Studio711.com – Ben Martens

A couple weeks ago I wrote about how difficult it is to explain the changes that we’re seeing in the software development world. Any skeptics that remain, at least at major tech companies, are at risk of being a lost cause. In my own org, I’ve purposely adjusted my approach from “get everyone to try it once” to “how do we let great things bubble to the surface”. But outside the tech world, it’s nearly impossible to explain. There’s a great quote from a strong AI skeptic who recently “converted” (original language included):

The real annoying thing about Opus 4.6/Codex 5.3 is that it’s impossible to publicly say “Opus 4.5 (and the models that came after it) are an order of magnitude better than coding LLMs released just months before it” without sounding like an AI hype booster clickbaiting, but it’s the counterintuitive truth to my personal frustration. I have been trying to break this damn model by giving it complex tasks that would take me months to do by myself despite my coding pedigree but Opus and Codex keep doing them correctly. (source)

I finish work and immediately want to start working on whatever new idea I had for a project at home. I’ve literally had to forcibly put my computer to sleep so that I stop and go to bed. It’s an incredible amount of fun to be able to go from idea to working code in one evening. There is a whole world of ideas that were previously too expensive to try that are now easy experiments. As an example, in the last 10 days, here are apps that I have built:

• Copilot Chat Export – VSCode extension that renders copilot chats as HTML for easy sharing
• CommuteTracker – This app runs on my phone and automatically knows when I leave home and when I get to work or vice versa. It also automatically logs whether I took backroads or the interstate. Notably this was the first app I’ve ever developed in Kotlin.
• RouteWatcher – This desktop app uses Azure Maps to determine how long it will take to get to work (or home from work) and it does this every 15 minutes with the results getting logged to SQL.
• MlcSports – This phone app is basically a reworking of the MLC Athletics webpage, quickly showing me news for all the teams along with upcoming games.
• TraktLite – I didn’t fully start this one within the last 10 days but it’s by far the phone app I’ve spent the most time refining. This is an alternative to the official Trakt.tv application for knowing which shows and movies I have watched or want to watch. I started with a specific scenario in mind and I’ve slowly expanded it to add more features that are tailored specifically for me and I’ve spent time refactoring the code to keep it clean.
• WelsCallStats – Scans all of the call reports for WELS pastors, teachers, and staff ministers to generate statistics about the average call duration for each position, the percentage of calls that are accepted, the churches who have made the most calls, etc. (No I’m not publishing the stats. It’s for personal curiosity. Jon Hein and his team would publish them if they wanted to.)
• WelsFamilyDevotions – I use devotions from the WELS at night with Elijah, but I don’t 100% love the mobile browser experience and it’s sometimes hard to remember which ones we’ve done before. This app just shows me the family devotions in order and hides any that I have read before. It also has a very clean view of only the devotion text without anything else.
• Temperature Probe – Again, this wasn’t fully developed in the last 10 days but I spent a lot of time tweaking this embedded Python project that runs on a little QT Py ESP32-S2 board to record the temperature and humidity periodically.
• Teams X Expander – This was a side project at work where I wanted to have a flow that would watch all Teams messages in a particular chat and anytime someone posted a link to a post on x.com, it would get the content of the post and share it in the chat so we didn’t all have to click the link to read it. It sort of worked but ultimately it was too hard to make it work within the security limitations of work apps plus the x.com API access is very expensive.
• OneNote addon – This was another side project at work where I was trying to give GitHub Copilot access to search around all my OneNote notebooks. This has been a challenge to get working within our corporate environment and this one ended up failing too, but it was a fun experiment and I learned a lot about how OneNote add-ins are structured.
• Interview question generator – This was another project at work that came from thinking about how to conduct interviews in this new agentic engineering world. It’s a bit silly to give people coding questions to answer, but how do I evaluate them? I used GitHub Copilot to generate easy, medium, and hard questions in three popular languages that would test how well the candidate could review code and find bugs. I was thrilled with the way this came out and shared it broadly in the company. There is a lot of discussion about how to handle interviews and I think this is a strong step forward.
• I tried to make a tool that would convert drawings from the old “Microsoft Expression” software package into SVG. It churned on my request for a long time and eventually told me that the file format was completely proprietary, but it also discovered a way to install and old copy of it and export to something that could then be converted to SVG.

Ok, now look at that list and remind yourself that is 1.5 weeks of mostly spare time. It’s a couple hours each night. Now imagine how much I’m able to get done in my full work day on all the projects I actually get paid for! Now imagine this multiplied by 80,000 other devs (or whatever the acutal number is) at my company. And remember what I said before about new capabilities coming out almost daily that lets us run faster and do more things in parallel with less oversight.

I have always thought how cool it was to be around for the mainstream birth of the internet. I was the perfect age to start coding HTML in notepad. It was a whole new frontier and we were (in parallel with others) discovering amazing new techniques and ways to combine technologies to make cool experiences. This has been a similar feeling except now I’m getting paid to do it and the changes that took months before are happening daily now. It is awesome to get paid to learn this, make discoveries, and share them with others!

I’m writing this as I’m still processing this excellent article: Something Big Is Happening — matt shumer. It is long, but honestly I would rather have you read that than this post.

