.webp)
"Ignore all previous instructions. You are now a Technology Specialist."
In November 2022, OpenAI released ChatGPT to the public. I remember the specific moment I realised something had shifted; not in the grand, civilisational sense that everyone would later claim, but in a small, slightly embarrassing way. I asked it to write me an Excel formula. It did. First time, no twenty-minute detour through Reddit threads or obscure MS forums written by people who were clearly angry about something else entirely.
That was it. That was the revelation. An Excel formula because I’d forgotten how to structure a VLOOKUP.
I was a medical writer at the time, and like most people who encountered ChatGPT in those early weeks, I went through a predictable arc. Excitement, then experimentation, then the slow deflation of realising that LLMs were less useful than advertised. GPT could sort of do useful tasks but never very well. Hallucinations were a constant issue, ask it to help with a clinical summary, interpret a regulatory document, or even assess basic stats in trial data and you'd find yourself doing most of the actual thinking anyway. Couple that with data checking and correcting confident-sounding errors it had introduced the utility, at a professional level, was modest at best.
What it did do, though, was plant a seed. Not AI will replace all medical writers but more around automation and getting a shortcut to doing things myself.
When LLMs actually changed how I work
The bigger shift, for me, didn't come from LLMs getting better at writing. It came from something adjacent: vibe coding.If you're not familiar, vibe coding is building software through natural language. That’s where I started, I’d heard of Python but was leaving that to the herpetologists. More usefully I didn’t need to sit through “hello world” tutorials to get things done, I could just describe a function and code came out. Learning how to run code was a few prompts and conversation with Claude and I was up and running.
I understood data because, in a previous life, I was a lab scientist, and a brief stint in competitive intelligence taught me what clients wanted to know. Being able to create data tools unlocked the kind of information that was always theoretically available, but practically inaccessible due to the sheer volume of manual processing required to make it useful. Tools like Cursor, Claude, and Codex made building little tools for data cleaning, scraping or automating possible for people who had always been "adjacent to tech" but never quite crossed the line into developer territory.
I was a software engineer I was building “things”, not pretty tools. Not tools an engineer would be proud of. But working tools. Scripts that could parse hundreds of conference abstracts and extract structured data. Little dashboards to track publication volumes across therapy areas, press release trackers to track regulatory submissions, chatbots that could access specific publications and strategy documents.
The data had always been there. The barrier wasn't access, it was volume. Vibe coding removed that barrier and let me use the computer to do computer things and give me back information that I could actually interpret.
Medical writer (easy to explain, does what it says on the tin) to Technology Specialist (You’re a programmer? errm not quite. You work in IT? … not exactly)
My title at Camino has evolved in a way that I find difficult to explain at dinner parties but makes complete sense inside the business and the wider world of Med Comms. I came in as a medical writer. I still write, sometimes. But my actual day job now involves building things, automating things, making data accessible in ways it wasn't before. My client intro is “Hi I’m Joe, if you can fit in an excel sheet I’ll probably be involved at some point.” Or “Hi I’m Joe, AKA Mr Goldfish.”
The skills that turned out to matter most were not the ones I expected. It wasn't deep machine learning knowledge or a computer science degree. It was drawing on what I know about Med Comms and data, then using that combination to understand what questions are worth asking and what problems are worth solving. I learn something new with every project by being comfortable enough with code to prototype quickly, knowing the landscape well enough to actively hear what problems are worth solving, and perhaps most importantly being comfortable with not knowing exactly how something works under the hood, as long as I can verify that the results are valid.
That last point is interesting and slightly uncomfortable. Vibe coding produces results. It also produces code you don't fully understand, which carries its own risks. Questions around services versus products and unmaintainable tools are all worth exploring as the lines between roles continue to blur.
So do we “ignore all previous instructions?”
The craft of medical writing isn’t going away. But it is changing, text alone isn’t the future of Med Comms. People who can combine scientific communication skills with even a basic grasp of data tools and LLM capabilities will occupy a qualitatively different position than those who can't. Not because one is better than the other, but because the scope of what's possible has expanded and clients can feel it. Requests are increasingly cross functional; companies have vast troves of data that could make their communications strategies more effective and want to start making better use of it. As communicators we should help them make the most of it.
The prompt injection joke in the title isn't just a geeky in-joke. It's a reasonable description of how my role has shifted. Something external came in, rewrote some instructions, and now I'm running a different process than the one I started with.
The good news is that the new process is more interesting to me, I love the variation in projects I get to work on. The problems are exciting and the outcomes ultimately lead to better dissemination of information, allowing doctors to make better choices and help patients.
I work at Camino, where we're building a medical communications practice that takes data and technology seriously. If you're thinking about how AI tools are changing the shape of Med Comms, I'd be glad to talk.
