It's been almost 2 years since I last posted here. A lot has changed: technology, my growth and development, projects, etc.

The last time I posted something about AI was AI and Me: What A Time to Be Alive in 2023. Looking back, I'm so glad I documented my headspace back then.

Consider this a similar headspace dump, not extensively thought out or curated. I just woke up this morning and thought it was time for another AI and dev dump on my cute little internet blog. I do hope it makes the most sense to people who aren't on the bleeding edge of this technology, or who are overwhelmed by the pace and sheer volume of it all.

In no particular order:

PRE AI

Pre AI, I was making an honest attempt to become a better developer, especially with frontend stuff, which was mostly new to me post graduation. So I documented building this portfolio site from scratch and adding Contentful to it. There's also this cute little Java tutorial. Reading all of it back makes me feel good, and I'm almost tempted to keep developing this site manually. But I doubt that will happen. What I will commit to is using AI sparingly to make improvements here, as a reminder of the journey so far. Plus, I kinda like the minimalism of this website.

WTF

In the 2023 post, I described ChatGPT as a tool that "plays more of a teacher-assistant role at this stage in my life." Well, a couple of years later, that's still the best way I can describe its role in my life. But the degree to which it now plays that role is hard to fully articulate.

At that time, we were talking about GPT-4 and GPT-4 Turbo. As of September 23, 2026, OpenAI's latest family is GPT-6, with names like Astra, Sol and Luna. You will not find a technical breakdown of how much better these models are compared to the older ones on this blog, but WTF! They know stuff (they've been trained on a massive amount of information) and with the right software around them, they can now do stuff for you on your computer. This "do stuff for you" part is why you're hearing so much about agents. An agent combines a model with tools and instructions so it can work through the steps needed to complete a task, instead of stopping at telling you how to do it. Guess my main agent's name.

NERDS VS NORMIES

Just an FYI: the nerds on X and in the general AI discourse refer to people who aren't on the bleeding edge of AI as normies lol. Just search the term "normies" on X and do not take offence if you think the shoe fits.

TOOLS

I almost left this part out because it's probably the opposite of what I'm trying to achieve with this post, which is to minimize the technical jargon. But I think it's important to be able to talk about where AI is at in 2026 as simply as possible, in any room.

As you can imagine, there are millions of pieces of software out there that can be called AI tools, and I obviously can't claim to know or have used them all. But the general picture looks something like this: there are the big AI labs, incumbent tech companies with their own AI plays, smaller labs and labs outside the US. Then there's everyone else. That last group includes people building software around AI (harnesses), with AI (vibe coders) and for AI (MCPs and other connectors).

Big AI: OpenAI and Anthropic are the two I spend the most time with. OpenAI makes ChatGPT, its general purpose AI chat app, and the GPT family of models that powers it. You'll hear the most advanced models called frontier models, which basically means they're at the leading edge of what AI can currently do. They also make models for generating images and working with speech.

Then there's Codex, its coding agent, and ChatGPT Work, its more general-purpose do stuff experience. I'll explain the word harness in a second because it comes up a lot when people talk about these tools.

Anthropic similarly makes Claude, its general use AI chat app, which has now absorbed its general purpose "do stuff" agent (formerly called Cowork). Its models are Fable, Opus and Sonnet, and Claude Code is its coding harness.

Across the industry, there are also increasingly specialized AI tools for fields like law and finance.

Incumbents with their own AI plays. Meta, Google, Amazon, Microsoft and xAI all have their own AI plays.

There are also smaller or boutique AI labs, like TypeSafe AI, working on their own approaches.

Foreign AI labs. A lot of China lol, although obviously not just China. Alibaba (China), who make Qwen, a model gaining serious popularity, DeepSeek (China) and Cohere (Canada), all with their own AI plays. A lot of these are open models. Open source (or "open weight") basically means the lab publishes the model itself so anyone can download it, run it on their own machines and tweak it, instead of only being able to use it through the company's app.

People building software around AI: This can mean building harnesses. If you think of the model as the brain, a harness gives it a body: eyes to see your screen and files, and hands to type, click and run things. Products like Codex and Claude Code include a harness. The agent is the working setup using that model and those tools to get something done.

It can also mean embedding AI into existing systems, like whatever is probably happening at your 9-5 right now. Or building an AI wrapper: an app that packages someone else's AI model into something people can use for a particular purpose. Think of everyone promoting an AI travel planner or an AI therapist. AI something.

There are also interesting photo and video tools like Runway and Higgsfield that help people create realistic and complex media projects, used alongside the frontier models obviously.

People building software with AI: I'm sure you've heard the term vibe coding, which caught on to describe the way AI changed coding. Until now, coding was a meticulous act of typing out syntax character by character and iterating until it worked. It was hectic, time consuming and fun, in the way I imagine pain as pleasure feels. With AI, we can now give instructions in plain English and have the model, harness or agent (depending on your workflow) write the code. Because natural language is loose and often subjective, "vibes" was elected as the official term for this loss of meticulousness in coding. Hence VIBE CODING.

