figcaption></figure><p>So to sum up this step, the more precise you are in your spec, the better results you will get.</p><p>The goal here is to outline exactly what we have in mind so that when AI starts writing the code, nothing that we didn’t anticipate happens.</p><p>And if you ever hit a brick wall while vibe coding, chances are that your spec is too vague.</p><h3>Oversee The Changes</h3><p>Next up, you must keep an eye on all the changes that AI makes.</p><p>This is obvious in hindsight, but our own innate human laziness gets in the way. Its easier to simply assume the new code works and get going.</p><p>However, AI still does not understand the full context of your project or catch every edge case. We still need to oversee the changes it makes.</p><p>So always keep in mind the Laffer curve we discussed earlier… If you do everything yourself, you’ll move too slowly and miss out on AI’s productivity boost, but if you rely 100% on AI, you will end up with useless slop.</p><p>So you guide the AI, but verify everything it does. The good thing is that if you have followed the previous steps (and also the next one), overseeing the changes will be very easy.</p><p>Now back to the Youtube data scraper example…</p><p>After I finally got the data I wanted by nailing my prompt down to a decent spec, I told AI to wrap it all in a single function.</p><p>So now I will simply make sure AI is precisely following my instructions by overseeing the code changes.</p><p>How you do this depends on the tool you use of course. I use Cursor, which is great at showing what AI has written and lets you accept or decline it.</p><p>But each tool is different, so adjust this step to fit with your tool.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*zs_zfCp1mVQ5KlnZrxX_Og.png" /><figcaption>Cursor shows the new generated code and asks for my acceptance</figcaption></figure><p>Again, at this point, my goal is simply to watch where AI is changing and creating code, see if everything is in fact abiding to the spec, and I might even run the code just in case.</p><p>As you can see, this is nothing revolutionary, but it does take time and focus. Which is why most of us skip it and just assume the code works.</p><p>Luckily, the next step will be about how you can reduce the time spent on verification…</p><h3>Have A Decent Test Suite</h3><p>As we discussed throughout this article, the thing with AI is that you need to make sure its output is correct.</p><p>If you rely on manual checks, or worse, don&#39;t check at all, your app will eventually break and/or go in a direction you don&#39;t want.</p><p>But if you have a good test suite, like unit tests or integration tests, you can easily test a lot of your code when you want AI to make changes.</p><p>Moreover, if you&#39;re using Claude Code agent or a Cursor agent, it can also run your tests to ensure nothing breaks. And if something does break, the agent can fix itself.</p><p>Thus offloading most of the effort you would have to do yourself and help steer the AI on the exact direction you want.<br><br>Back to our scraper example… I decided to create 2 tests, one to check if the web request returns a successful response and the other to check if the data is returned correctly (ie. all videos have titles, thumbnails and etc…)</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*e-RLuDLPC1D-R5Dq2t_5fQ.png" /></figure><p>And If I want to add more features to this scraper (ie. sort the playlist playlist videos by recency), I can simply vibe code this new feature and then run the already existing tests.</p><p>If no test fails, it means all is good and I can keep vibe coding.</p><p>Else, I can manually inspect what went wrong or simply have the Agent read the error logs, keep iterating and correct itself.</p><p>As you see, having a good test suite saves you a lot of mental effort.</p><p>Though in all fairness, not all projects need a test suite. Especially if you don’t have any long term plans for it. But the more complex your code gets, the more you&#39;ll benefit from a good test suite.</p><h3><strong>The Prompt That Ties It All Together</strong></h3><p>As promised, here’s the exact prompt I use that enforces all the principles we’ve discussed.</p><p>This prompt easily covers steps 1, 2 and 3.</p><pre>I need help with a small coding task. Before you write any code:<br><br>1. First, analyze and explain what you understand about the requirements<br><br>2. Then, outline your proposed approach with the specific changes you would make<br><br>3. Ask for my confirmation before implementing any code<br><br>4. Only after I approve, provide the actual code implementation<br><br>Please start by explaining your understanding of the task and suggesting an approach before writing any actual code.<br><br>Here&#39;s the task I am working on: [briefly describe the task]</pre><p>This prompt will allow you to easily break down the task, get a clear spec from the AI, and maintain oversight of the changes.</p><p>It is designed for Cursor and Claude Code.</p><p>Other platforms might be different, but the overarching principles of being precise and incremental won’t change.</p><p>Use this template as your starting point, and your vibe coding sessions will be more productive with fewer detours.</p><p>Thanks for reading this far! I hope these 4 steps help you find the sweet spot between AI productivity and code quality.</p><p>Remember: incremental changes, clear specs, active oversight, and good tests. Use the prompt template as your starting point, but adapt it to your own workflow.</p><p>Please also check out the <a href="https://github.com/luc-pimentel/YT_crawler">YouTube data scraper I’m building here</a> (work in progress, feel free to contribute and suggest features/issues).</p><p>Happy vibe coding!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=335af93ada6c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Auto-Generate Pydantic Models From JSON Data]]></title>
