My Personal Experience using GenAI

10

October

2025

No ratings yet.

As a Master’s student who also works part-time as a software engineer, I’ve been relying more and more on generative AI tools in my daily life. At first I treated them as a novelty, but over time they’ve become a multipurpose tool for both my academic and professional life. Still, GenAI is not a magical tool that solves everything, its limitations have become increasingly clear to me over time.

One of the earliest ways I used GenAI was for search. Instead of digging through ten different Google results, I could ask a direct question and get a straight answer. This saved time when I was researching for papers, looking up a software library, or just wanting an answer to random questions I had (as one does). However, I was cautioned against trusting the answers blindly. I quickly experienced first hand that sometimes the AI gives outdated information or confidently states something incorrect. So I do additional research, depending on how high the stakes are.

Another major use case for me is summarizing. During my studies I often have to digest long articles, papers, or lecture notes. Letting the AI condense 20 pages into a quick summary was a game changer. Of course I still have to do more in-depth reading when I need to fully understand an argument, but it gives me a head start and helps me prioritize what to focus on.

GenAI is quite good at brainstorming and drafting too. Whether for a group project in class or when sketching ideas for a feature at work, it provided prompts and perspectives I wouldn’t have come up with myself. The downside is that its creativity can be surface-level, models often just regurgitate variations of ideas that were in their training data. So if I want to come up with something truly novel, I try to think of it myself and then use GenAI for “validation”.

In terms of drafting, I’ve used it to outline essays, emails, and even software documentation for my work. It’s great for overcoming writer’s block and speeding up the initial phase. Still, if I don’t rewrite and refine the draft myself, it’s easy to see that it was generated by AI because it sounds generic.

In my work, I use GenAI mostly for boilerplate code and bug explanations. It saves me time on repetitive tasks. But in complex systems, its contextual capabilities fall short. I’ve had it produce code that looked correct but had subtle flaws, not in syntax but how it used other functions in the codebase. Also, at times it has difficulty adhering to the design philosophy and stylistic choices of larger projects.

Finally, I’ve even used GenAI for language learning (currently Dutch). It’s particularly good for practicing small conversations and checking grammar. That said, I heard from a couple Dutch friends that it sometimes uses phrases that feel unnatural to native speakers. However, for now its Dutch is definitely better than mine, so I will continue using it for learning.

In sum, I already use GenAI for a variety of tasks, and I’m sure I will continue to discover new ways it can be useful. Have you tried any of the use cases I mentioned? What was your experience? I’m curious to hear.

Please rate this

Generative AI as a Co-pilot?

10

October

2025

No ratings yet.

My first real use of generative AI was to plan a trivia night for my friends. Not only did ChatGPT save me from hours of work, but it also gave me personalized questions based on the information I gave it. Furthermore, I used it to make a PowerPoint presentation in a jeopardy game show-like format with over 30 slides. It gave me questions with varying difficulty, themes, tones, and styles to keep the game night fun but not too easy. I have also used Dall-E to help a friend design marketing graphics for their bagel start-up. I am not a very artistic person, but this helped curate my ideas into figures. It only took some short prompts and a few seconds, and I was delivered multiple mockups. This made me realise that generative AI tools have helped me do more with less.

The most helpful generative AI tool for me has been NotebookLM. I attach my lecture notes, readings, and personal notes to convert them into podcasts. It is quick, and it helps me review concepts even while commuting. I believe that I learn better while listening than while reading. So, NotebookLM has definitely been my saving grace while studying for exams.

Yet, I wonder whether I am learning or just consuming what the AI tool decides is important. Across these experiences, I have felt both smarter and dependent. I believe that generative AI improves my ideas and productivity, but it also reduces the natural creativity aspect of being human. By this I mean experiences such as brainstorming, reflecting, discovering, writing, etc. I realize the solution is not rejecting AI completely, but rather using it consciously. Hence, I use generative AI tools as my co-pilot, who helps me navigate through the proccess more efficently rather than it being my captain.

Please rate this

Building Applications Has Become Easier Than Ever

10

October

2025

No ratings yet.

A couple of months ago, a client at my student consultancy job asked us to automate a document anonymization process at their real estate agency. Due to data protection requirements, the processing had to be done locally or within their Microsoft environment.

After some unfruitful experimenting with Power Automate, we decided to give building our own tool with Claude (an LLM by Anthropic) a try. With a bachelor’s degree in international business, I had no coding knowledge whatsoever. The results were amazing. Within a few hours, we had some basic capabilities established.

