My ten-year-old daughter became obsessed with the show Wednesday and decided she wanted to play cello like the show's protagonist. I’m a firm believer in any kind of enrichment, so we rented a cello, found a teacher and suddenly there was this big, beautiful instrument sitting in our house.
Music has always been an important hobby of mine. I play piano and guitar as quick intermissions during work, when my mind gets stuck on seemingly unsolvable problems. Thus, once the cello was in the house, I couldn’t resist trying it too. I learned fast just how hard it is to get a string instrument to produce even a decent sound unlike piano and guitar.
Piano is a world of clear boundaries. Keys are fixed. Press the key and you get the note. Guitar is similar in spirit, frets create a grid that keeps you honest. Yes, there’s technique and expression, but the instrument gives you guardrails. It nudges you toward “correct.”
The cello doesn’t do that.
Between any two notes, there are infinite possible notes. You don’t “hit” pitch so much as you hunt for it. Then the bow adds another layer: pressure, speed, angle, contact point, how you start a note, how you end it. The same intended note can come out timid, harsh, warm, nasal, confident, depending on tiny variables you barely notice until the sound exposes them.
It was my first visceral lesson in how much the player matters. The same intended note can come out ten different ways and the instrument doesn’t protect you from that variability, it exposes it. To get consistency, you don’t follow a formula. You earn it through practice.
And here’s the observation that clicked for me: that’s exactly why so many users are having such uneven experiences with AI.
I come from a programming background. Programming rewards defined paths and formulas. Inputs go in, logic runs, outputs come out. If you run the same code with the same inputs, you expect the same result. Constraints are the feature. They’re how you get reliability.
AI isn’t like that. AI is more like a cello it reflects the quality of the operator and the environment.
The output depends on how you approach it: how you frame the prompt, what context you provide, what constraints you include, which model you’re using, and sometimes you can keep everything “the same” and still get a different response on a different run. People interpret that as instability, and they’re not wrong. But they’re often expecting piano-key behavior from a bow-and-string tool.
Vague prompts produce vague answers. Conflicting goals produce mush. But when you bring a clear intent, a strong brief, real constraints, and a definition of what “good” looks like, the model suddenly seems smarter. In reality, you’ve just started playing it with technique instead of an input/output machine.
So the takeaway I’m circling isn’t “AI is unpredictable.” It’s more practical: AI is not a single lane to follow. It’s a wide expressive surface. If you want it to behave consistently inside a business, you don’t only buy the instrument, you develop the ability to play it and you build the room around it: standards, guardrails, review and feedback loops that turn variability into performance. And like any instrument, if you want results that most resemble predictability, then you must practice it yourself religiously until the patterns become familiar, instead of implementing cookie-cutter "solutions" from prompt engineers or field experts.
Don't be frustrated by its unpredictable nature. Embrace it to enjoy the medium to forever improve your outcomes.
— Yas Dalkilic
Head of AI, RAB2B
