Current pharma layoffs and the push for AI automation

I’ve seen the demo like everyone else, and for a few minutes it was hard not to be impressed.

The agent built a pre-call brief in seconds. What had changed since the last visit, the prescribing trend ticking up, the open objective, the affiliated accounts, a next best action and a clinical paper to share. After the call it transcribes the conversation, writes the summary and fills the compliance fields. The company behind this tool even has a tagline for it: From burnout to breakthroughs.

This is not about the future

Takeda is cutting around 4,500 roles this year, a 245 year old company stripping out layers as it leans on technology. Novartis, Moderna, Pfizer and more than a hundred other organizations have signed on to use AI agents that will decide what happens in front of healthcare professionals. Takeda bought the engagement engine a year before it announced the cuts which supports the notion that the restructuring and the AI-automation are not two separate stories, they are one and the same.

So the question is not whether this is coming, it is here. The question is what it replaces.

What the agent actually does

Strip the demo down and the agent does two things and does them well. It prepares, and it writes up. It tells the rep what changed, what to say next, and then files the report afterwards. For the hours a rep loses to logging calls and chasing compliance fields, this is a real gift, nobody will complain about too little admin.

But look at what got automated. The preparation and the paperwork. The start and end of the meeting, not the meeting itself.

The part the demo never mentions

The agent is only as clever as the data it learns from. And we know what’s is in our CRM, because we’ve all been filling it.

Our CRM holds what reps are allowed and rewarded to put there. Topics discussed, products mentioned, outcomes, follow-ups. It is product focused because that is the only thing they are permitted to log. It is bloated in places, because we built metrics that reward logging a visit, so visits got logged. And it is silent on the things we know will make a meeting great; what this account is really struggling with, where the patients fall out of the pathway, what the stakeholder said once the laptop was closed.

That silence is not something we can fix, it is built in. Storing qualitative information from a named doctor is a compliance liability, so the rule is simple, you do not write it down. The best thing a rep knows never enters the system.

So the agent uses what’s in there and builds a script based on very basic and generic data.

The vendor answer to this is to pull in outside sources, run the model over the free text notes. Fine. But it cannot recover what was never observed, or what compliance never let anyone record. You cannot model a thought that only ever existed in a rep’s head on the walk back to the car.

Soon everyone will sound the same

Fore example, Agentforce’s pre-call brief and next best action is hanging there, in a digital void, stripped out of context, because there’s no true feedback mechanism it can learn from other than what’s allowed to be entered into the system.

The thing AI cannot touch is the live adjustment to the person in front of you, the hypothesis put in the open and corrected on the spot. Forty years of research keeps landing on the same finding: The rep who adapts to what’s happening in the meeting beat the reps who run the same type of call every time.

And the doctors are voting. They now keep three companies and filter out the rest. They are not keeping the three with AI derived pre-call briefs. Soon every brief will look the same anyway, same summary, same suggested paper, same next best action, all generated by the same engine that sits within the same software platform. When the script is automated, the script stops being a differentiator.

I have said the same thing about the Challenger sales model; when the whole field is trained to teach, tailor and take control, the whole field starts to sound alike. Any model that scales is a model that removes the edges off everyone who uses it. Compared to Challenger, the agent just does it faster, and to more of us at once.

The good news hiding in this

The machine does not threaten the capability I keep coming back to in my articles, it clears away the work that was always getting in the way of the real job, and leaves the one thing that was never in its scope: the understanding, the relationship, the judgement and intuition that today lives in your head and nowhere in the system.

Everyone is about to have the same agent, AI is not a competitive advantage, it is the floor. The advantage will come out of what you feed it.

An agent running on product-focused call notes will hand you product-focused moves, faster. An agent running on a real picture of the account, what the system around the patient is actually struggling with, where people fall out of the pathway, what this stakeholder is trying to fix, will hand you something a stakeholder leans forward for. Same model. Different input.

Here’s what you need to do: start capturing the intelligence the standard CRM was built to ignore, in a place fit to hold it. And keep working on turning that captured intelligence into a conversation your stakeholders would like to have. The company that does both will stand out in a sea of look-alikes that follows the same script. Because the companies that buys the agent and feeds it what’s currently in the CRM will get faster versions of the calls that were already not working.

A note to you, reading this

If an agent reading your call notes can reproduce what you bring to a meeting, then in time it will.

But if what you bring is a real understanding of the account, and the understanding of how to use it, then the agent becomes the most useful colleague you have ever had, because you are finally giving it something worth working with.

The meeting was always yours. Start with the next one.