<?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 Chris Tomich on Medium]]></title>
        <description><![CDATA[Stories by Chris Tomich on Medium]]></description>
        <link>https://medium.com/@chris.tomich?source=rss-c7024afa133c------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*DWPkZSj5yK9vXEOCTE1AHw.jpeg</url>
            <title>Stories by Chris Tomich on Medium</title>
            <link>https://medium.com/@chris.tomich?source=rss-c7024afa133c------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 22:36:50 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@chris.tomich/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[How AI has turned me from hating TDD to loving TDD]]></title>
            <link>https://medium.com/@chris.tomich/how-ai-has-turned-me-from-hating-tdd-to-loving-tdd-d0a1cdfc1d4f?source=rss-c7024afa133c------2</link>
            <guid isPermaLink="false">https://medium.com/p/d0a1cdfc1d4f</guid>
            <category><![CDATA[test-driven-development]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[unit-testing]]></category>
            <dc:creator><![CDATA[Chris Tomich]]></dc:creator>
            <pubDate>Sun, 25 Feb 2024 02:04:27 GMT</pubDate>
            <atom:updated>2024-02-25T02:04:27.342Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*9COJ051KBBrlFZZuLfieJg.png" /><figcaption>Human boxer training with AI boxer — Created with Copilot / DALL-E 3</figcaption></figure><p>For a long time now, I’ve firmly been on the hater side of the TDD debate. It’s just never really fit with how I work. My way of working is to understand a problem by writing the wrong solution as quickly as possible and iterating early and often till I have the right solution. TDD has always felt like overhead in this process. Stopping to write a test rather than spending time on writing a solution feels like wasted time especially when I know the test is wrong and will have to be re-written. It feels faster to print to stdout/stderr and rely on myself to be the “unit test”.</p><h3>Writing unit tests is so boring…</h3><p>So here’s a hot take, if it’s not required (by my client or by my employer) I don’t even like to do unit testing. I think most people probably feel this way. Building a solution is the fun part. Writing the tests afterwards is the boring part and I’m only doing it if I’m getting paid to do it.</p><p>One of the arguments I’ve heard for TDD is this idea that if you write unit tests while creating your code, unit testing becomes an integral part of your workflow and so you don’t have a “unit test writing” phase and so it doesn’t feel boring. If this works for you, great, but for me anytime writing tests is time I could’ve spent writing more features.</p><h3>The turning point</h3><p>If you just crawled out of a cave, you wouldn’t have yet seen the reports that AI is here and that we’re all going to lose our jobs. In particular software engineers have engineered their own downfall. The truth is AI, and more specifically LLMs, aren’t even close to being that good. Although one thing we’ve all found is that they really excel at writing boiler plate. For me specifically, it’s been incredible for writing unit tests.</p><p>In both my job and my own personal projects, I’ve been using LLM-based editor extensions to help me write the majority of my unit tests. It’s not perfect and I often need to tweak the unit tests that are generated, but it reduces huge amounts of manual copy/paste/edit. The time reduction is so huge I have finally embraced TDD!</p><p>But more than the time reduction, because I’m using natural language to tell the LLM extensions what I want the unit test to do, TDD is finally feeling natural to me. As I mentioned at the beginning of this article, my workflow is to build the wrong solution first. In the past I would have done just that. The difference now is that I’ll build my skeletal entry point (like a method or class) and then I’ll ask the AI extension to build a unit test for this with the list of behaviours my code should exhibit.</p><p>In a recent task at work, I needed to build a feature in Go that would accept configuration and an instance of a third party SDK. The configuration would contain a mixture of locations for data, either in-memory, on the file system, or on the network, and this data would be used to configure the third party SDK.</p><p>I started by writing an empty method that would accept the in-memory data and the object representing the third party SDK. I then directed the LLM to create me a unit test that would call this method and pass in my sample data as well as a new instance of the third party SDK, and then check that the output object had particular settings set that were similar to the input data. I did have to be explicit about some of the names on the input and output objects, and there was a little tweaking that came with the final output, but for the most part, it did a pretty good job given very little context about my problem and the third party SDK.