The new programming language: English
With AI's rise, effective communication in English is now vital for programming, shifting focus from coding syntax to clarity and context.
In case you haven’t noticed, the pedantic annoying colleagues around you who constantly correct your grammar mid-speech are having a moment. As AI models become a big part of our work and personal lives, all of us have realized the one thing we absolutely need to prompt an AI model to do our bidding is absolute command over plain old English. An ability to express what we want, in words, phrases and complete sentences, in brief, with clarity. The more specific your prompt, the more likely it is that you will be able to cajole the right answer, the right image and video, or as software programmers are realizing, the right code from an AI model.
This is dramatically different from what computer programming used to be five years ago. For the longest time, interacting with machines was through a special programming language that the machine could understand and humans had to learn. It was based in mathematics, a set of instructions, zeros and ones, syntax, data, variables, functions and code that a computer could understand and execute to perform specific tasks. In the 1960s, engineers instructed computers using COBOL and BASIC which quickly consolidated into C++ in the 1980s, giving way to Internet languages of Python and Java. Computer engineers spent years learning these syntaxes and functions, excelling in the logical art of conversing with a computer to create software, websites, applications, and other technologies.
STEM universities focused all their energies in educating their students in these languages so they could join the workforce to interact with computers, to build digital systems as we know it. So much so that the programming language training global market ballooned from $ 3.32 billion in 2018, to a projected $ 8.53 billion in 2028, an increase of around 10% year-on-year, according to a Technavio report.
All this while, a dictionary and a command over vocabulary began to gather dust in the digital world of instant messaging. Complete sentences and words gave way to distorted language, misspellings, abbreviations, memes and emojis. The aforementioned pedantic colleagues and editors were told they suffered an ailment similar to OCD, Grammatical Pedantry Syndrome (GPS).
AI models like it when you’re precise and concise
Just when we had resigned ourselves to breakups and business communication with thumbs up and hearts, along came AI models, and software engineers and the rest of us, suddenly needed to brush up our language skills. In February this year, Andrej Karpathy, OpenAI’s co-founder and AI age’s favourite software philosopher, was the first one to name this phenomenon of co-coding with an AI model. He created a rather poetic term, ‘vibe coding’, to describing his bromance with LLMs (Large Language Models), where he would cajole the model to create a code for an application he wanted to build. “You fully give in to the vibes, embrace exponentials and forget that the code even exists,” he said in a tweet, adding that all he needed to interact with the LLM was a command over English language so he could give the right prompts to the model. “The hottest new programming language,” he concluded, “is English.”
Along with prompt engineering, vibe coding soon became an adjective to describe a relaxed and intuitive approach to programming or coding through constant conversation and iterations with an AI models such as Claude, Gemini or ChatGPT. A style of building software not by writing lines of codes or syntax, but by describing what you want in succinct, plain, simple words and phrases of English language.
“I don’t remember ever, in the history of computer science, where we’ve abdicated logic to machines,” exclaimed Martin Casado, general partner at Andreesen Horowitz, one of the most influential venture capital firms in Silicon Valley. Talking with his colleagues in a recent podcast, he wondered if we require to rethink what it means to be a programmer, or what it means to create a software. What Casado and most of the AI community are trying to grapple with is how AI models have dramatically upended developer workflows, tooling and code, blurring lines between technical know-how and communication, between human and machine.
Spec-writing and context is the new superpower
For it is communicating with an AI model, that has quickly become an important artifact of the modern engineer. In the recently concluded AI Engineer World’s Fair in San Francisco, software engineers discussed how writing code was only 10-20% of what they would produce. “The most valuable skill is communicating intent with precision,” said Sean Grove from Open AI, giving a talk on ‘The New Code’.
Last month, the AI community developed yet another term to describe this concise interaction with an AI model to develop software code, perhaps first used in a tweet by Tobi Lutke, CEO at Shopify: context engineering. Karpathy jumped on the lexicon, further popularizing it. “Context engineering is the delicate art and science of filling the context window with just the right information for the next step,” he explained in a tweet. By tailoring information AI systems receive, context engineering is a way to make LLM output more effective. Give too much information, or a wrong description and a model will not give you accurate answers. Prompt it with too little and again, the output would be too general. Already the proponents of context engineering are building its lexicon: strategies such as Retrieval-Augmented Generation (RAG), quarantine, pruning, summarization, poisoning, distraction, and offloading.
Job requirements in the Valley and across the world have changed to reflect this. As AI models become part of an engineer’s life and role requirements, companies are looking for context engineers who can not only understand code, architect a product but also converse with an AI model in plain English. Last week, grappling with AI onslaught on their business, Wipro Ltd announced that it has mandated English competency tests for its senior executives, adding that those who fare poorly, might be put on performance improvement plans.
Does a context engineer need to write code?
As more and more students rely on AI models to write code for them, will the software developers of the future not be able to write code? As one of the test engineers in India who visited a university to recruit told me: “Students are coding using AI models and don’t even know how to write a ‘for’ loop in Python.” But with context becoming everything, do they even need to know what a ‘for’ loop is? This is the looming question for STEM universities today.
One thing’s for sure. The awkwardly mumbling geek using faltering English is dead. The new age software programmers will need to have a better grasp at vocabulary (unless that is outsourced to AI models as well), be able to communicate with AI models and agents and be concise and clear in their intentions.
Though some AI models do understand other languages such as Chinese, Japanese and even French and German, their contextual understanding is limited for now, with higher instances of hallucinations and errors. Most of the models have trained on English language data (this might change as well, in the quicksand world of AI). While the other languages catch up, it’s time to brush up on your skills to instruct, describe and express your intent in plain English. For writing code right now means full-sentence conservations with an AI system.
Shweta Taneja is an author and journalist based in the Bay Area. Views expressed are personal.
