This is definitely not a financial advice article. It simply begins with an example to illustrate how generative AI integrates into daily life. We have always needed access to knowledge: we bought books, asked questions on IRC channels and forums, became regulars on StackOverflow, and now we have tools far more practical than all of them combined.
Türkiye is currently going through a mutual fund crisis, and for the past few days, I’ve been sitting back trying to figure out what saved me from getting caught in it. Was it sheer instinct, the reluctance to deal with reporting income to the Finanzamt and filling out an Anlage KAP-INV here in Germany, or simply the glaring asymmetry between risk and reward? I can’t quite tell. After all, I used these investment vehicles for a while myself, and what ultimately kept me selective (and helped me steer clear) was a virtual advisor.
I remember thinking to myself: “I have a financial advisor in Germany; why not have one for Türkiye too?” A few months ago, I built a custom “GEM” in Gemini, configuring it to act as if it were a real financial consultant, and asked it every single question on my mind. We would debate back and forth in lengthy sessions. Through these discussions, I realized I had been making critical mistakes by overlooking aspects of the decision-making process I had never previously paid attention to, such as how a fund’s risk rating is calculated on TEFAS or KAP, the relevance of inception dates, and why reverse repos matter.
At no point during these conversations did I ask for or receive fund recommendations. In every question I framed, I made sure that agency and final judgment remained firmly with me. I also cross-referenced its responses with additional research to ensure the information was accurate and consistent. This process evolved into a loop with two distinct bottlenecks:
The first bottleneck is ourselves. I realize calling this a “bottleneck” carries a negative connotation, but here it functions more like a supervision gate. Think of it like a turn in a turn-based game such as Civilization. We could surrender all control to the computer and settle for merely watching the game play out (sitting outside the loop as a passive observer), but this is not a game. You are putting years of hard-earned savings on the line, and your trust in the other party is inherently limited; after all, it has nothing to lose. That is why you must retain your agency and frame your questions accordingly. This question of agency becomes even clearer when applied to your own domain. I am not a financial expert, but I am a software engineer; I can immediately tell whether AI-generated code is garbage or not. But how much can we trust the output in domains outside our expertise? By asking the right questions. And asking the right questions demands expertise, experience, and mental productivity of its own, because the model only knows what you prompt it to consider.
The second bottleneck is the counterpart’s boundary of authority. By “counterpart,” I mean the persona assigned to the AI, such as that of a financial expert. What it can do on your behalf is currently very limited, even though the infrastructure for agentic workflows is being actively built. For now, it cannot speak with your bank branch manager, negotiate quotes, execute your fund decisions, or proactively deliver reports detailing profit-and-loss scenarios for managed funds unless explicitly requested.
I experience this exact same “agency and advisor” relationship in a far more personal aspect of my life: my hearing aid journey. When I got my first device, the best source of knowledge available was an acoustician. An ENT specialist examines you, refers you to an audiologist to determine whether you need a hearing aid or a cochlear implant, and the audiologist runs tests and shares the results back with the doctor. If the doctor concludes that a device is necessary (the doctor holds the ultimate medical authority; I only decide whether I am mentally ready for a device or surgery), you take those test results to a hearing aid acoustician.
An acoustician is essentially a craftsperson, practicing a trade much like writing code. In the past, the acoustician would recommend a device, and you either liked it or you didn’t. If you weren’t satisfied, they would fit a more powerful receiver, configure presets for different noisy environments or music listening, service the device when things went wrong, update firmware, take ear mold impressions, provide loaner units, and so forth.
And today? You guessed it: I have a virtual acoustician as well. It doesn’t have the physical equipment to take an ear impression, of course, but when I visit a real acoustician, I can walk in as an informed patient, fully aware of what material the ear mold should be made of and what size venting hole I need, to clearly articulate what I’m looking for.
When collaborating with these virtual craftspeople, editors, or advisors, positioning yourself as the decision-maker within the loop makes the entire feedback cycle far more potent. It is not enough to be merely an inspector or a passive observer; you have to remain intellectually active and direct the process.