</p><h3>Whoa!</h3><p>The first “whoa!” moment for me was during the iterations on my solution. As I built more of my solution, I would ask the LLM to update the unit test providing it both the new code I had written plus the old unit test. The results were incredible. I found very little need to tweak the subsequent unit tests.</p><p>The second “whoa!” moment was when I needed to write additional tests for edge cases. With the solution code and existing unit tests as an example, I literally asked the LLM “in the style of the following unit test, could you create new unit tests that tests &lt;new edge case&gt;”. It then proceeded to spit out the unit tests almost near perfectly. Of course for some of the more complicated edge cases still needed tweaking, but the effort on my part was minimal.</p><p>I think the most positive thing in this whole experience is that because they were tests, most of the time when something was wrong either in the test or my solution, it would cause the test to fail. Debugging the test would then either lead me to learn something more about my solution or it would lead me to fixing the test. Working in this way, I finally felt that TDD supported by LLMs was finally providing value to me and it was no longer boring because I was always learning something new about the problem I was solving.</p><h3>TDD is supercharged by AI</h3><p>IMO the very nature of the unit tests in TDD is to be adversarial to the code being written. It’s like playing chess against yourself. I know some people are smart enough to use this as a way to challenge yourself. For the kind of person I am, it just feels pointless and boring to fight myself. I don’t think I’m alone.</p><p>With LLMs I now have an opponent to write my tests and challenge my designs. I discuss with it the rules of engagement but it does all the grunt work. This backwards and forwards is no longer just me playing all sides, it’s me playing with a slightly more ignorant opponent. But like in any debate, more ignorant opponents can sometimes challenge your assumptions and that leads to your own growth.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d0a1cdfc1d4f" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How AI Could Ruin the Software Industry (and How to Shield Yourself)]]></title>
            <link>https://medium.com/@chris.tomich/how-ai-could-ruin-the-software-industry-and-how-to-shield-yourself-63462f092fd9?source=rss-c7024afa133c------2</link>
            <guid isPermaLink="false">https://medium.com/p/63462f092fd9</guid>
            <category><![CDATA[programming]]></category>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Chris Tomich]]></dc:creator>
            <pubDate>Sun, 03 Sep 2023 09:59:02 GMT</pubDate>
            <atom:updated>2023-09-03T09:59:28.045Z</atom:updated>
            <content:encoded><![CDATA[<p>I know it’s a sensational title, but stay till the end and maybe you’ll have observed the same problems. The problem I’m referring to is that each successive generation of engineers understands less and less about the platforms they’re working on and how to learn resilience. Everything in this article is anecdotal but I don’t think I’m alone in the experience.</p><h3>The Tooling Problem</h3><p>Although I started working as a software developer around 2002, I had been learning how to write software through high school in the 90s. It had always been my dream to become a PC game developer (dreams change) so I did my best to try and learn and understand C++, Win32 APIs, OpenGL, and DirectX.</p><p>The benefit of that dream was that I forced myself to learn quite a bit about how PC hardware operates, how the OS interacts with that hardware, and how I need to work with it from userland. In hindsight, it’s easy to see where I had a lot of misunderstandings, but nonetheless the experience of research and learning has had a big impact on how I now operate as a software engineer.</p><p>Fast forward to today, and I had an interesting conversation with one of the younger engineers I work with. We were discussing app development using Electron and how JavaScript and HTML work within the browser DOM. We then got onto the subject of other tools like Flutter and .NET MAUI and their performance. It was at this point the engineer continued to talk about the browser DOM at which point I explained those frameworks aren’t browser based and had their own rendering engines. This seemed a bit confusing to the engineer.</p><p>After a bit of backwards and forwards, it turns out they had never learnt (or at least didn’t remember) about how graphics are rendered to the screen. When I asked them follow-up questions about kernel land and userland, they admitted they knew very little, if anything, about how the OS operates and interacts with applications.