            <link>https://medium.com/ai-creators/auto-generate-pydantic-models-from-json-data-91834c352e48?source=rss-a99bef0c8cc6------2</link>
            <guid isPermaLink="false">https://medium.com/p/91834c352e48</guid>
            <category><![CDATA[pydantic]]></category>
            <category><![CDATA[prompt-engineering]]></category>
            <category><![CDATA[claude]]></category>
            <category><![CDATA[ai-agent]]></category>
            <category><![CDATA[api-development]]></category>
            <dc:creator><![CDATA[Luke AI Dev]]></dc:creator>
            <pubDate>Wed, 06 Aug 2025 21:55:27 GMT</pubDate>
            <atom:updated>2025-08-06T21:55:27.513Z</atom:updated>
            <content:encoded><![CDATA[<p>If you’ve ever dealt with JSON data in Python, you know the pain of working with raw dictionaries…</p><p>One typo in a key name or wrong data type and your code breaks silently.</p><p>Pydantic solves this by bringing type safety and automatic validation to your data structures, catching errors right when data enters your system.</p><p>In short, Pydantic transforms messy dictionary wrangling into clean, reliable, and maintainable object code, and because of that, it’s a godsend for everyone who works with external API data on a regular basis.</p><p>Now, in this article, I’ll show you how to auto-generate Pydantic models from JSON data. And for this, we will only need Claude Code, and a custom agent I built.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/720/1*jtitR7_TFBzm3gGM6MBSmA.jpeg" /></figure><h3><strong>Getting Started with the Auto-Generator</strong></h3><p>First, you’ll need Claude Code installed...</p><p>Then, head over to my GitHub repo at <a href="https://github.com/luc-pimentel/api-to-pydantic">https://github.com/luc-pimentel/api-to-pydantic</a> and download the agets/api-to-pydantic.md file. This is the custom sub-agent that does all the heavy lifting.</p><p>To set it up, create a folder named .claude/agent/ in your project directory and place the api-to-pydantic.md file inside it. This follows Claude Code&#39;s sub-agent structure where agents are discovered from the agent/ folder.</p><p>If you want to use it across multiple projects, you can install it system-wide following the <a href="https://docs.anthropic.com/en/docs/claude-code/sub-agents">Claude Code sub-agents documentation</a>.</p><h3><strong>How to Use It</strong></h3><p>Once set up, using the agent is straightforward:</p><ol><li>Place your JSON file in your project directory (e.g., data.json)</li><li>Open Claude Code and instruct it to use the api-to-pydantic sub-agent:</li></ol><ul><li>“Use the api-to-pydantic agent to generate Pydantic models from data.json”</li></ul><p>3. The agent will analyze your JSON structure, generate the appropriate Pydantic models, and test them iteratively until they perfectly match your data</p><p>4. Copy the generated models into your Python project</p><p>The beauty of this approach is that the agent doesn’t just generate models — it validates them against your actual data, ensuring perfect compatibility.</p><p>No more manual tweaking or discovering edge cases in production.</p><h3>Live Example</h3><p>Let’s walk through a real example using actual YouTube data.</p><p>I scraped the comments from a YouTube video, which gave me a video_comments.json file containing nested comment structures with thumbnails, badges, and various metadata fields—exactly the kind of complex JSON you get from real-world APIs.</p><p>First, I navigate to my project directory and run the command:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*cbY1a3aibYD_oBkHQ5bv0w.png" /></figure><pre>&gt; run the video_comments.json through the api-to-pydantic</pre><p>The agent immediately starts working, creating a task list that shows exactly what it’s going to do:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*UZCr3WnNGY70uAA0vO000w.png" /></figure><ul><li>Locate and validate the JSON file</li><li>Examine existing models to understand the codebase conventions</li><li>Extract the schema from the JSON data</li><li>Generate Pydantic models that match your existing patterns</li><li>Create test files to validate everything works</li></ul><p>After generating the models, the agent automatically creates validation tests to ensure everything works correctly with the original scraped JSON data. The entire process takes about 30 seconds for a complex nested YouTube comment structure.