After a while, I wanted to make the process more efficient than having to copy paste Claude’s changes into my project files. A new version, named Claude Code, had been released. It enables the AI to work in your project files directly. I had to watch a few tutorials and do some error-fixing with ChatGPT to get it to work in container (a sealed-off environment on my laptop). After about two hours we were ready to go.

The result was a developer working at lightning speed. It could code, test, readjust and retest until it works, all in one go. You see it break down the task into sub-tasks and tackle them one-by-one. Alternatively, you can put it in plan-mode so it will brainstorm about what you want, come up with multiple alternatives with pros and cons and execute one when you give the word. While it is executing that piece, you can open a second, third or fourth window to work on a different issue. You can quite literally run an entire team of coders at the same time, while you only manage them.

However, it’s not perfect. Especially fixing more complex bugs can be an issue. Sometimes, after showing the problem and asking for a solution several times, it won’t be able to fix it. Since I do not know anything about code myself, I had to be creative.

Firstly, working modularly helps you pinpoint the issue to a specific module. You can then ask Claude to zoom in on that module and come up with possible causes. With just logic you can often judge its suggestions. That way you can help Claude get closer to fixing the issue.

Sometimes, it gets stuck in a certain thinking path it has gone down. In that case, it can help to get a second opinion. You open a second window or ask a different LLM (e.g. ChatGPT) to look at the issue. This way it is not biased by the context in your current conversation or its LLM specific knowledge. This has more than once resulted in it immediately recognizing the real issue, and me being frustrated with the fact that I spent half an hour trying to fix it in the initial chat.

All in all, I was really amazed with the possibilities. Getting it all set up was a bit of trial and error, and it takes quite some time to brainstorm about the implications of architectural choices. But once you have done that, it builds full-fledged applications in minutes.

New AI tools are being released quicker than we can learn to use them, so adaptability seems more important than ever. Just being able to build applications is not enough either. Just like before coding became so much easier, you need a business case for the application too. All in all, I think it’s a great time to be a business student with an interest in technology.

To anyone else who has been experimenting with AI tools for coding: what tools do you use and what best practices have you discovered?

Please rate this

AI can turn floor plans into 3D housing models. What does it take?

10

October

2025

5/5 (1)

I read “Automatic Reconstruction of Semantic 3D Models from 2D Floor Plans” by Cambeiro Barreiro et al. (2023). The goal they had was to turn scanned plans into vector 3D models that you can export as IFC, a building data format. It lets different BIM (Building Information Modeling) tools share the same data, like walls, rooms, and doors, etc. This matters especially for big real estate portfolio holders, like housing associations. These parties would be able to gather insights from the vast amounts of data, which are normally only available on paper.

The core problem

Symbols and techniques differ, and old drawings are clumsy. Windows and doors break through walls. The scans are biased. Redrawing by hand is labour-intensive, expensive, and error-prone. The stakes are higher for large portfolio holders, such as housing associations. They oversee thousands of residences. They are unable to compare layouts, energy features, or risks across the stock in the absence of structured data.

How it actually works

You begin with floor plan scans or PDFs. Like a human, the system reads the plan. It recognises windows, doors, and walls. It corrects gaps and crooked lines. After that, it creates clean vectors from these shapes. Lastly, it exports IFC and creates a basic 3D model. Common BIM tools can use that file. You can then measure areas, tag units, and count rooms. Additionally, you can link maintenance notes or energy data. Structure is crucial. You can perform the same checks on buildings in the portfolio at large scale once the data is in a standardised format.

source: Cambeiro Barreiro et al., 2023

Why portfolio owners should care

Data is locked in drawers by paper plans. They become structured BIM thanks to this pipeline. This opens up the portfolio’s unit mix, wall types, window counts, and layout features. It facilitates large-scale renovation planning, accessibility assessments, and energy audits. You receive a cleaner handoff to designers and contractors and expedited due diligence.

Takeaway

Digitising plans is the first step if you own a sizeable housing portfolio and would like to digitize. Segment walls, identify symbols, and carefully translate everything using floor-plan AI. To prevent lock-in and maintain the data’s utility across tools, export to a common data type like IFC. The improvements are easy to understand in terms of speed, consistency, and insight at the portfolio level.

Reference: Cambeiro Barreiro, A., Trzeciakiewicz, M., Hilsmann, A., & Eisert, P. (2023). Automatic Reconstruction of Semantic 3D Models from 2D Floor Plans.