</p><p>This isn’t an isolated incident. One of the grads we hired in a previous round also exhibited similar gaps in knowledge. In my discussions with both engineers, I’ve noticed a commonality in their experience at university. Both went to two different universities in two different cities but both had projects working with 2D/3D that involved using tools like Blender or WYSIWYG tooling rather than being forced to put pixels on a screen like I had experienced during my university days.</p><p>As universities are forced by market forces and industry to prepare graduates for the workforce, fundamental knowledge is often skipped in favour of tooling. This isn’t new. When I got out of university, I had very little understanding about how to use assembly at the occasional ire of the senior engineers I was working with at the time. Most of my university course was using Java with the occasional use of C, it only touched on assembly for very brief moments and not enough to be useful in a professional environment.</p><p>Project this tooling problem out another 10 years and as the students of today are taught to be “prompt engineers” with AI, it’s likely they’ll have even greater gaps in their knowledge. Heavy reliance on a tool like ChatGPT to generate working code and to debug problematic code will become an even greater crutch as students will never have had to experience solving a complicated logic error using a debugger and println.</p><h3>The Learning Problem</h3><p>The tooling problem is a big one, but I think generative AI tools bring an even bigger problem with them. As many have said, generative AI tools are not like any software tool we’ve ever seen in the past. I’ve spoken previously about how they won’t replace creativity, but their ability to complete menial coding tasks within context of the described problem domain is incredible. This leap in effectiveness comes as a hidden cost to engineers, particularly young inexperienced engineers.</p><p>As I was often told by my parents and teachers as a young child, your brain is like any muscle in your body. If you don’t train it, it’ll be weak. People go to the gym to put in the hours to do sets and reps to get the body and fitness level they want. The same applies to software engineering. It may not be glamorous, but putting in the hours of debugging, reading through (passive aggressive) Stack Overflow posts, trawling through blog posts, and even trying to think of different search terms for Google, helps build those engineering brain muscles. AI tools like ChatGPT take all that away.</p><p>A common experience I’ve had with younger engineers is “how did you know to search for that in Google?”. As I’ve explained many a time, it’s only through continuous researching and learning of concepts tangental to the problem I’m actually trying to solve that I’ve been able to train the brain muscles. When the need arises to tackle a complicated problem, I have a large enough breadth of knowledge to call upon to help understand and analyse the problem and to then synthesise a solution.</p><p>The problem with AI and younger engineers is that they aren’t going to ever experience that breadth of research the majority of the time. This will be compounded with generative AI as they’ll also have less experience synthesising a solution. The menial programming tasks they will be assigned will very easily be solved by AI. They’ll feel competent and their managers will think they’re competent. The problem is, when something becomes mildly difficult and requires a creative solution, they’ll be as stumped as the AI and without the “brain muscles” to do the heavy lifting.</p><h3>The Solution — Embrace the Struggle</h3><p>If you project my conclusions further into the future, it leads to a pretty dire situation for the software industry as a whole. We end up in a situation in an even worse resourcing situation than we are currently in. We already have difficulty hiring graduates that are good at problem solving, but what if we’re at a point where only a few can problem solve because for most they’ve been able to rely on ChatGPT.</p><p>This all just leads to the further infantilisation of our graduates. It will take longer for them to learn how to solve problems and synthesise solutions. The infantilisation of the youth isn’t just confined to the software industry. I’ve experienced this first hand more generally in my own life. My parents already had a home in their 20s. It’s taken most of my generation till their 30s to get a home. The next generations will probably take till their 40s.</p><p>For the younger readers of this article, my hope is that you’ll take away the importance of a “learning journey”. In every minor struggle, there is a lesson to be learnt and you’ll come out the other side better for it. Even if it’s just something as minor as having to read through a few blog posts before you get an answer or taking an extra day to find a solution.