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*fADjQOo1xjPLaTJfw08_qA.png" /></figure><p>The generated models include proper type hints, optional fields, and even import statements from your existing codebase.</p><p>Notice how it picks up on your project’s structure and imports types like Thumbnail, TextWithRuns, and SimpleText from your .search module—this is the agent understanding your codebase context.</p><p><strong>So, in short:</strong></p><ul><li>Placed a proper .json file in the project root (included in the repo for testing)</li><li>The agent generates models directly in your project, following your existing structure</li><li>Test files are created alongside the models for immediate validation</li></ul><h3>Wrapping Up</h3><p>I’ve been using this in my own projects for handling various API responses, and it’s saved me countless hours of manual model creation.</p><p>The iterative validation approach means the models just works every time.</p><p>If you find this useful, give the repo a star on GitHub, and feel free to contribute improvements or share your use cases. <a href="https://github.com/luc-pimentel/api-to-pydantic">https://github.com/luc-pimentel/api-to-pydantic</a></p><p>Got questions, found a bug, or have ideas for making it even better? Create an issue on the repo, drop a comment below, or reach out directly.</p><p>I’m always looking to improve the tool based on real-world usage.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=91834c352e48" width="1" height="1" alt=""><hr><p><a href="https://medium.com/ai-creators/auto-generate-pydantic-models-from-json-data-91834c352e48">Auto-Generate Pydantic Models From JSON Data</a> was originally published in <a href="https://medium.com/ai-creators">AI Creators</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Youtube Data Extraction: Easily Scrape All Videos And Comments From A Channel]]></title>
            <link>https://medium.com/ai-creators/youtube-data-extraction-easily-scrape-all-videos-and-comments-from-a-channel-b97886bf72a6?source=rss-a99bef0c8cc6------2</link>
            <guid isPermaLink="false">https://medium.com/p/b97886bf72a6</guid>
            <category><![CDATA[data]]></category>
            <category><![CDATA[scraping]]></category>
            <category><![CDATA[data-analysis]]></category>
            <category><![CDATA[youtube]]></category>
            <category><![CDATA[python]]></category>
            <dc:creator><![CDATA[Luke AI Dev]]></dc:creator>
            <pubDate>Mon, 01 Jul 2024 01:00:18 GMT</pubDate>
            <atom:updated>2024-07-01T01:00:18.917Z</atom:updated>
            <content:encoded><![CDATA[<p>YouTube is one of the biggest platforms for content creation and consumption.</p><p>And whether you’re a data scientist, market researcher, or content creator, the ability to extract comprehensive data from a YouTube channel can provide invaluable insights.</p><p>This tutorial introduces a powerful yet straightforward method to scrape all videos and comments from any YouTube channel.</p><p>Using Python and the YouTube Data API v3, we’ll build a robust scraper that can:</p><p>1. Fetch all videos from a specified channel</p><p>2. Extract detailed information for each video</p><p>3. Collect all comments and replies for every video</p><p>4. Store the data in an easily analyzable format</p><p>Our approach prioritizes efficiency and simplicity, making it accessible for programmers of all skill levels. By the end of this guide, you’ll have a versatile tool at your disposal, capable of gathering extensive YouTube data with minimal effort.</p><p>Let’s dive into the world of YouTube data extraction and unlock the potential of channel-wide content analysis!</p><p>For this tutorial, all you’ll need is a YouTube Data API v3 key to download the <em>api_crawler</em> package by using <strong>`pip install api_crawler`</strong>.</p><p><strong>PS: </strong><a href="https://github.com/luc-pimentel/api_crawler/blob/7f0e0ff3eb1cbf28bcd0c5896ce32861cf4628fb/docs/Youtube%20Data%20Extraction.ipynb">Visit this link for the code notebook</a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*GUSnB8N--R4vDctJeC0Aeg.jpeg" /></figure><h3>Setting Up</h3><p>First, let’s import the necessary libraries and set up our API key:</p><pre>from api_crawler import YoutubeAPI<br>import time<br>import pandas as pd<br><br><br># OpenAI youtube channel ID<br>channels_to_monitor = [&#39;UCXZCJLdBC09xxGZ6gcdrc6A&#39;]<br><br>YOUTUBE_V3_API_KEY = &#39;YOUR_YOUTUBE_DATA_V3_API_KEY&#39;<br><br>youtube_api = YoutubeAPI(api_key=YOUTUBE_V3_API_KEY)</pre><p>Make sure to replace <strong>’YOUR_YOUTUBE_DATA_V3_API_KEY’ </strong>with your actual API key.</p><h3>Fetching Videos from the Channel</h3><p>Now that we’ve set up our environment, let’s start by fetching all the videos from our target channel:</p><pre>videos = youtube_api.get_videos_from_channel(channels_to_monitor[0], get_full_info=True)</pre><p>This code retrieves all videos from the specified channel, including detailed information for each video.