Please rate this

Film Photography and AI: My experience

10

October

2025

No ratings yet.

I started using ChatGPT for school, then realized it could help my photography too. With film, every click costs money and time. Having a quick second brain lowered the stress and helped me make better choices before I even loaded a roll.

Most of my shoots are low light or mixed neon. I ask for a quick plan: likely shutter speeds for Cinestill 800T or my MARIX 135 T800 and Amber T800, what EV to expect at blue hour, and how far I can push before motion blur ruins the look. It is not magic. It just gives me a sensible starting point so I do not waste half a roll testing the obvious.

I also use it for composition practice. I describe a scene from my contact sheet, like “subject under a shop sign, bright window behind, messy foreground.” It suggests two or three framings to try next time. Step left to kill a distraction. Drop the angle to separate the subject from the background. Add a leading line from the curb. Simple ideas, but it keeps me iterating. My contact sheets feel less random and more like a series with intent.

Metering and color are where it saves me the most. If I am debating 1 stop over for skin indoors, or how much to bias exposure for tungsten under mixed LEDs, I ask for trade-offs. It reminds me what will happen to highlights on 800T and what to expect from halation. When a scan comes back with a green cast, I run a quick checklist for likely causes and fixes. It is the same with push or pull. I still note my lab’s advice, but I go in with clearer expectations.

Trust grew with results. The more useful the output, the more I tried. I still keep guardrails. I verify technical claims, write shot lists, and never paste personal data. The goal is not to outsource taste. The goal is to give my taste more chances to show up.

If you shoot film, try this next roll. Write a one paragraph brief, ask for two lighting setups and a backup plan, and make a tiny shot list. Then compare that contact sheet to your usual one. Did you see more, or just shoot faster?

Please rate this

Agentic AI in Customer Relationship Management (CRM)

10

October

2025

5/5 (1)

One of the most transformative developments in enterprise technology today is the emergence of Agentic AI in the field of CRM. The way campaigns are being designed, executed and how data is being processed fundamentally shifts towards the integration of Agentic Systems. 

But how does that actually work in practice?

Unlike conversational AI tools that only assist users through predictive analytics or content generation, agentic AI actively makes decisions and executes tasks, without constant human intervention or the need for prompts. In todays practice, this means CRM systems are evolving from static databases into intelligent ecosystems where AI agents autonomously manage lead follow-ups, orchestrate personalized customer journeys and interestingly, also initiate retention campaigns when churn risk is detected. When companies decide to implement those agentic capabilities into their CRM, the implications for efficiency and scalability are profound. Those companies can engage customer continuously and react to behavioural changes in real time. The most important aspect is that a level of personalization for the individual consumer can be achieved, which was previously impossible at scale. 

How to implement that into existing workflows?

Many firms overestimate their technology readiness, meaning that the often launch isolated pilots, rather than focussing on clean data, orchestration frameworks, or proper human oversight. To be able to implement this technology successfully companies need to follow a balanced approach between bottom-up and top-down. Only when the employees are being enabled and empowered to identify areas where the agentic AI can help, the implementation will work out. Especially in CRM it is of high importance, that the system development begins with clear process mapping, well-defined guardrails, and incremental deployment. This way the firms can expand the given autonomy as trust in the system grows. If Agentic AI in CRM is implemented right the CRM moves from a reporting tool about campaign success, customer churn, or CLV into a living, learning collaborator that augments every stage of the customer lifecycle. 

How does the agentic workflow look like in practice?

First, the Agent sets a Budget-Goal for a campaign (Increase of Abonnement-Conversion by 15%). Then it accumulates data from CRM and other sources. Third, the agent analyses and prioritizes the given data (Decision who, when, on which channel and with which offer we can contact the client). The fourth step is about Asset generation such as creating personalized text or visuals. Here the highlight is, that the agentic AI personalizes every Client contact based on the accumulated data in the previous steps. The next step is about Optimizing the output and flushes the campaign to the client base. The next step is crucial for the agentic system since feedback loops, as well as deep learning capabilities come into play. Here the agent will focus on the interpretation of the performance of the previous campaigns and adjusts where it is necessary. 

Conclusion

Agentic AI in CRM is without any doubt the biggest transformation in today’s business-world. Companies are constantly searching for better ways to run client campaigns, to reduce churn and to increase interaction with clients to consequently generate more revenue. With the integration of an Agentic CRM system, topics like scalability and marginal costs are important. Companies need to focus on the implementation now, instead of falling behind competitors.