</p><p>For the experienced readers of this article, my advice is to encourage the juniors to appreciate the struggle and embrace it. We also have a responsibility to build that struggle into our estimates. Failing to give our juniors the time and space to experience the same learning journeys we were afforded means we’re failing ourselves in the long run. There are no shortcuts in engineering, just bad decisions.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/700/1*R1RAkK-rQ3SM_C68DE7ALQ.jpeg" /><figcaption>Me vs My Parents</figcaption></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=63462f092fd9" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Step-by-Step: How to expose a QEMU VM on a CentOS 9 Stream host to the external network?]]></title>
            <link>https://medium.com/@chris.tomich/step-by-step-how-to-expose-a-qemu-vm-on-a-centos-9-stream-host-to-the-external-network-1f481336378e?source=rss-c7024afa133c------2</link>
            <guid isPermaLink="false">https://medium.com/p/1f481336378e</guid>
            <dc:creator><![CDATA[Chris Tomich]]></dc:creator>
            <pubDate>Sun, 12 Mar 2023 06:55:18 GMT</pubDate>
            <atom:updated>2023-03-12T06:57:45.259Z</atom:updated>
            <content:encoded><![CDATA[<p>I’m writing this primarily as a reminder to myself how I did this for future references as I’m bound to forget everything.</p><p>First the context, I have the following environment setup -</p><ul><li>The host OS is CentOS 9 Stream.</li><li>I’ve installed it using the Virtualization Host in the software selection.</li><li>I enabled Cockpit immediately after installation.</li><li>I installed the virtual machines extension for Cockpit.</li><li>I created a new virtual machine through Cockpit with the default options and using a <em>User session</em> connection, not a <em>System</em> connection.</li></ul><p>What follows are the steps I took to enable bridging for the new VM.</p><h3>Step 1 — Create bridge device</h3><p>First thing you need to do is change the main ethernet NIC to be a bridge. You can do that easily by going to the <em>Networking</em> section in Cockpit and clicking on the <em>Add Bridge</em> button as illustrated in the following screenshot.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*mRYFMFxCQC8Za-WLXvbTlA.png" /><figcaption>Adding a new bridge device</figcaption></figure><p>A dialog will appear as illustrated below. Give your new bridge device a name like <em>bridge0</em>. In the ports section select your ethernet device. In my example, it’s <em>eno1</em>. Save the new bridge device.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*YhAleJYMJgO-NNMN0piKQw.png" /><figcaption>Configure new bridge device</figcaption></figure><h3>Step 2 — Create virtual network</h3><p>This is the part that got a bit confusing. The next step is to create a virtual network. As my virtual machine has been configured using a <em>User session</em> connection, I need to create a virtual network also using a <em>User session</em> connection. For some reason, you <strong>can’t</strong> do this from Cockpit in the <em>Networks</em> section found in the <em>Virtual machines</em> tab. The alternative I found was to do it from the console of the host machine. I followed the instructions I found on <a href="https://linuxconfig.org/how-to-use-bridged-networking-with-libvirt-and-kvm#:~:text=Creating%20a%20new%20virtual%20network">this page</a>. For posterity, I’ve replicated those instructions here.</p><p>Create a new <em>bridged-network.xml</em> file with the following contents. If your bridge is named something other than <em>bridge0</em>, replace the bridge name to your bridge device.</p><pre>&lt;network&gt;<br>    &lt;name&gt;bridged-network&lt;/name&gt;<br>    &lt;forward mode=&quot;bridge&quot; /&gt;<br>    &lt;bridge name=&quot;br0&quot; /&gt;<br>&lt;/network&gt;</pre><p>We then use <em>virsh</em> to create a new virtual network with the following command. It’s important not to use <em>sudo</em> here as we want this network to be created in the <em>User session</em>.</p><pre>$ virsh net-define bridged-network.xml</pre><p>Once you’ve done that, we need to activate it and set it to auto-start.</p><pre>$ virsh net-start bridged-network<br>$ virsh net-autostart bridged-network</pre><p>You can verify the network is activated.</p><pre>$ virsh net-list<br>Name              State    Autostart   Persistent<br>----------------------------------------------------<br>bridged-network   active   yes         yes</pre><h3>Step 3 — Update VM NIC</h3><p>Now that the virtual network has been created and bound to the bridge device, we just need to assign the virtual network to our VM. Go to your virtual machine in Cockpit and click <em>Edit</em> on the network interface for the VM in the <em>Network interfaces</em> section. You should see something similar to the following screenshot.