</p><h4><strong>Collecting Comments for Each Video</strong></h4><p>Next, we’ll iterate through each video and collect all comments and replies:</p><pre>for i, video in enumerate(videos):<br>    comments = youtube_api.get_all_comments(video[&#39;id&#39;], include_metadata=True, include_replies = False)<br><br>    comment_threads = [comment[&#39;id&#39;] for comment in comments if comment.get(&#39;snippet&#39;, {}).get(&#39;totalReplyCount&#39;, 0) &gt; 0]<br>    <br>    for comment_thread in comment_threads:<br>        youtube_api.get_all_comment_replies(comment_thread)<br><br>    if i % 25 == 0:<br>        print(f&#39;Processed {i} videos&#39;)<br>        time.sleep(60*3)</pre><p>This loop collects all comments for each video and fetches replies for comments that have them. We’ve included a pause every 25 videos to avoid hitting API rate limits (which happens quite frequently if you’re making too many requests).</p><h4><strong>Processing the Collected Data</strong></h4><p>After collecting the data, we’ll process it into pandas DataFrames for easy analysis.</p><p>Notice that, in the code above, I am not saving the output to a list. And that’s because the `api_crawler` package already saves the output to a json file to serves as a data lake.</p><p>You could save the output to a list if you wanted, and that would not be wrong, but the entire context and metadata would be lost whenever you leave the notebook. That’s why I prefer to use the local json data lake to save the output.</p><p>In the code below, we will read the json data lake and process the data into a pandas DataFrame:</p><pre>from api_crawler.data_lake import read_log<br><br>videos_log = [video <br>              for output in read_log(&#39;lake/json_lakes/YoutubeAPI_get_videos_from_channel.json&#39;)<br>              if output.get(&#39;output&#39;) is not None<br>              for video in output[&#39;output&#39;]]<br><br><br>videos_df = pd.json_normalize(videos_log)[[&#39;id&#39;, &#39;channel.id&#39;, &#39;channel.name&#39;,&#39;title&#39;, &#39;publishDate&#39;, &#39;duration.secondsText&#39;, &#39;viewCount.text&#39;]]</pre><h4><strong>Process comment data</strong></h4><pre>comments_log = [comment <br>                for output in read_log(&#39;lake/json_lakes/YoutubeAPI_get_all_comments.json&#39;)<br>                if output.get(&#39;output&#39;) is not None<br>                for comment in output[&#39;output&#39;]]<br><br>comments_df = pd.json_normalize(comments_log)[[&#39;id&#39;, &#39;snippet.videoId&#39;,&#39;snippet.topLevelComment.snippet.authorDisplayName&#39;,<br>                                               &#39;snippet.topLevelComment.snippet.textDisplay&#39;, &#39;snippet.topLevelComment.snippet.publishedAt&#39;,<br>                                               &#39;snippet.topLevelComment.snippet.likeCount&#39;, &#39;snippet.totalReplyCount&#39;]]</pre><h4><strong>Process comment replies data</strong></h4><pre>flattened_replies = [reply <br>                    for comment in read_log(&#39;lake/json_lakes/YoutubeAPI_get_all_comment_replies.json&#39;)<br>                    for reply in comment.get(&#39;output&#39;, []) if comment.get(&#39;output&#39;)]<br><br>comment_thread_replies_df = pd.json_normalize(flattened_replies)[[&#39;id&#39;, &#39;snippet.parentId&#39;,&#39;snippet.authorDisplayName&#39;,<br>                                                                  &#39;snippet.textOriginal&#39;, &#39;snippet.publishedAt&#39;, &#39;snippet.likeCount&#39;]]</pre><h4><strong>Saving the Results</strong></h4><p>Finally, we’ll save our processed data to Excel files for further analysis:</p><pre>videos_df.to_excel(&#39;openai_yt_videos.xlsx&#39;)<br>comments_df.to_excel(&#39;openai_yt_comments.xlsx&#39;)<br>comment_thread_replies_df.to_excel(&#39;openai_yt_comment_thread_replies.xlsx&#39;)</pre><p>And there you have it!</p><p>You’ve successfully scraped all videos and comments from a YouTube channel. The data is now ready for your analysis, whether you’re looking at content trends, engagement metrics, or performing sentiment analysis on comments.</p><p>For any channel you want to scrape, you can just replace/add the channel ID in the `channels_to_monitor` list and run the code.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b97886bf72a6" width="1" height="1" alt=""><hr><p><a href="https://medium.com/ai-creators/youtube-data-extraction-easily-scrape-all-videos-and-comments-from-a-channel-b97886bf72a6">Youtube Data Extraction: Easily Scrape All Videos And Comments From A Channel</a> was originally published in <a href="https://medium.com/ai-creators">AI Creators</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Original Content With AI — The Only Tool You Need]]></title>
            <link>https://medium.com/ai-creators/original-content-with-ai-the-only-tool-you-need-cb79fccf1233?source=rss-a99bef0c8cc6------2</link>
            <guid isPermaLink="false">https://medium.com/p/cb79fccf1233</guid>
            <category><![CDATA[content-marketing]]></category>
            <category><![CDATA[content-creation]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[social-media-marketing]]></category>
            <category><![CDATA[ai-content-creation]]></category>
            <dc:creator><![CDATA[Luke AI Dev]]></dc:creator>
            <pubDate>Sun, 24 Mar 2024 22:02:59 GMT</pubDate>