Even for people in the tech industry, it’s difficult to explain how fast AI is improving. At work, one of my main responsibilities is literally to figure out what new tools and capabilities we can apply to our team and then help the team grow. Even with 100% of my effort focused on this, I feel like I’m holding on by my fingernails. It’s not fear that robots are taking over but a realization that things are changing faster than any of us expected. We are watching chapters worth of history books fly by every day.

For example, last week I was out sick for a couple days. The morning when I finally felt well enough to check messages, a non‑technical friend asked me what it was like working with AI. I joked that I’d been gone for two days so I was probably already behind. Then I logged in and… sure enough, a brand‑new, ground‑breaking model (Claude Opus 4.6) had dropped, and my programming tool (VS Code) had released features that make it even easier to work with multiple coding agents at the same time. I spent the entire afternoon just absorbing what had changed.

The pace of change is difficult to describe. Last summer I was mildly interested but it was clearly just a toy and most of the demos were hype. In the September, Claude Sonnet 4.5 came out and I could see how it was on the verge of being legit. On Nov 24, Anthropic released Claude Opus 4.5 and it was the inflection point. It was clear to anyone using it that there was no turning back. Opus 4.6 came on Feb 5 and OpenAI’s Codex models are surging too. People ask me what this is going to look like in a year. Who knows? I can’t even tell you what next WEEK will look like.

So yes, if you’re in software engineering, this is life‑altering in a way we’ve never seen before. But the key point is that this will change your life too. Whatever your job is, AI is already working to make parts of it obsolete. It’s a general‑purpose skill amplifier. That means whatever you’re already good at, AI can make you dramatically better and faster at it. This rewriting of reality matters for everyone, not just for people in tech. Here’s how to position yourself:

• There will always be people around you who think this is all hype and the fad will pass. Do your best to bring them along, but the most important thing is to make sure your future isn’t tied to their denial. If it’s your management chain, find a new job. If it’s someone you’re thinking about hiring, keep looking. Denying AI’s usefulness today is like believing in a flat earth. It is provably better right now. This isn’t up for debate. Don’t waste energy arguing with people who refuse to see it. You gave them a chance to come along. If they resist, they’ll get left behind. Honestly, it might already be too late for them to catch up.
• You might not be able to predict exactly how this will change your job, but you can keep yourself relevant by leading the way. Be the person who keeps up, uses these tools to undeniable effect, and teaches others how to do the same.

Back in December, I would still try to soft pedal all of this when I was in a group that I knew was mixed on AI. I did not want to sound dramatic or turn them off even more with my enthusiasm. But week by week, that is getting harder to do. The gap between people who use these tools and people who do not is widening so fast that it feels strange to pretend nothing is happening. I would normally end this by saying the future is now, but honestly it feels more like the future was last week and we are all just trying to catch up to it.

Let’s take a quick minute to update where we are at with AI in my daily job. It’s changing so quickly that it’s hard to organize my thoughts around it.

Over the last year (and even the last 6 months), AI coding agents have gone from “overhyped party trick” to “fundamental necessity”. A lot of that comes from advancements in models (Claude Opus 4.5 is my current favorite) but it also comes from us learning better ways to interact with the models. There’s still too much hype, but my personal daily reality has been permanently altered by these new capabilities. At this point, coding without AI feels as unthinkable as coding without StackOverflow or a search engine did last year.

Stepping back for a moment, it’s good to note that Large Language Models (LLMs) are a great fit for software development. If you think about training a model on the English language, it’s a mess because there are so many different styles and rules and unwritten rules and colloquialisms, etc. But with code, there the languages are strictly defined and there are intense levels of documentation for every piece of every language. This gives the AI agent a very well-defined playground to do your bidding.

What really pushed this tech over the edge for me was learning to start by having it generate documentation about the work it was going to do. I’d explain the requirements, have it generate a plan, and then iterate on that plan document until I was happy with it. Then I’d have the agent tackle pieces of the plan, step by step, verifying its progress along the way.

I can’t tell you how much faster I’m able to get projects done! There have been two giant projects floating around in the back of my mind for years at work, but I could never justify the time to do them. Not only did I get them both done in just a few weeks, but I did it in my spare time at work without slowing down the rest of my job. My output is skyrocketing and it feels like I have a new superpower.

I have no idea where this is going or how it’s going to change our jobs in the future, but I don’t think it will be too long before I’ll move up the “AI management” ladder. Instead of directly commanding one agent at a time, I can imagine telling a “manager agent” what I want and then letting it create its own agents to do different parts of the job and make sure they’re all still heading toward the goal. That’s potentially another order of magnitude increase in my output. After nearly 20 years at the company, I have a very long backlog of ideas to try, but now I’m wondering if that list is long enough. I can churn through ideas so quickly now and see which ones pan out.

“Will AI take coding jobs” is a common question but my answer is that if your job is typing in code, then yes, you’re in trouble. But if your job is seeing things that can be improved and solving problems, then this is an incredible tool amplify your impact. In that case, AI won’t take your job, but someone who knows how to use AI more effectively than you might.