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/866/1*NdjJXEWMoGnyRdujpN0mHw.png" /><figcaption>Before adding the VM guest to the bridged virtual network</figcaption></figure><p>In the interface type select the <em>Virtual network</em> option and in the source you should see the option for the virtual network you created in Step 2, select this. In my example it’s <em>bridge0</em>. Save your changes.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/866/1*ZXTUJT36zGVbSuTLHVlgqg.png" /><figcaption>After changing the VM guest NIC to use the bridged virtual network</figcaption></figure><p>At this point you’re almost finished. If you shutdown your VM and start it back up, you’ll now be able to access the guest VM from external systems. The only issue is finding the IP of the guest VM is you’d need to use the terminal in the guest VM to find this as the <em>IP Address</em> column on the Cockpit page for the virtual machine will show <em>Unknown.</em></p><p>Step 4 — Display VM guest IP Address in Cockpit</p><p>After reading through some Github forums, it became apparent that there needs to be an additional service installed to the VM guest called <em>qemu-guest-agent</em>. I found <a href="https://access.redhat.com/solutions/732773#:~:text=Guest%20side%20modifications">these instructions</a> on how to install that agent but found I only need to do the Guest side modifications as Cockpit already configures the host VM to communicate with the <em>qemu-guest-agent</em>.</p><p>Login to the terminal for the guest VM and install <em>qemu-guest-agent</em> with <em>dnf</em> as shown below.</p><pre>$ sudo dnf install qemu-guest-agent</pre><p>Once installed, you’ll need to start and enable it using the following commands.</p><pre>$ systemctl start qemu-guest-agent<br>$ systemctl enable qemu-guest-agent</pre><p>At this point, all you need to do is shutdown your VM and run it again and you should now find you get an IP shown in Cockpit.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/256/1*WRJhxNwiOoJvE7fUjAjOTg.png" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1f481336378e" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[To my fellow engineers: AI art will never be art]]></title>
            <link>https://medium.com/@chris.tomich/to-my-fellow-engineers-ai-art-will-never-be-art-a22828a6e811?source=rss-c7024afa133c------2</link>
            <guid isPermaLink="false">https://medium.com/p/a22828a6e811</guid>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[art]]></category>
            <category><![CDATA[philosophy]]></category>
            <category><![CDATA[arts-and-culture]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Chris Tomich]]></dc:creator>
            <pubDate>Mon, 26 Dec 2022 03:02:23 GMT</pubDate>
            <atom:updated>2022-12-26T03:03:17.637Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/512/1*cpONz-jP6J68-PYdaF3nhg.jpeg" /><figcaption>Created in <a href="https://creator.nightcafe.studio/">https://creator.nightcafe.studio/</a> with the words “something abstract and artistic”</figcaption></figure><p>There’s a tweet that I saw recently which referenced an old video of Hayao Miyazaki, a famous Japanese animator, which started me philosophising about AI and it’s relevance in art. In the video, Miyazaki makes some strong comments about the AI generated footage he is shown to the obvious dismay of the software engineers giving the presentation. Even though I have been like those software engineers sitting across from a potential client being told “your software is garbage” and can empathise with how that may hurt, I definitely agree with Miyazaki — AI art will <strong>never</strong> be art.</p><h3>Tofu on Twitter: &quot;miyazaki of studio ghibli, upon seeing an artificial intelligence presentation- &quot;i strongly feel that this is an insult to life itself&quot;. pic.twitter.com/UmOOWvsfjV / Twitter&quot;</h3><p>miyazaki of studio ghibli, upon seeing an artificial intelligence presentation- &quot;i strongly feel that this is an insult to life itself&quot;. pic.twitter.com/UmOOWvsfjV</p><h3>You’re missing the point</h3><p>I think it’s fair to say, art has never been about the outcome. As my high school literature teacher would often say “how does it make you feel?”. The fact that an artist looks to evoke an emotional response in another is what makes something art.</p><p>The engineers in the video perceive art like writing software. As a software engineer, I am only interested in the outcome. If an artist chose to stop working on their artwork half way, it would still be art. The audience would ask questions like “why did the artist stop?” or “what made them think this was sufficient?”. If I stop coding a solution half way, my solution wouldn’t even work. My audience (my employer or client) would probably fire me.