            <atom:updated>2024-03-24T22:02:59.777Z</atom:updated>
            <content:encoded><![CDATA[<h3>Original Content With AI — The Only Tool You Need</h3><p>Content creation with AI can often come across as robotic and impersonal.</p><p>Because of this, most content creators still struggle to keep their audience engaged all the while fully leveraging this technology.</p><p>After all, the stuff that makes us humans — our unique tone of voice, ideas, and perspectives — still cannot be replicated by AI.</p><p>But just because AI can’t fully replicate these qualities doesn’t mean you should avoid using it altogether.</p><p>So please allow me to introduce you to the only app you will ever need to create original content with AI at scale.</p><p>And the best of all, it’s free and open source.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ZJxhue_CDi3i4riVaLo3VQ.jpeg" /></figure><h3><strong>How This App Works</strong></h3><p>You see, I’ve been advocating for a<strong> “voice-first” approach to content creation</strong>…</p><p>1. You start rambling about your ideas and point of views;<br>2. Use AI to get a transcription;<br>3. Use AI to structure the transcription into a coherent blog post.</p><p>This way, you can get your ideas out of your head as fast as humanly possible, and then use AI only to structure and organize them.</p><p>Thus keeping your authentic voice and personality intact.</p><p>What’s great about the voice-first content creation approach is that you don’t really need a lot of AI tools or fancy software. All you need is:</p><p>• OpenAI’s Whisper AI for voice-to-text transcription;</p><p>• A Large Language Model like Claude 3 or ChatGPT to help organize and clarify your transcribed thoughts.</p><p>I was using ChatGPT, but have recently switched to Claude 3 and noticed a great improvement.</p><p>But honestly, any large language model will do.</p><p>Again, this workflow is optimized for speed and originality.</p><p>We’re not using this to create low-quality SEO articles. The goal is to produce original, attention-grabbing content for your audience.</p><h3>Free Open-Source Python App</h3><p>Now, most AI content creation apps overcomplicate things with useless features.</p><p>That’s why I created a simple app with only the essential tools for AI voice-first content creation.</p><p>• Record your voice and get a transcription</p><p>• Have AI edit it in coherent format.</p><p>However, I realized most people don’t know how to code or don’t want to. So, I decided to make this app open source and completely free.</p><p>With this Python app, you can easily leverage the “voice-first” content creation workflow and create engaging content as fast as humanly possible.</p><p>Here’s how to get started:</p><p>Visit this <a href="https://github.com/luc-pimentel/VoiceContent-AI">GitHub repo</a> and clone it, then follow the step-by-step guidelines outlined in the README.</p><p>The app has three pages:</p><ul><li>The first is the homepage, which briefly explains how the app works, similar to the explanation I’m giving you now.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1018/1*9qpvc_pmRCI_EqThR_TJIA.gif" /><figcaption>Home page of the App</figcaption></figure><ul><li>The second page is the “content editor” page, where you’ll paste your rough drafts or ideas, and GPT will refine and clarify your writing.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1018/1*9BEr34Ce5uWyHVHzKofjtg.gif" /><figcaption>Edit your transcript into a coherent piece</figcaption></figure><ul><li>The third page is the “Transcribe Audio” page. Here, you can record yourself rambling out loud, and the app will transcribe your spoken words into text.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1018/1*E47lyj7XDGVoDeOF-fq2xg.gif" /><figcaption>Click to record and start rambling</figcaption></figure><p>To customize the underlying prompt for more specific needs, check the .env file. You can tweak the prompt there to tailor the content creation process.</p><p>Also, don’t forget to enter your Anthropic and OpenAI API keys in the .env file. This will allow you to access and use both models.</p><p>That’s all there is to it.</p><p>You’ll save a lot of money on AI tools and — most importantly — a lot of time creating content.</p><p>All you need are API keys for Anthropic’s AI and OpenAI. Even if you create a ton of content per month, you’ll rarely spend more than $5 in my experience.</p><p>I hope this open-source app proves useful for you. <a href="https://github.com/luc-pimentel/VoiceContent-AI">https://github.com/luc-pimentel/VoiceContent-AI</a></p><p>Take care!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=cb79fccf1233" width="1" height="1" alt=""><hr><p><a href="https://medium.com/ai-creators/original-content-with-ai-the-only-tool-you-need-cb79fccf1233">Original Content With AI — The Only Tool You Need</a> was originally published in <a href="https://medium.com/ai-creators">AI Creators</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How To Use Claude 3 on Cursor AI]]></title>
            <link>https://medium.com/ai-creators/how-to-use-claude-3-on-cursor-ai-855edee19c6e?source=rss-a99bef0c8cc6------2</link>
