Recently, I had a conversation with someone in which we challenged each other on some AI points but also found many areas we agreed on.

While I myself am neither a raging supporter of “everything AI”, I am also not against it. But in general, it frustrates me.

We spoke about an example of Spotify which allows AI artists and even gives them a blue checkmark verified artist.
One view on that is to treat AI music generation as an instrument. Someone learned a guitar and made a track, someone learned Ableton and made a track and someone used some prompts and made a track. Theoretically, all of these are just instruments producing “same” result. Practically - very different paths to this “same” result.

Same with other media - a photographer who got a camera, figured out the settings, found a good spot, interesting perspective, was there at the right time and produced a shot, versus creative prompting that would produce, possibly very similar shot.

So we started debating about “who is the true artist”, who is the true creative. The one who puts in time and effort into a more complicated process (classic intstruments/tools) or the one who solely uses AI, or even just digital (music DAW, Photoshop, 3D, etc.).

And then it sort of dawned on me that I have no issue with, for example, musicians who either learned and played a real piano their whole life, versus someone using Logic/Ableton/etc. with even sample packs, loops, etc. I see them both as musicians. Though “AI musician” gives me “emotional rash”. Why?

Pretence! That’s why.

AI tools and their use have not yet fully settled into a normal slot. It is still “pretending” to be someone else. It still “hides” its identity and “pretends” that the results were created by the “old process”, the human process.

And I realized that, for me, if I am browsing the offerings (music, images, videos, etc.) with a clear differentiator of what’s the classical guitar, what’s the electro mix, and what is AI gen, I actually wouldn’t mind that. I would see this indeed as a new form of creative tooling that helps people create more. And if I want to listen to some classical music because I want to hear how, back in time, composers created their masterpieces - I would filter for them, and when I want to consume volumes of AI-generated content and explore that - I would also filter for that.

But in today’s state, we have not yet fully designated a spot for AI tools. They are there, but everyone uses them to basically “pretend” to still be of human, manual origin.

I can imagine AI music festivals, AI photo exhibitions, AI books and poems, AI cooking shows, and so on. I just want to be able to filter it out or in, depending on what I need at this very moment.

This line of thinking started to excite me; it relieved my frustration, and I thought that, yes, I could even watch an AI movie if one came, if that’s what I wanted to watch at that moment. If I have the choice of choosing which content I consume, instead of being force-fed something pretending to be something else.

Not so long ago Bungie released Marathon. The extraction shooter.

Marathon had a few setbacks on its way to release. One notable was about the stolen art. Glad this was solved.

Because the art direction of this game took on a big leap and a serious challenge. Instead of going into the realism with its textures and models, they introduced a very distinct graphic style. Not unique, as you could draw a lot of parallels to many anime and alike. But still brave and colorful.

I was (maybe still am?) in love with the trailer:

Trailer was directed by Alberto Mielgo.

And if you watched Jibaro in “Love, Death and Robots” on Netflix, you’ll immediately recognize his style. Jibaro is his, too.

Jibaro short preview:

But back to Marathon.

This post has no intention of reviewing or discussing the game, as if people still need to know anything about it. No. I purely wanted to put down in writing my affection for it. At level 50, maining the “medic” and logging 102 hours with the game, I still enjoy its style, its soundtrack, and its game loop. I sort of ran out of things to do except PVP and trying to get through the Cryo Archive, but still, I have to say this game helped me heal from Tarkov (1500+ hours logged). Probably I’ll be back on the “heavy drugs” (Tarkov) soon, but for now, it’s a nice stroll in the park (Marathon).

And really - visuals are a pure art:

Intro for ROOK:

But then there is also a soundtrack, which also tends to trigger some meditative states:

That’s it. No big conclusions, no big statements, no revelations. The game is out; it had ups and downs, and it’s still developing its audience. But I hope it continues its growth as I love to keep enjoying its style.

I use CLAUDE (sounds like a confession…). Tons of private projects (which will inevitably die due to missing any valuable use case) and professionally.