Please rate this

Trying on Clothes with Chat

10

October

2025

No ratings yet.

I feel like every day I’m introduced to a new model of AI, and I need to make a conscious effort to keep up with it, especially to get the most value out of my ChatGPT subscription. Before starting this course, I had never really thought about the differences between General AI, Generative AI, Deep Learning AI, and so on. However, I’ve become much more aware of the types of AI I use in my daily life and how I can apply them to everyday questions and tasks.

To get more hands-on and fully utilize ChatGPT’s generative capabilities, I decided to ask him to help choosing which clothing items I should purchase. For a task to be considered “generative,” the AI needs to create new content (e.g., images, music, code) rather than simply analyze existing information. So, I asked it to generate images of the outfits I had in mind so I could better visualize them.

I’ve been wanting to buy a new coat for a while, but I kept putting it off because it usually takes me a long time to decide on the fit and color I want, and then even longer to find a store that sells something similar. To make this process more efficient, I first shared a link to my Pinterest board so he could know my preferences. Then, I asked it to search for coats I might like, and it returned a list of potential options with images and store links. This part of the process was more of a general AI function, involving search and curation.

^^ One of the suggestions that he gave me

Based on the items I liked most, I then provided ChatGPT with a screenshot of a dress I already owned and asked it to generate images of full outfits that included different coats. Before generating, I also gave my height and clothing size so that the proportions would look more realistic. This allowed me to clearly imagine how each piece might look on me, without actually having to try them on.

^^ Output of the generated image.

I found this use of ChatGPT incredibly helpful because it saves me a lot of time. Normally, it would take me at least 18 minutes to narrow down my options to a few stores, and then even more time to visit the shops and try things on. With generative AI, I was able to visualize my outfit options quickly and efficiently.

One way I could improve this process even further would be to provide ChatGPT with a full-body photo of myself. That way, the generated outfits could be customized to my actual body shape and features. In the future, it would be great if more clothing websites started integrating GenAI tools that allow customers to virtually try on clothes. This could completely change the online shopping experience, making it faster, more personal, and much more convenient.

Please rate this

It Wasn’t Easier with AI – Just Finally Possible

9

October

2025

No ratings yet.

When I first started with Power BI, I spent days trying to make sense of DAX, the formula language used to calculate and analyze data inside Power BI. I even joined a full-day training once, only to leave more confused than before. Getting one formula to work felt like winning the lottery – rare and mostly luck. Most of the time, I’d ask someone from IT for help, wait a few days, get a fix, and then forget half of it by the time I tried again. It was slow, frustrating, and I always felt dependent on others to move forward.

That changed the moment I started using generative AI tools.

Suddenly, I could ask why something didn’t work, not just copy a fix that a more experienced colleague or someone from an online forum came up with. I could try different versions of the same formula, see how each behaved, and actually understand the logic behind it. Instead of watching people like Alberto Ferrari on YouTube and trying to force their examples onto my own messy dataset, I could finally learn by doing. In my data at my pace.

What surprised me most was how quickly small insights added up. A few lines of explanation from ChatGPT often cleared up what hours of tutorials couldn’t. Over time, I stopped treating DAX as a collection of tricks to memorize and started to see the structure behind it.

AI didn’t make things easy. It just made them possible. It gave me the sense that, if I stayed curious long enough, I could figure almost anything out.

When Research Says the Opposite

Funny enough, that’s not what the studies say.

An MIT Media Lab paper (Kosmyna et al., 2025) found that people using AI for writing were actually less mentally engaged. Their memory recall dropped, and they felt less ownership of their work. Another study by Becker et al. (2025) found that even professional developers got slower when using AI tools – by almost 20%. So, if people think less and produce worse results with AI, why was I having the opposite experience?

The Difference Between Copying and Learning

I think it comes down to ownership.

Those studies gave people pre-defined tasks – write this essay, fix that code. In my case, I wasn’t completing someone else’s assignment. I was solving my own problems, things that mattered to me and my work. When I use ChatGPT for DAX, I’m not outsourcing my brain. Each time I test a formula, I understand a bit more about how Power BI “thinks.” The difference is that the feedback loop is instant. I don’t have to wait days for an answer.

Before, I often hesitated before starting something new: Do I even know enough to try this?
Now it’s more like: Let’s see what happens.

The biggest shift for me isn’t technical – it’s psychological.

Maybe that’s what the research doesn’t capture. For me, AI hasn’t made me think less. It’s made me more curious, more confident, and more willing to experiment.