            <guid isPermaLink="false">https://medium.com/p/855edee19c6e</guid>
            <category><![CDATA[ai-coding]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[claude-ai]]></category>
            <category><![CDATA[anthropic-ai]]></category>
            <dc:creator><![CDATA[Luke AI Dev]]></dc:creator>
            <pubDate>Sat, 16 Mar 2024 18:09:37 GMT</pubDate>
            <atom:updated>2024-03-16T18:09:37.316Z</atom:updated>
            <content:encoded><![CDATA[<p>Everyone’s talking about Claude 3.</p><p>They say it’s better than ChatGPT 4 for coding.</p><p>But if you’re using Cursor AI as your main IDE, you probably haven’t tried it yet.</p><p>So let me show you how to configure Cursor to use Claude 3 instead of GPT.</p><p>You don’t even need to make an Antropic account. (Which is great news for those not in the US)</p><p>Let’s dive in.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*GUSnB8N--R4vDctJeC0Aeg.jpeg" /></figure><h3>1. <strong>Use OpenRouter AI</strong></h3><p>OpenRouter AI is an AI model aggregator.</p><p>You can access models like GPT-4 32K, Google PaLM, Anthropic via their API.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*qOGdr-lRe4g-y3a96qTZgA.gif" /></figure><h3>2. <strong>Fund Your Open Router Account</strong></h3><p>Next step is to add funds to your OpenRouter account just as you would to any AI API service.</p><p>Keep in mind that you’re not paying for any subscription service.</p><p>You’re paying for how much you use it, just like if you were using Anthropic AI directly.</p><h3>3. <strong>Get Your API Key</strong></h3><p>After funding your account, the next step is to get your API key.</p><p>On the top right corner, go to: Account -&gt; Keys -&gt; Create Key.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*1qrJnSZjgGg1zfsX7FAecQ.gif" /></figure><h3>4. <strong>Go to Cursor AI and Open the Settings</strong></h3><p>Next, head over to Cursor AI and open the AI settings.</p><p>This is where you’ll configure the available models to include Claude 3 models.</p><p>Scroll down in the Cursor AI settings, and click on the “Configure Models” tab.</p><p>Then, click on “Add Models” -&gt; Override OpenAI base URL and add the the following URL:</p><p><a href="https://openrouter.ai/api/v1">https://openrouter.ai/api/v1</a></p><p>Next, click on add models and add the following models:</p><p><strong><em>• anthropic/claude-3-opus<br>• anthropic/claude-3-haiku<br>• anthropic/claude-3-sonnet</em></strong></p><p>It’s crucial that you use the exact name for both the model and the base URL.</p><p>Because this is how Cursor AI will connect to the API endpoint.</p><p>Don’t change a single letter. If you do, the API connection will fail and you won’t be able to use the model.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/505/1*m5PZ4XpGr_qdHqxEz8CYZg.gif" /></figure><p>5. Enter Your OpenRouter API key</p><p>Now, enter the API key you received from OpenRouter below the OpenAI API key section, and click to test.</p><p>Then, you’ll be ready to start using Claude3 through CursorAI.</p><p>This method also works for any other model you want to try, as long as it’s available on OpenRouter.</p><p>So whenever there is a new important AI release, integrating it with Cursor will be very easy.</p><p>And that’s it for this tutorial!</p><p>If you liked it, you’ll also like my <a href="https://lucasaibuilder.gumroad.com/l/ultimate_prompting">Ultimate Prompting Guide.</a></p><p>Where I will teach you how to create highly effective prompts and AI automations.</p><p><strong>Here’s what you will learn:</strong></p><ul><li>Easy step-by-step approach to create highly-reliable super prompts.</li><li>The right way to create prompts that are highly contextual to you;</li><li>Discover the best prompt engineering guidelines available today;</li><li>The simplest step-by-step approach to create AI automation workflows;</li><li>How to create AI apps with Python</li></ul><p>Please check it out: <a href="https://lucasaibuilder.gumroad.com/l/ultimate_prompting">https://lucasaibuilder.gumroad.com/l/ultimate_prompting</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=855edee19c6e" width="1" height="1" alt=""><hr><p><a href="https://medium.com/ai-creators/how-to-use-claude-3-on-cursor-ai-855edee19c6e">How To Use Claude 3 on Cursor AI</a> was originally published in <a href="https://medium.com/ai-creators">AI Creators</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Professional Photos In Under 5 Minutes Wi]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/ai-creators/professional-photos-in-under-5-minutes-wi-297c0987d7fe?source=rss-a99bef0c8cc6------2"><img src="https://cdn-images-1.medium.com/max/600/1*pTyzaHTffOXZU5ICgMUjjQ.jpeg" width="600"></a></p><p class="medium-feed-snippet">Generate professional looking photos in under 5 minutes&#x2026;</p><p class="medium-feed-link"><a href="https://medium.com/ai-creators/professional-photos-in-under-5-minutes-wi-297c0987d7fe?source=rss-a99bef0c8cc6------2">Continue reading on AI Creators »</a></p></div>]]></description>
            <link>https://medium.com/ai-creators/professional-photos-in-under-5-minutes-wi-297c0987d7fe?source=rss-a99bef0c8cc6------2</link>