I tried setting up multiple agents, created my own MCP server(s) for different apps, and delved into skills and plugins.

But in the end, I found all of it to be an overkill for the needs I have. Even when my project is rather complex, I really prefer to work WITH CLAUDE, not let CLAUDE work for me. I do trust that it’ll do a decent job, actually, if I just let it loose, but at the same time, I want to know what’s going on. I want to follow in CLAUDE’s footsteps.

The most useful for “extras” for my CLAUDE envrionment are the MS Azure plugins:

Just so that when I need to deal with tons of AZURE APIs, CLAUDE helps me to nail down the right ones.

As for the rest, I stick to a live CLAUDE session within the repo I’m working on, and I mix between manual edits (10-20%) and CLAUDE edits (80-90%).

I love the /plan mode. I mostly work with it. But honestly, I rarely read the plan fully. I skim through it, if it makes sense to me, general logic and approach, I don’t care about exact details. At least I know which decisions CLAUDE took to realize a certain logic.

I prompt - CLAUDE edits - I edit - we progress.

Do I need my projects to go much faster?

Well, they are going much faster already, and I still feel like I maintain a level of control.

Even faster? Leave it all to CLAUDE?

I have an existential dread that a client or a colleague asks me about a piece of the product, and its logic, and I have zero knowledge of how and why, even though I “shipped” it. If CLAUDE vibes coded it in full, I will then have to VIBE TALK myself out of those questions. I don’t want that.

I still feel solely responsible for the solution. Hence, I need to know what, how, and why.

I’m yet to find a project where I would actually consider deploying multiple agents to multitask at the level of a full dev team. Yes, maybe the scale of my operation is small. But at the same time, back to the existential dread, I want to be in control. If not by decision-making, then at least by the knowledge of the whole codebase.

Maybe I’m clinging to the old world a bit too much. Maybe I’ll let it go.
But for now, co-working with CLAUDE is my favorite thing to do. I want to be in the loop.

Subscribe now

Year 2005+

The Internet is “free” (compared to the sad state of things we have today…). You browse it, you discover something new. Something unique. Something interesting. Information feels more like a nicely cooked meal, rather than a fast food that was force-fed to you on the go (what we have today).

And suddenly you stumble upon this video, the new ad from SONY.

“WHAT?”

First, it hits you with a wonderful rendition of The Knife - Heartbeats by Jose Gonzales. If you survive that joy, you are finished by an unspeakable show of countless bouncy balls pouring down the hills of San Francisco. Something that many would have thought of, but never imagined could be scaled up to such an extent!

You watch once. You watch twice. You watch until the time tells you that tomorrow’s early morning wake-up just got harder.

You watch it tomorrow. You share it with friends. You express how amazed you are and how cool and simple the idea is, and how brave and creative the execution is.

Later, you find the behind-the-scenes:

You watch, you enjoy, you share, you talk about it.

You dream of changing jobs and working for marketing agencies, movie studios, creative industry. You want to execute such ideas!

This video, this idea, and its execution now occupy a small corner of your mind, and it never leaves.

Through the years, you stumble upon many commercials, and from time to time, the creativity of them reminds you of that SONY ad.

BURBERRY (behind the scenes)

GUINNESS (behind the scenes)

Old Spice (behind the scenes)

There is more…

And every time, each of them shares the creative genius of production planning, directing, acting, editing, etc.

All of them rely on HUMAN CREATIVITY. Many people work together to realize something out of the ordinary, something out of place, something dreamy or brave, or risky. The use of real tools, real world, with its opportunities and limitations.

And it’s a pleasure to watch. It inspires you.

Year 2026

AI-generated commercials. They have not yet taken off at full scale. But brands and companies try them. People also explore - how would it look?

