Twenty years ago, as I prepared to graduate with a degree in English, everyone asked me the same question: “What are you going to do with that?”
At the time, it was fair. Specializing in language wasn’t a direct path into business or tech.
Today, that same skill has a new label: prompt engineering.
The ability to write clear, explicit instructions that provide context and guide interpretation has become valuable again in the age of large language models, because AI systems are only as precise as the directions and context they’re given.
Eventually though, the model “forgets” what you told it previously. Even with stored documents to “remind” the AI of past progress, at a certain point, the model can only process so much at a time.
And that’s the key: to get the most out of your AI models at work, it’s less about having the smartest model, and more about the context you provide.
Without enterprise data, your AI models are commodities
That context often comes with access to enterprise data. Whether it’s structured or unstructured, that data—including historical performance data, customer behavior, or business constraints—can provide much more specific outputs around your needs.
Ask an AI coding assistant without access to your data to build you a bespoke analytics application, and it’ll give you something polished, possibly even technically sound, but it won’t reflect your reality.
Give that same model governed access to marketing performance, customer cohorts, pricing dynamics, inventory signals and sentiment trends within a secure environment, and the output changes dramatically.
This is the shift many organizations are underestimating. Foundation models like Claude, OpenAI, and Gemini—which are trained on vast datasets and are designed to generalize across use cases—are rapidly becoming commoditized.
Your data is not. With marketing data and business data in the same enterprise environment, you can move beyond dashboards and into real machine learning workflows much faster than most teams expect.
If this English major can do it, so can you
I’ve seen this firsthand.
I was able to do work that would ordinarily require a month—coordinating across teams, standing up environments, tuning models—in one week of focused work, in between my day-to-day responsibilities.
An AI coding assistant worked behind the scenes, configuring hyperparameter variations and writing lines of code, as I focused on defining the business question, evaluating the output, and iterating.
While this doesn’t make us all data scientists overnight, it absolutely changes the pace at which teams can explore, test, and operationalize predictive models.