            <guid isPermaLink="false">https://medium.com/p/297c0987d7fe</guid>
            <category><![CDATA[photography]]></category>
            <category><![CDATA[branding]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Luke AI Dev]]></dc:creator>
            <pubDate>Thu, 08 Feb 2024 01:56:02 GMT</pubDate>
            <atom:updated>2024-03-02T21:56:55.058Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[My AI Content Creation Setup]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/ai-creators/my-ai-content-creation-setup-f5be1e42397f?source=rss-a99bef0c8cc6------2"><img src="https://cdn-images-1.medium.com/max/1728/1*qrC9-4mP6bCkZzUukTsngQ.jpeg" width="1728"></a></p><p class="medium-feed-snippet">My Setup To Become a Content Machine</p><p class="medium-feed-link"><a href="https://medium.com/ai-creators/my-ai-content-creation-setup-f5be1e42397f?source=rss-a99bef0c8cc6------2">Continue reading on AI Creators »</a></p></div>]]></description>
            <link>https://medium.com/ai-creators/my-ai-content-creation-setup-f5be1e42397f?source=rss-a99bef0c8cc6------2</link>
            <guid isPermaLink="false">https://medium.com/p/f5be1e42397f</guid>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[content-marketing]]></category>
            <category><![CDATA[content-creation]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Luke AI Dev]]></dc:creator>
            <pubDate>Sat, 27 Jan 2024 20:28:47 GMT</pubDate>
            <atom:updated>2024-03-02T21:58:29.271Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Most AI Output Is Hallucination]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/ai-creators/most-ai-output-is-hallucination-fb7d6a2b78cd?source=rss-a99bef0c8cc6------2"><img src="https://cdn-images-1.medium.com/max/1152/1*MV4ZNJhRMAsOrWQ3eDzbfA.jpeg" width="1152"></a></p><p class="medium-feed-snippet">In this post I want to talk about an important issue with ChatGPT and LLMs in general that&#x2019;s often overlooked.</p><p class="medium-feed-link"><a href="https://medium.com/ai-creators/most-ai-output-is-hallucination-fb7d6a2b78cd?source=rss-a99bef0c8cc6------2">Continue reading on AI Creators »</a></p></div>]]></description>
            <link>https://medium.com/ai-creators/most-ai-output-is-hallucination-fb7d6a2b78cd?source=rss-a99bef0c8cc6------2</link>
            <guid isPermaLink="false">https://medium.com/p/fb7d6a2b78cd</guid>
            <category><![CDATA[prompt-engineering]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[chatgpt]]></category>
            <dc:creator><![CDATA[Luke AI Dev]]></dc:creator>
            <pubDate>Fri, 26 Jan 2024 13:57:24 GMT</pubDate>
            <atom:updated>2024-01-26T13:57:24.394Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Writing Content Should Be The Easiest Task Of Your Day]]></title>
            <link>https://medium.com/ai-creators/writing-content-should-be-the-easiest-task-of-your-day-c52e434c5650?source=rss-a99bef0c8cc6------2</link>
            <guid isPermaLink="false">https://medium.com/p/c52e434c5650</guid>
            <category><![CDATA[content-creation]]></category>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[content-marketing]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Luke AI Dev]]></dc:creator>
            <pubDate>Thu, 25 Jan 2024 18:53:45 GMT</pubDate>
            <atom:updated>2024-03-02T21:59:09.489Z</atom:updated>
            <content:encoded><![CDATA[<p>Writing content used to feel like a drag. Time consuming and mentally exhausting.</p><p>I felt like quitting and burning out more times than I’d like to admit.</p><p>But since I’ve automated much of my content creation workflow with AI, I cut down the time I spend writing to a fraction of what it used to be.</p><p>To the point that it became the easiest task of my day. And it finally became a fun activity that I look forward to.</p><p>And it’s not like I am just writing a bunch of fluff, quite the contrary…</p><p>I am putting out content that is interesting, original, high-quality, and that you can build an audience on.</p><p>So in this post, I’m going to explain why I believe writing should be the easiest part of your day if you don’t want to burn out from the content game.</p><p>And I will also share my AI workflow so you can achieve the same.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xdmCRzBQC2lphuprgMWcjw.jpeg" /></figure><h3>“I Hate Writing, But I Love Having Written”</h3><p>Writing is a mentally exhausting task like no other. After a good writing session, you are left with no creative energy for anything else.</p><p>And if you’re anything like me, you’ll need a good break, a nap, or some good ol’ caffeine (or all three) after an intense writing session.</p><p>This is clearly not sustainable if you have other obligations besides creating content.</p><p>After all, your writing habits shouldn’t get in the way of you working on your business, your job, and honing your skills.</p><p>However, writing is important if you want to have a voice in the world today, so you also don’t want to stop writing all together.</p><p>But thanks to the AI automations I built, I no longer face this trade-off. Now, I can create a ton of content effortlessly.