Before ChatGPT made all of this mainstream, I had beta access to GPT-3 while working as a data analyst. My day job was large-scale datasets for household brands. My nights were spent poking a language model that could suddenly do things software was not supposed to do.
Lesson one arrived fast: the model is only as good as what you feed it. With clean, structured context it looked like magic. With the messy reality of most company data it looked drunk. Everyone was staring at the model. The advantage was obviously going to belong to whoever had their data act together.
That is why I still open client engagements with the least exciting sentence in AI: data is the backbone of every AI strategy. It was true in the beta, it is true now, and it will be true for whatever model ships next quarter.
Lesson two took longer: capability arrives years before adoption. GPT-3 could draft, classify, and summarize in 2021 well enough to be useful. Almost nobody in a normal business was using it. The gap between what is possible and what is deployed is where a services company lives. That gap is still enormous.
I did not know it yet, but those two lessons were the founding documents of Friday Labs. Get the data right, then go close the gap.
