By
She is the only person left in quality at her organization. Not the only one on her team. The only one.
She told me, without drama, that there is no way she does her job without her agents. Not that they make her faster. That the work does not get done otherwise.
She is not an outlier. She is the norm, and we have all watched it happen.
Three days before KENX AI in GXP in Raleigh, the people building the frontier models asked the world to slow down. I want to take that seriously, and I want to ask: slow down who?
Because in life sciences the function that would do the slowing has been cut to one person. The industry has spent the last two decades slowly reducing quality oversight, quietly, one reorganization at a time, and calling the result efficiency. We renamed quality as overhead. Then we acted surprised when the organization started treating it like overhead. (note: All the while IT budgets have surpassed all other parts of the organization)
Here's why this matters: a frontier lab pacing capability development and a quality organization pacing adoption are managing entirely different risks. One is about what a model might become. The other is about whether the last trained person in the building has the time, the budget, the colleagues, and the tools to build quality into the process rather than inspect for it at the end. You cannot pace a function you have already cut to one.
This is not an AI story. It is not a technology story. It is not even a compliance story. It is the cumulative result of a hundred headcount decisions, each one defensible on its own quarter, none of them ever added up.
And so the agents arrive. Not as adoption. As life support.
That distinction matters more than anything else I heard this week, and she is not alone. When a lone quality professional builds agents to survive the job, that is not an organization embracing AI. That is an organization discovering that the thing it deleted was load-bearing, and quietly outsourcing it to software without ever writing down that it made that choice. No conscious decision. No documented rationale. No defensible criteria. Just a person holding the line with tools she configured herself, on her own time, because nobody else was coming.
Ask her to slow down. See what happens.
What discipline actually looks like
Two conversations from the same week ran the other direction, and both were about deciding rather than reacting.
The first was with a product manager at an AI company whose software watches manufacturing work through vision models. Her team had just been through validation. They could have bluffed it. She knew it, and she said so. Instead they brought in quality advisors, because she understood how the bluff ends: the first customer reads the materials, gets halfway down a page, and asks what any of this actually means.
She has used generative AI since early 2023 and does not trust its output. Not from fear. From use.
What I told her is what I tell any software company being sold a validation package right now. The word validation is heavy and it is being sold to you. You do not need validation. You need a software development lifecycle you actually run. Written requirements, a ticket per change, defined stages, evidence it moved through them. Run that and you do not have a problem. A vendor does not become compliant by buying a label.
We keep buying validation when what we lack is discipline. And we keep cutting the people who would supply it.
(Sidenote: Within the last six months I have witnessed a surge of new 'apps' appear in online feeds and come to market, have they actually produced a 'quality' product robust enough to support the life sciences? this is a separate blog post for another day)
What constraint teaches
The second conversation was with a small manufacturing team who arrived at the conference convinced they were years behind everyone in the building. By the second afternoon they had worked out they were ahead of organizations a hundred times their size.
Not better tools. They still keep paper master records. Constraint forced them to fix the process first and add the technology after. That is the order, and scale is precisely what lets an organization avoid it.
One of them had run both manufacturing and QA. He got QA down to a quarter of its previous size, not by cutting corners but by resolving the exception at the moment it occurred, on the floor, in real time, so that by release everything had already been through what mattered.
That is quality by design. Shift left as far as the process allows, and the function contracts to genuine exceptions. Note the difference: he reduced QA by building the process. Again because he HAD to. There is no one else to do this left.
Their IT lead does not believe for a minute that AI removes the human, and he is right. The models cannot keep your controls current. Somebody reviews. Somebody approves. The labor has not disappeared.
And what convinced him it was worth doing was not a return-on-investment model. It was the relief on a young operator's face when her backlog cleared and she wanted to come to work again.
Life Sciences are reducing QA by removing people, despite the overwhelming need for Quality.
Where the investment should go
So, to the frontier labs: the call for pacing is credible and I am glad it was made. But pacing is the cheapest possible contribution. If the concern is that the world is not ready, then fund readiness. Put the money into education. Into curricula that do not exist. Into training the people who will have to make these judgments in regulated environments for the next thirty years. We are graduating scientists who have never seen a batch record and will spend their first month reading SOPs, wondering what happened to the job they trained for. That gap is fixable with money and attention, and you have both.
And to the global life sciences industry: stop buying shareholder value with headcount. It is the one lever you control completely. Not the models. Not the regulators. Not the market. You decide whether there is one person left in quality or a function capable of thinking.
Every layoff announced as discipline is a decision to have less judgment in the building next year. We have made that decision, over and over, and then expressed concern about whether we are ready for AI.
We are not ready. Not because the technology moved too fast. Because we spent a decade removing the people who would have been ready.
Invest in people. That is the whole answer. It has been the whole answer for a while.
Thank you to those that shared your stories with me and I'm glad that I provided, even for a moment, was a space for you to seen, heard and valued for who you are and grateful for the work you do!