</p><p>The engineers have projected their own beliefs into what they think an artist does. They see it as purely a mechanical skill. Towards the end of the clip, the engineers make it clear that their goal is to build software that can draw like a human. What they’ve failed to understand is all the reasoning by the artist that has gone into the drawing.</p><h3>AI doesn’t reason to create, it seeks your approval</h3><p>When an AI algorithm, like those in the video or more recent advances like ChatGPT, “create” something, they don’t do it with the intended goal of evoking an emotion. Using pre-trained probability distributions made from terabytes of data and a set of rules, they seek to find the most optimal solution with the provided inputs.</p><p>To answer why an AI algorithm created something is actually pretty simple — because we created a probability distribution in n-th dimensional space using training data and this was the “solution” we found in that distribution. Unfortunately this isn’t the actual “why” we’re asking as a human being. That human “why” can’t be answered because current AI algorithms work on probability distributions, not reason.</p><h3>Art is the domain of consciousness</h3><p>Maybe someday humans will create <strong>real</strong> artificial intelligence. An artificial intelligence with real consciousness that will make its own decisions and driven by its own desires. When that day comes and that entity creates art, that will be real art. The questions then will be, “what message is it attempting to convey?” and “could we ever hope to understand that message given our organic origin and its inorganic origin?”. Similar to the questions we ask our human artists today.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a22828a6e811" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[ChatGPT Will Change Software Engineering — But Not in the Way You’re Hoping]]></title>
            <link>https://medium.com/better-programming/chatgpt-will-change-software-engineering-but-not-in-the-way-youre-hoping-2d7dfbc3c6c0?source=rss-c7024afa133c------2</link>
            <guid isPermaLink="false">https://medium.com/p/2d7dfbc3c6c0</guid>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[work]]></category>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[careers]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Chris Tomich]]></dc:creator>
            <pubDate>Sat, 03 Dec 2022 07:07:57 GMT</pubDate>
            <atom:updated>2022-12-04T11:49:54.009Z</atom:updated>
            <content:encoded><![CDATA[<h4>The future is bright for those that adapt</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*R-FeCeOtOX_FpDh-PKJINw.png" /><figcaption><a href="https://chat.openai.com/chat">Source</a></figcaption></figure><p>I’m always dubious of claims made in the media, both social and mainstream, around AI and ML developments. With so much hype over the past decade, many people are.</p><p>Whilst I don’t have a deep background in applying AI and ML models, I understand the fundamentals of the technology better than most. The appreciation of the technical and mathematical achievements in the area is why I feel a lot of the claims made are a little (if not completely) over-exaggerated.</p><p>So when I started seeing comments about ChatGPT over the past day, I felt that these were another exaggeration. As I have an OpenAI account that I opened for Dall-E 2, I thought I may as well give ChatGPT a go and experience the “amazingness” for myself.</p><p>I hate to admit that my cynical “amazement” quickly dissipated to real amazement. ChatGPT really is something incredible and provides a glimpse of how software engineering could be in the future, but not in the way I think some people are thinking.</p><p>In fact for some people, this increase in productivity is going to be so great, they’re actually going to be out of a job.</p><h3>But First Dall-E 2…</h3><p>A <a href="https://www.youtube.com/watch?v=oqamdXxdfSA">YouTube video</a> I watched on Dall-E 2 showed, from the perspective of a designer and artist, how it could change the way they worked. The summary was that it wouldn’t replace designers or artists.</p><p>Instead, it would become an assistance tool that would increase productivity and could help act as a virtual muse for users.</p><p>After watching this, I needed to experience Dall-E 2 for myself and I wanted to see if the video was overhyping the technology. Once I got access, I’ll admit, it was impressive.</p><p>At the same time, I could see that the images being produced from the input I was giving weren’t close to what I was imagining and I could appreciate the points made in the video about how it wouldn’t replace designers and artists but could assist them.</p><p>Even with this experience, I couldn’t fully appreciate the gravity of how this technology might change an artist’s or designer’s way of working. I’ve never worked in any art or design field so I don’t know the first thing about how they work. The way I interacted with Dall-E 2 was an extremely contrived use case as I was just using it to generate narrative-based scenes.