People building software for AI: Because AI can now do stuff, not just tell us stuff, there was a need for software that makes doing stuff easier for AI. This is where MCPs come in. Connectors have always been a thing in tech: developers have always used APIs (Application Programming Interfaces) to plug different software into each other. MCP (Model Context Protocol) is a newer standard, that gives AI models one common way to plug into external tools and apps, like your calendar, your email or Slack.

Okay, enough technical jargon.

MY AI

Which tools do I use, and how?

I pay for ChatGPT Plus, the $20 USD a month plan, so I can access the better models and use Codex at higher limits. This is my preferred tool for everyday chat.

I also pay for Claude Pro at $20 USD a month for similar reasons, even though I use it less because it tends to run out faster than ChatGPT, and I don't see the need to upgrade to the $100 USD a month plan yet. I keep it for Claude's design feature when I need a quick mockup of anything, and for Claude Code when I want an alternate opinion from the GPT models. I find Claude is great for more sensitive things like financial analysis, and I don't mind it for coding either.

I also use a harness called Hermes Agent, which is comparable to Codex and Claude Code. I use it because it lets me switch out models. Anthropic won't let you use your Claude subscription in 3rd party harnesses, but that's fine, I just switch to Claude Code if I need to. With Hermes, I can quickly swap OpenAI models for the open source Chinese models if I run out of tokens. You'll hear people talk about running out of tokens. For text, tokens are the small chunks a model reads and generates. A token might be a whole word, part of a word or a punctuation mark. Companies use tokens to measure usage and calculate charges, although a subscription's limits aren't always shown as a simple token balance.

I also use Hermes because it lets me set it up in Slack. Hi, Pierre. So I pretty much vibe code from Slack. It does everything it needs to do on my computer while I'm doing the dishes or at the park with Feranmi. It opens pull requests and has computer use too. Honestly, there isn't much Hermes does that Claude Code or Codex can't do these days, but I have very few reasons to complain about Hermes, so I still use it. These things are starting to boil down to preference, if I'm being honest. I built app.vibewise.ca a group travel planning app all from Slack.

In terms of media, I'm still not fully sold. I don't use AI for APT 300 (my DJ collective with my brothers) event posters, because honestly, they suck. But I've been experimenting with getting Hermes, running an advanced model like GPT-6 Astra, to use my computer to build graphics in Affinity. All from Slack on my phone lol. Affinity is Canva's free design software, with tools comparable to Adobe Illustrator and Photoshop. Honestly the results are encouraging.

The next project on my to-do list is getting it to make me a movie in Premiere Pro. Will keep you posted on how that goes.

I recently started exploring Muse which is supposed to be Meta's take on a personal AI agent. So far so good. Will report more next time

That's it for now. So I'm not as AI pilled as I sound, but I'm trying to keep up and find more reasons to use it to break down any barriers to productivity and, more importantly, to financial freedom and prosperity.

FEARS

Safety has been a big discussion in AI. AI ending all of humanity used to seem like fiction or folklore. But when you think about it, you can kind of extrapolate the likelihood. There were recent reports of models breaking out of their testing sandboxes and compromising outside systems. A sandbox is a restricted environment meant to limit what software can access or affect. There have also been reports of rogue agents ignoring instructions or doing things they weren't told to do.

And in Anthropic's recent release post for Opus 5.5, they said, verbatim:

"We see signs that Opus 5.5 often suspects it is being evaluated, which challenges our ability to assess how it will act in the vast variety of real-world settings it is deployed in."

That scares the shit out of me, and it's evidence that the technology has indeed gotten slightly away from the developers.

Another thing that scares the shit out of me is misuse. Anthropic has been raising flags about biological risks. God forbid another biological disaster like Covid. Then there's AI in warfare, and scammers using deepfakes, which are convincing fake images, videos or audio of people. The whole thing is giving Frankenstein.

I've also said in the past that tech as a whole is giving "Tower of Babel," and some know how that ended. I call these my macro fears: fears about my big picture perception of the world and human consciousness.

I also have micro fears. The main one comes from seeing the number of AI induced layoffs around me, and being fully aware that it can happen to me and anyone. I fear letting go of the drudgery needed to become good at knowledge work without the help of AI.

If knowledge is now commoditized, does that mean humans don't have to get good at knowledge stuff? And what happens if I can't afford access to big knowledge? We definitely see it with Big Pharma: drugs people need to survive are sometimes out of reach because of cost.

What happens when knowledge becomes an inaccessible commodity? Higher education already made knowledge expensive, and now technology wants to make it widely available, as long as you can afford tokens. Should I just keep getting better at manual coding as a backup plan?

Then there's Feranmi and the generation born into this world. What does a career look like for them? What fields will be left? Is making money from labour even going to be a thing in 10 years?

POLITICS

I don't think anyone should shy away from the politics of anything. As with everything in these end times, the politics of AI has heated all the way up. From people raising concerns about AI's impact on the environment, to people calling on government to do more to regulate AI and protect jobs. My political views will hopefully be their own blog post someday, because I want to put in more research and come up with more concise statements and views. But for now, I strongly believe everyone in government needs to be as educated and up to speed as the people on the bleeding edge.

This has gone longer than I thought, but I feel much lighter after this dump.

Till next time.