References

Kosmyna, N., MIT Media Lab, Hauptmann, E., MIT, Yuan, Y. T., Wellesley College, Situ, J., MIT, Liao, X.-H., Mass. College of Art and Design (MassArt), Beresnitzky, A. V., MIT, Braunstein, I., MIT, Maes, P., & MIT Media Lab. (2025). Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task. In MIT Media Lab. https://arxiv.org/pdf/2506.08872

Becker, J., Rush, N., Beth Barnes, David Rein, & Model Evaluation & Threat Research (METR). (2025). Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity [Journal-article]. arXiv. https://arxiv.org/pdf/2507.09089

Please rate this

How Generative AI Became My Job-Search Partner.

9

October

2025

No ratings yet.

I first started using GenAI tools for internship search during my bachelor’s studies, mainly to save time when preparing applications. What began as a small shortcut quickly turned into something much more valuable. Over time, I learned to use AI not only for writing but for analysing vacancies, identifying missing skills, and tailoring my applications to what employers were really looking for. I often asked it to highlight key competencies and keywords that could make my CV or motivation letter stand out. Combined with insights from HR professionals and career influencers I follow online, I developed a method that feels strategic and personalised, where AI acts as a sparring partner rather than a content generator.

GenAI also became part of my preparation for interviews and assessments. I’ve used it to simulate common behavioural and technical interview questions, structure my answers, and gain feedback on how clearly my reasoning comes across. When preparing for the TestGorilla assessment, I applied the same study habits I used for university exams by asking ChatGPT to create logical reasoning and situational judgment exercises and to explain how recruiters might evaluate responses. It made me feel more confident, structured, and aware of my performance.

Looking back, I’d say GenAI has become a real career companion by helping me work smarter, reflect more deeply, and stay organised throughout the job-search process. However, this experience changed how I view AI in the recruitment process. While it saves time and helps formulate ideas clearly, it doesn’t remove the need to put effort into each application. The outcome still depends on how well I can express my own story, experiences, and motivation. AI can refine my words, but it can’t replace authenticity or individuality. In the end, standing out still comes from the human touch behind the screen. Do you think AI will ever fully understand what makes a candidate truly unique?

Please rate this

The Creative Cost of Convenience

9

October

2025

No ratings yet.

While it’s now part of an exercise to write blog posts in which we reflect on our experience with AI tools, this topic has been on my mind for quite some time. I often find myself wondering: who am I without the assistance of these AI things? What is my added value?

Most interesting is that it hasn’t been long since the first consumer-facing AI tools entered our lives, and yet I increasingly hear from people, both within and beyond the university environment, that they can hardly remember what life was like before these tools existed. That’s wild, especially considering how quickly we’ve become reliant on them.

And I’m no exception. Although I was a late adopter of the technology, I now use AI regularly, across a wide range of tasks. In academic settings, I mostly use Copilot and ChatGPT for refining my ideas, generating feedback on my work, and correcting my writing, especially since English isn’t my first language. Outside of the university environment, I recently started to experiment with image and video generation tools, as they allow me to express my creativity in a completely new and faster way. In those moments, AI feels like a complementary creative partner, and using it that way is genuinely fun and empowering.

However, I also notice a growing trend: more and more people are outsourcing entire workflows to AI, from brainstorming all the way to final drafts. And honestly, I can’t blame them. The efficiency gains from automating repetitive tasks are undeniable. But as others have pointed out in their blogs as well, this shift comes at a cost. I really feel like we’re gradually offloading our human capabilities, our creativity, our critical thinking, to machines. If we continue down this path, those skills may slowly fade away.

And I feel like it’s already affecting most of us, at least I know it has affected me. It changes how we think. A part of our problem-solving ability simply gets outsourced, and we don’t always notice it happening. That’s why I’ve started to pull back a bit in how I use generative AI tools. I’m worried that if I rely too much on these tools for my day-to-day tasks, my own creative and critical thinking skills might fade. And if that happens, what do I really bring to the table in future workplaces? If I depend on machines for everything, how can others depend on me?

That’s why, currently, I am working to restore my balance between thinking on my own and seeking ‘advice’ from these tools. And I believe we all should. AI can really be a great support, but it shouldn’t take over the parts that make us human. It should solely complement us, for instance to actually BE creative. The real challenge is to keep thinking for ourselves, stay curious and critical, and make sure we’re still using our own minds, even if it is so tempting to outsource our whole existence to these machines.

What do you think, is AI a step toward empowerment in the human environment, or could it become our downfall?

Please rate this