</p><h3>ChatGPT and Conversational Software Development</h3><p>At the time of using Dall-E 2, I immediately wondered if the OpenAI team was working on something similar for software development. I wanted to experience an AI-based tool like Dall-E 2 but in a use case, I’m familiar with it.</p><p>With this latest ChatGPT release they were clearly doing that and I’ll say straight away, this really will change the way we work as software developers, and, for some, this could be the beginning of the end of their software development careers.</p><p>On first use, I just did what I saw people doing online — which was to ask ChatGPT contrived questions about software development like building a well known algorithm or asking it questions about code.</p><p>The output from this was sort of impressive but, as an experienced software engineer, I could probably write the solutions just as quickly and would have more confidence about their correctness. The primary algorithms powering all AI tools are pattern based, not rationally or logically based, so there are bound to be errors in the implementations.</p><p>For my next attempt, I chose to approach it from a different angle. I’m currently working on a simple note-taking application in Flutter and I wanted to see how I might do it with ChatGPT. I asked the question “How do I build a simple note-taking application in Flutter?”. I’m assuming that because the OpenAI APIs are rate limited, it only did half the application, but what it did was a good start for a basic note-taking app. Most importantly it completed a number of basic tasks for me -</p><ol><li>Suggestions on a basic starting architecture for my application.</li><li>Suggesting boilerplate code for my application.</li><li>Suggesting names for my classes.</li></ol><p>The value here isn’t huge, and I wouldn’t expect it to be. All of this is something I could’ve done myself, but it does save me a lot of time. On top of this, I could continue “chatting” with ChatGPT and was able to have it build out more features and further refine the generated code. It was like having a slightly stupid human assistant doing the boring coding for me.</p><p>In addition to these basic tasks, like with Dall-E 2 for designers, interacting with it gave me inspiration for other parts of the note-taking app that I hadn’t yet considered. The power of acting as a virtual muse shouldn’t be understated. The concept is so powerful, an old work colleague that used to work at IBM told me how on his first day he was given a teddy bear and was told “This is Ted and he’s a great problem solver.</p><p>If you get stuck on a problem tell Ted and if you haven’t solved it after that, seek help from others.”. This wasn’t a backhanded attempt to tell a newbie not to bother other people. It was teaching the important lesson that sometimes when we narrate and verbalize our problems, the solution can sometimes become apparent. ChatGPT is Ted on steroids!</p><h3>A Glimpse of the Future of Software Development and The Existential Question “Will I Be Needed”?</h3><p>Just this initial interaction shows so much promise for what is to come. I’m starting to see how an AI model is actually providing a more significant positive value to being part of my daily workflow beyond writing simple comparison if statements (one of the “amazing” things I’ve had people show me with Github Co-pilot).</p><p>This now beckons the question, do I think software developers are still going to be needed in the future? The short answer of “definitely” should come as no surprise. The disheartening extension to that answer is that the price of these tools will be the jobs of entry-level and lower-level software developers.</p><p>Amongst software development communities, there is the continuous joke about how Google, StackOverflow, and “copy and paste” write most people’s code. As the old idiom goes, there’s a grain of truth to every joke. I know for a subset of people out there, copy and pasting code from the internet is how they operate daily.</p><p>Infamous events like <a href="https://www.theregister.com/2016/03/23/npm_left_pad_chaos/">“left-pad” with NPM</a> show that many people often choose to outsource basic thinking. The danger with this is not the act of doing it, it’s the reason why. There will be many that do it because it saves time. There will also be many that do it because they don’t know how to do it. It’s this latter group that needs to be worried about these recent advances in AI.</p><p>As these advances in AI progress, the most basic programming tasks, like creating boilerplate classes, implementing algorithms, building simple web forms, doing simple data processing, etc. are increasingly going to be computer-aided. At the moment some of these tasks are time-consuming and companies will hire entry-level or lower level software developers to do this work. This will inevitably change.