Here is one:

Well.
It tries to be funny. It has all the ingredients of “how the hell did they film that?” (whale?:). You watch it. But once is enough. And you might even just skip it. You know that no one did an insane plan of somehow getting the whale onto the highway. Maybe a dead one, maybe would then use some cranes with straps to “animate” it. No one dancing around the fire while being ON fire. No one had to flip a truck and set the scene as if it were a crash.

You know that HUMAN CREATIVITY here was expressed only via prompting. Only “wishing” to see something, but not actually planning and doing that something.

You also know that you can create the same type of video if you just spend a few hours with the AI gen tools.

This won’t be shared as much and won’t be talked about.

Are we killing our own creativity by making things easier where they should stay difficult? Maybe we need obstacles to come up with a cool way to overcome them. Maybe we need limitations that require time, require taking risks, require trying and failing in order to get to something great?

Can a prompt be as admired and spoken of as SONY’s filming day at SF, where gazillions of bouncing balls stormed the streets?

Personally, I am not impressed by prompting.

Every prompt that generates a video kills a creative production that could have involved many and inspired millions.

I’m not impressed. Sadly.

Lately, I was watching/re-watching The Office (US). Yes, it’s a comedy show, but it’s still set in a sort of real world with real issues, problems, and ideas. And in the show, when the salespeople of Dunder Mifflin travel to the potential client (or existing) and want to close a deal, they quite often use a phrase:

With us you’ll get our full attention and exceptional customer support!

Ok, probably most salespeople say that. But it’s a good sales point – next to the main product you sell, offer the customer service that’s better than anyone else can offer.

I think I know what they mean. I used to be a “user” of such customer support services. We used to have more of them when support was more personal and more accessible. Maybe just because we had to call and have a human conversation with the representative. Of course, we had negative experiences, but the mindset of customer support used to be different – it used to be an extension of the main product. It used to target potential gaps and pitfalls in the core product, solve edge cases, and take care of the customer. It used to be an extra “human” interface to the core product.

Then we started saving money. E-mails, tickets, extensive FAQs, “if – then” chatbots, and now AI.

These days, getting to talk to a human feels refreshing. Once you have a question, you first need to do a series of quests to be granted access to a human being. Send this here, show that there, chat about something else here: “Did this solve your problem or do you need further assistance?”…. Yes, I do need further assistance!

But the mindset of human help and support has also changed. People on the other end of the line these days are siloed, protected, and hidden. If you get to them, they know that “you are a problem” as nothing before (docs, FAQs, AI bot) could help you, so now you are going to “ruin their day”.

Lately, due to personal circumstances, I had to interact with many new services and had multiple issues, questions, etc. – I had to interact with customer support. And the predominant mindset I felt was that I always needed to “excuse” myself, defend my problem, plead, and really try my best to “fit” into the customer support employee expectations. Rarely (almost never) did I have a person who would CARE. Who would genuinely try to see this from my perspective, start thinking along with me, and work with me on the case? Who would actually try to make my day a bit better, given that the shortcomings of the core product made it worse.

IWG – Yes, I’m calling them out directly because I’m stunned by the level of service this company provides. I experience rudeness, cluelessness, and absolute disregard for customer needs, while still expecting the customer to bend in any direction and adhere to every line of the signed contracts, regardless of the value the customer experiences. I reserve the possibility that others might have had better luck than I did.

This wave of customer support experience kept bringing me back to the Dunder Mifflin sales pitch…

…you’ll get our full attention and exceptional customer support…

This made me think that, actually, today you can probably win growth by just improving customer support. You can offer the same core product like many other companies do, but bring back humans who care. Create a better culture and mindset. Make sure your customer support is not a moat, but a bridge that keeps delivering positive value to your customers. Don’t hide behind an AI assistant, don’t hide behind walls of FAQs – give me a person who would go down into the “trenches” with me and solve that “incorrect account number” while being positive, informative, supportive.

I would switch multiple services if I knew that I had a “partner” when I need help.

July 2022 – that’s when, somewhere on Reddit, I stumbled upon AI art generated with Disco Diffusion.