</p><p>For instance, my publishing schedule now consists of:</p><ul><li>At least 4 long-form pieces of content every week;</li><li>1 newsletter article</li><li>Multiple tweets per week;</li><li>2 new modules to my product per week;</li></ul><p>That on top of my day job. So writing and publishing is not intererting with my other obligations.</p><h3>My Best AI Content creation Workflow</h3><p>You see, my AI content creation workflow is designed for quick and original content creation, while requiring minimal creative effort from me.</p><p>In other words, I optimized it for laziness without losing originality.</p><p>And if you’re a writer who feels exhausted after a writing session, you’ll appreciate this workflow.</p><h3>1. Record Yourself Talking</h3><p>To start, I turn on the microphone and begin speaking freely.</p><p>At this point, I have nothing more than a brief outline just to keep track of the points I want to cover in the article.</p><p>I don’t worry about structure at this stage; I just turn on the mic and talk extensively about the topic I want to cover.</p><p>You see, most people speak at a rate of 150 words per minute, while they type at only 40~50 words per minute (and feel drained afterwards)…</p><p>So by speaking instead of writing, you can effectively triple your writing speed all the while saving most of your mental energy.</p><h3>2. Get a Transcription Of Your Recording</h3><p>The second step is to transcribe your spoken words using AI.</p><p>This is finally where AI comes into play in our content creation process.</p><p>You’ll use AI to get an accurate transcription of what you’ve just said. There are many speech-to-text AI tools available.</p><p>I use Whisper, which is an OpenAI’s speech-to-text model. If you use the GPT Mobile app, you can access Whisper through the built-in microphone of the GPT mobile app.</p><p>You can also use the API if you know how to code.</p><p>You can use other speech-to-text AI tools too; the important thing is to get a transcription of what you’ve said.</p><h3>3. Format Everything Using GPT</h3><p>The third step is to take your incoherent ramblings and send it to ChatGPT to make it coherent.</p><p>As you can see, by the third step you will get a pretty good blog post almost as fast as you could get your thoughts out.</p><h3>4. Edit</h3><p>Then, the last step is to read through the post and make minor edits.</p><p>After all, the final piece won’t be perfect, and you might want to add or remove things.</p><p>However, you’ll rarely need more than two rounds of editing, and since GPT has already structured most of the transcription, each round will be very quick.</p><p>And after that, you’re ready to publish.</p><p>For reference, this workflow lets me publish 1000+ words articles within an hour. And if I let go of my perfectionism in editing, it could be even faster.</p><p>Now compare it with the typical workflow without AI, where one would spend at least 6+ hours to write a thousand words blog post from scratch.</p><h3>Words Are Abundant, Ideas Are Scarce</h3><p>All in all, I’ve learned that if writing is the hardest and most time-consuming part of your job, you’re not working as efficiently as you could be.</p><p>Creating content should be about taking ideas from your mind and sharing them with the world in an engaging and captivating way.</p><p>While writing is crucial and a part of the whole content creation process, it shouldn’t be the main focus.</p><p>I’ve found that I should spend most of my time as a creator generating ideas, indulging my curiosity, and interacting with clients to understand their needs and challenges.</p><p>In short, I should dedicate my time to uncover new information and come up with ideas. And when I have that “eureka” moment, I start writing as a way to document it.</p><p>If writing drains all your creative energy (as it does with me), you are not generating new interesting ideas. A route that will lead to stagancy an burn out sooner or later.</p><p>And that’s all for today’s article!</p><p>If you liked it, you’ll also like my <a href="https://lucasaibuilder.gumroad.com/l/ultimate_prompting">Ultimate Prompting Guide.</a></p><p>Where I will teach you how to create highly effective prompts and AI automations.</p><p><strong>Here’s what you will learn:</strong></p><ul><li>Easy step-by-step approach to create highly-reliable super prompts.</li><li>The right way to create prompts that are highly contextual to you;</li><li>Discover the best prompt engineering guidelines available today;</li><li>The simplest step-by-step approach to create AI automation workflows;</li><li>How to create AI apps with Python</li></ul><p>Please check it out: <a href="https://lucasaibuilder.gumroad.com/l/ultimate_prompting">https://lucasaibuilder.gumroad.com/l/ultimate_prompting</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c52e434c5650" width="1" height="1" alt=""><hr><p><a href="https://medium.com/ai-creators/writing-content-should-be-the-easiest-task-of-your-day-c52e434c5650">Writing Content Should Be The Easiest Task Of Your Day</a> was originally published in <a href="https://medium.com/ai-creators">AI Creators</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>