</p><p>With the rapid progress made by tools like ChatGPT, people with no technical experience, like business analysts or designers, will do this work themselves. The supporting tools like IDEs, SDKs, and compilers will evolve to the point that they’ll be hidden from view and will more closely resemble Excel where anyone can get set up immediately and with little training, only their existing knowledge of the business domain.</p><p>Assuming this future is inevitably approaching where does that leave technical people? As the basic repetitive actions are performed increasingly by AI, the work that will be left will be highly technical and environment specific. These problems will be beyond the reach of a pattern-based AI due to the unique nature of that work requiring greater reliance on logical and rational thinking.</p><p>As an example, the majority of work I do in my day job isn’t boilerplate work. It’s understanding the business needs and then figuring out a way to use multiple, unrelated systems to achieve that business requirement in a reliable and scalable manner.</p><p>My daily work requires me to have an understanding of networking, distributed systems, and data layer technologies and leverage these technologies to build solutions for business problems that I often have to research and define myself. I rarely use Google or StackOverflow for my daily work because most of the problems I deal with are too bespoke to have a ready-made solution.</p><h3>How to Future Proof Yourself?</h3><p>The most interesting thing about this question is that I don’t think it changes much from what I would’ve told someone 10 years ago before things like Dall-E and ChatGPT were even on my radar. The crux of that advice would be two major principles -</p><ol><li>Take time to understand people and the problems they are facing.</li><li>If you’re going to spend time doing something technical, understand the fundamentals and understand them well.</li></ol><p>To that second principle, if you’re new to software development, make sure you understand your primary tool — computers. When I say “computers” I mean everything about them — mobile phones, PCs, servers, operating systems, browsers, networking, software architectures, distributed algorithms, etc. You don’t need an encyclopedic knowledge of everything, but you need to be at least aware of the different facets. A good test is this — if someone asked you the question “What happens when you type google.com into your browser?” is your answer more <a href="https://github.com/Tyrrrz/interview-questions/blob/master/!-%20General/What%20happens%20when%20you%20type%20google.com%20in%20your%20browser.md">like this</a> or is it more <a href="https://github.com/alex/what-happens-when">like this</a>?</p><p>If you’re the latter, you don’t have anything to be worried about. If your answer looks more like the former, work to get yourself to the latter because ChatGPT and its pals are coming for your job.</p><p>If you’ve been working for a few years in software development, the same applies to someone new. It’s never too late to learn more about the tools you work with day-to-day. But if this isn’t where your strength lies, if your strength lies in your communication skills and working with people, focus on this side of things.</p><p>There will always be a need for people at the intersection between where people problems meet technical problems. The explosive growth in the no-code community is proof of this. There will always be a need for great business analysts and great project managers. I highlight “great” because there’s already plenty of mediocre business analysts and mediocre project managers out there and they’ve been struggling for decades.</p><h3>The Future Is Bright For Those Who Adapt</h3><p>ChatGPT and whatever other tools may arrive in the near future are extremely positive for software development. There will be people at the lower end of the software development skill spectrum that will need to find a different role.</p><p>If those developers are willing to up-skill, there are going to be even greater and higher rewarding opportunities on the horizon. These technologies are going to be a massive productivity boost for knowledgeable software developers.</p><p>The AI software developers are coming and they will take your job, thankfully that job is pretty boring and you’ll find something more interesting.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/600/1*U0U5KUW453QR8jcqMSBRVw.jpeg" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2d7dfbc3c6c0" width="1" height="1" alt=""><hr><p><a href="https://medium.com/better-programming/chatgpt-will-change-software-engineering-but-not-in-the-way-youre-hoping-2d7dfbc3c6c0">ChatGPT Will Change Software Engineering — But Not in the Way You’re Hoping</a> was originally published in <a href="https://betterprogramming.pub/">Better Programming</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>