That was sort of a “splash screen” / poster on the page of Disco Diffusion. I saw all the art generated by other people, and I got very excited about it and what it does.

I didn’t know much about the detailed technicalities of AI image generation. I knew the general principles of diffusion generation. Read some articles here and there. This, for example, gave me some idea (the image below is borrowed from the article).

I followed some guides and set up on my PC the Disco Diffusion. But after doing the generations manually, one by one, I thought I wanted to scale it up a little bit. Or at least automate to see more results and make them wildly different. I was also tired of coming up with prompts:)

I’ve set up a few things:

• I got the text of Snow Crash by Neal Stephenson (I own a copy of the physical book…)

• Ran some logic through the whole text to randomly split it into chunks of text. Approximately 200 +/- chunks in total.

• Each chunk I ran through the regular expression to extract long enough, random sentences and collect those sentences into a list (array).

• Fed each sentence as a prompt to the Disco Diffusion with settings set for a certain number of steps and also safe .gif of the generation progress

• Created a Telegram bot that would send each generated GIF to me.

• Ran that through the night.

In the morning, I got ± 200 gifs. From awesome to weird. And each of them showed how the diffusion was generating those images. I just sat for hours watching all of them.

Recently, 4 years later, I stumbled upon this Telegram bot, remembered the good old times and thought I should document this somewhere. Hence, below is a subset of these GIFs with the relevant prompts (sentences). First a prompt, then the result. Just to be clear:)

And there are more… ±200:) They are all mostly strange. I used some random model/dataset that was being discussed back then online. Downloaded god knows from where. It surely had a preference for architecture, but there was zero information about what it was trained on. I would suspect it was not respecting any rights of any sources… It was a bit like the Wild West back then.

But I liked that experimentation. Taught me a lot of things. Demystified a lot of AI magic back then, and I’ve stayed skeptical ever since!:)

I will download the whole archive and make another post with a proper gallery. Maybe I will also compress the GIFs a bit or convert them to a smaller format.

Subscribe now

So many of the same, but different colours… Colour tones.

I’m not a painter or any type of artist or graphic professional who pays attention to the colour details or knows colour theory. I know some basic stuff for day-to-day life; white pants are a challenging choice – brave style, hell, to keep clean, dressing all in black is very not cool to some and very stylish to others (funerals excluded), guys don’t buy pink cars, if someone’s face turns blue – better get a doctor on the case asap.

And living my whole life in different big cities – I also absorbed the “city pallets”. From century-old red bricks, to boring grey concrete, to strange soft tones, to bold attempts to stand out.

And when you live in the cities for so long, you also notice that keeping up the colour of any structure seems to be a tough task for the city services.

Well, I get it. Even without any deeper research, the reasons can be relatively obvious.

• Lost original paint (even its formula or ID) and grabbed something “similar.”

• Weather elements make the original colour fade/change, hence even the same paint looks different.

• Dirt on the original layer (not everyone washes the whole wall before painting it over…)

• Can’t be bothered…

And doing a bit of research, the list continues:

• The way you apply the paint can influence the final colour

• Regulatory changes might start dictating what is light-blue in 2010 vs light-blue in 2020 (never trust your eyes, trust the stamped and signed paper!)

• Metamerism

• Intentional temporary fix due to the plans for a full repaint in the near future

Well, the temporary fix falls under one of my most favourite sayings:

There is nothing more permanent than temporary.

But if you walk the streets of any city and pay attention to this, you’ll notice this case a lot.

I kept thinking about how to avoid it, how to fix it.

And some fixes are obvious (not all are simple, though):

• Colour-matching

  • On site, in a different light, a couple of measurements, a couple of tests – can get you the right formula.

• Gradient edges / blend it in

  • Stop with hard edges. Just blend it in. However, there are such massive colour mismatches that it will just look like a massive stain. But in cases with “we are pretty close,” – might work out well.

• Reduce the pallet, keep stock, keep a track, and a knowledge base on how paint fades and which tone to use after 1 year, 5, 10 …

I’m sure the list can go on with much smarter and better solutions. But, there is one solution that rules them all:

Just don’t care!

