The best way to discover why you're irreplaceable is to try and automate your job. That's the counter-intuitive truth I stumbled upon while building what I call "Vanilla Agents."
Building the scaffolding that powers your Vanilla Agent is an exercise that will help you define what parts of your work are truly about you thinking and making “your magic happen”, and what parts are time-consuming and don’t really leverage your unique skills.
Two more things will become evident in this process, and will delineate what’s your real value added: 1. The questions you ask, 2. The insights you extract from information. More and more it’s becoming clear to me that good “prompt engineering” is not so important because of how the model understands (most models are now re-writing your prompt in the background to optimize it to their own way of “thinking”), it’s important because it uniquely shapes the result that you want to obtain. This is what will eventually guarantee that your competitors won’t reach the same conclusion as you, even if they’re using the same model or agent.
Part 1: codify your thinking, leverage the agent to do the research for you
Let's bring this concept to life with a practical scenario from the industry I work mostly with, Pharma. Imagine two competing business development managers, both tasked with assessing the market potential for a new immunology drug. Both have access to the exact same AI agent. They’d also have different proprietary data which will give them an additional edge, but let’s exclude that to simplify the discussion - at least for now.
The first manager takes a conventional approach. He asks the agent: "Summarize the current market for rheumatoid arthritis treatments and list the top competitors." He receives a competent, accurate, but generic report. It's a summary of publicly available data—a good starting point, but it's information anyone could get. His value-add is minimal.
The second manager embodies the new paradigm. She doesn't ask for a summary; she begins a dialogue. Her "prompts" are a series of strategic interrogations, born from her unique experience:
- "Analyze the prescribing patterns of the top 50 Key Opinion Leaders in rheumatology over the last 24 months. Cross-reference this with their published research and identify any discrepancies between what they write about and what they prescribe."
- "Now, segment the patient population from recent clinical trial data by genetic markers. Identify which sub-population shows the highest efficacy for our competitor's drug and model the unmet need if our drug demonstrates a 15% improvement in that specific cohort."
- "Finally, scrape the abstracts from the last three major immunology conferences and run a sentiment analysis on any mention of 'unmet needs' or 'treatment challenges'. Synthesize these findings into three potential messaging pillars for our drug launch."
The two managers used the same tool, but only one of them is using the agent as an augmentation of her own experience and expertise. The other one, is using it as a chatbot.
This is the crucial shift. The value is not in finding the information per se. The enduring, defensible value is in the architecture of the questions. The expertise is not just in knowing immunology; it's in knowing what to ask.
This process of building and refining a Vanilla Agent forces you to codify that unique line of questioning. The agent becomes more than a tool; it becomes a personalized extension of your own strategic mind, a partner trained to navigate complexity with your specific brand of curiosity.
Part 2: Add your company data to the mix
The second layer of complexity, and value, we can add, relates to connecting this agent to your proprietary data. That’s where the agent goes from “thinking like Lucrezia” to “thinking like Google”.
The process is no longer just about how you query the world’s information, but how you cross-pollinate it with your own. It’s about drawing conclusions from two distinct but now interconnected streams of evidence:
- On one hand, the vast, dynamic landscape of public knowledge that the agent finds via a well-grounded Google Search—the clinical studies, the competitor press releases, the market analyses.
- On the other, the treasure trove of proprietary, internal documents and information you've connected to the system—the unpublished trial data, the confidential sales reports, the internal R&D strategy notes that hold the secrets to your past successes and future ambitions.
When you ask your next defining question, the agent doesn't just look outward; it looks inward, too. It can now find the faint signal in a public research paper and amplify it with the hard evidence from your internal Phase II results. It can see a nascent trend forming in the market and immediately contextualize it with your own sales team’s on-the-ground feedback.
The insight you extract is no longer just a smartly-filtered version of what everyone else can see. It's a true synthesis, a novel conclusion born from the unique intersection of public fact and private knowledge. This is what truly guarantees no one will reach your same conclusion. They don't have your questions, they don’t think like you, and they certainly don't have your data.
Let’s make this less abstract. Imagine you’re a Medical Affairs lead for a successful drug that has been on the market for a few years. Recently, you’ve noticed a plateau in prescriptions in a key region. You have a hunch it's due to physician objections, but you don’t know why.
You start by pointing your agent outwards. You ask it to scan the public chatter: transcripts from recent medical conferences, physician forums, and review articles. The agent identifies a recurring, low-level concern. Doctors are citing a perceived side effect that, while on the label, is known to be extremely rare. It’s a soft signal, but it’s there.
In the past, this is where you'd start a long, manual process of surveys and focus groups.
But now, you turn the agent inward. You ask it a question that’s impossible for anyone outside your company to answer: “Cross-reference the list of doctors raising this objection with our internal CRM. Pull up all the call notes from our MSLs and sales reps from their meetings. At the same time, query our medical inquiry database for any direct questions we've received about this side effect.”
This is where the picture snaps into focus.
The agent reveals that the objection isn't random. It’s being driven by a single, influential, and slightly outdated review article from a well-respected KOL. Doctors aren’t seeing the side effect in their own practice; they’re echoing what they read.
But here’s the knockout blow: the agent also finds in the MSL call notes that when our medical team has a chance to proactively show doctors our latest Phase IV real-world evidence—data that lives only on your internal servers—the objection disappears, and their prescribing behavior changes within a quarter.
You’ve just gone from a vague problem to a precise, actionable strategy in minutes. The issue isn't the drug; it's a piece of misinformation. And the solution is a targeted educational campaign, led by your MSLs, armed with your specific proprietary data, aimed at the exact doctors who are being influenced by that one article.
This is the real magic of cross-pollination. It’s not just finding answers. It's connecting a public conversation to your private knowledge to uncover the why behind the what. That's an insight you can build a multi-million dollar strategy on, and it’s one your competitors will never find.
Closing remarks
So this is where the journey has led. What started as a simple exercise in automating the tedious parts of my work has revealed itself to be something far more profound. We've moved beyond just identifying tasks to offload, and into the realm of truly augmenting our own strategic thinking.
The agent, at this final stage, is no longer a productivity tool. It’s a strategic partner. It becomes an extension of your own mind, now fused with the collective memory and intelligence of your entire organization. The result isn’t just about working faster or more efficiently. It’s about thinking on a completely different scale, armed with insights that are uniquely, deeply, and defensibly your own.
And that thought brings me to a new, even bigger question. If this is the new frontier for creating individual value, what does it mean for how we build and lead our teams? What does an organization look like when every key player is empowered with their own agentic companion?
That’s the topic I’ll be exploring in my next article.
For those special ones who got this far into the reading
This article has been co-written with my personal Agentspace Agent. Here is, at a high level, what I produced and what I used its help for:
- I wrote the first paragraph
- I experienced a creative block so I asked Agentspace to suggest how to continue my flow
- I asked to change the writing style of its suggested continuation to be aligned with how I write
- I removed part of its suggestions and added a couple of more pointed sentences
- Agentspace wrote the closing remarks
- I re-read the article and asked it to change the very first sentence to make it punchier.
- I re-read it again and change the last few things directly
- If you're wondering about the podcast... that's still a secret (and no, it's not generated with NotebookLM).
That's all for today folks.
Let me know what you think.
Lucrezia