Cities are… busy. Vibrant. Moving. Changing. Things get built, things break. A patchwork of paint is probably not as critical an issue. Some cities still try, some have given up. Doing all those solutions and more equals time, money, patience, and commitment.

And after all, I was not going to try and solve it here. I was only going to share a small set of photos I took from 100s of patches that I’ve noticed through the years. These are just fun artifacts we get to live among.

• Happy

• Curious

• Worried

First, let’s cover Happy and Worried:)

AI creations are getting marked! You can start noticing this on LinkedIn as they introduced Content Credentials.

It looks something like this:

Then, relatively recently, there was a proposal to add HTTP headers clarifying whether the served content/page contains AI-generated content.

It’s still a proposal, and I’m not sure it’s moving forward much.

We have also lived through a short era of text checkers for AI generation. It was fun, until those checkers confirmed that the Constitution of the USA was written by AI.

We have trained our eyes at this point to spot the AI-generated imagery most of the time, and we sort of can get a sense of whether a chunk of text from someone was generated by AI or written genuinely by them. Our accuracy is not 100% and maybe even not 80%. But we got good at it.

As we try to identify what is genuine human creation and what is AI, we seem to still be confused about what to do in the cases of AI generation.

On one hand, we sell multiple solutions that are based on AI generation. All shapes and forms. Mixed media. Endless use cases.

On the other hand, we also hunt down with pitchforks those who use AI-generated content in some industries. I.e., Clair Obscur case.

On Hacker News, there is a comment from the user nrmitchi with a good point. Here is a piece of the comment and a link to the full one:

Yes, there is a huge problem with AI content flooding the field, and being able to identify/exclude it would be nice (for a variety of purposes)

However, the issue isn’t that content was “AI generated”; as long as the content is correct, and is what the user was looking for, they don’t really care.

This is a cornerstone of the confusion, as it seems to me. AI is still a probabilistic technology with its precision getting better and better; however, knowing that it’s not deterministic, we tend not to give it our full trust.

All above explains my happiness (being able to have a regulatory way of marking human vs. AI) and worry (aren’t we over-regulating? If it’s valuable, then who cares how it was created?).

And now to the case that I’m curious about. It’s the case that’s not yet TRUE. It’s not yet implemented or enforced. But as I observe the developments, emotions, EU’s love for regulations (moats…) – I can see the scenario where marking the AI-generated content becomes a regulation/law.

And in that scenario, I’m wondering what it would do to the fashion industry, specifically the e-com part. How would that impact the decision-making of a common consumer?

If today we shop like this:

Review items like this:

But then, with the marking in place, if we get to shop with something like this (AI mark mock-up):

And like this:

What would this do to the consumer choices, trust, and preference? I’m very curious to see the A/B test on items like this. As nrmitchi said above – if the result is correct, no one would care if it’s AI or not. But, how would our consumer psychology work if we are looking at something that maybe is correct, however, it’s still MARKED as AI. Wouldn’t we think twice? Thinking “Hmm..they mark it as AI because it might deviate from how it actually is?”

It could also take one friend or friend of a friend who “bought an AI-marked item, and it was really off – I don’t trust these images anymore” – however, in reality, it could still be the usual e-com consumer experience, as even photos of the real items don’t always convey how the items will really look ON YOU.

If I were to make predictions and place bets, I think that if we get to the state where marking AI-generated content becomes a must, fashion would need to rethink its usage in direct-to-consumer content. I would imagine that an online shop mixed with real shots and AI shots wouldn’t be the best of ideas, and brands would have to either go all AI and build the trust with the consumer that their AI shots are CORRECT, or keep it “old skool” and keep the craftsmanship of the photoshoot alive.

I do want to witness that “conundrum” even though it might cause some negative impact to my professional plans… But I think it’s an interesting experience that we will all learn from. So… Bring on the markings!!!