Can AI cut medicine costs? *AI in drug manufacturing* | Bryden Wood Podcast \\Martin Wood and Adrian La Porta
There has been significant investment in AI-driven drug discovery over the past decade, and the results are beginning to come through. New molecules are entering development pipelines faster than before. The question that has received far less attention – and that Martin Wood and Adrian La Porta sit down to examine in this episode – is what happens next. If drug development accelerates, can manufacturing keep up?
More questions than answers:
This conversation was recorded ahead of Bryden Wood's Accelerate event on the role of artificial intelligence in drug manufacturing, held at 101 Euston Road in November. It is an honest, exploratory discussion – both speakers are clear that they have more questions than answers at this stage – but those questions are worth taking seriously.
The quality problem
Adrian's central argument is that the pharmaceutical industry's current approach to quality is, in a fundamental sense, a compensation for ignorance. The paradigm of batch manufacturing – static processes, locked parameters, standardised sampling – exists not because it is the best way to guarantee quality, but because manufacturers do not have sufficient visibility into what is actually happening inside their processes at any given moment.
AI and advanced data systems change that equation. If you can gather vast amounts of process data in real time, correlate it with quality outcomes, and build a genuine understanding of what is happening inside the equipment, the static control model becomes unnecessary. Quality becomes something you know rather than something you enforce. That is, as Adrian puts it, a completely different way of looking at it – and a better one. But it requires a reversal of decades of ingrained practice, and a regulatory framework that is currently not designed to accommodate it.
The regulatory conundrum
Martin frames this as one of the episode's central tensions: quality frameworks are static, and the situation they are being asked to govern is dynamic and accelerating. AI-augmented quality systems are already being deployed in commercial manufacturing – Adrian cites an AI-powered final inspection system for sterile products that is increasing both yield and quality in live production. But these implementations are pockets of augmentation within an existing structure, not a transformation of the structure itself.
The deeper question is whether that structure needs to change. Regulation, Adrian notes, is fundamentally about trust – and trust is a human quality. The public taking these drugs wants to know they are safe, and the instinct to equate safety with human oversight is not irrational. Martin and Adrian are careful not to dismiss it: the goal is AI as human enhancement, not AI as a black box. But the current framework may be preventing the kind of systemic change that would actually improve quality, not just maintain it.
A disconnected pipeline
Beyond manufacturing, the episode touches on a larger structural problem: the pharmaceutical pipeline from strategy through development, manufacturing, and into healthcare is not, in practice, an integrated system. Decisions are made in silos. The feedback loop from drug use in the population back into manufacturing is extremely slow, in an era when wearable technology and real-time diagnostics could theoretically make it near-continuous.
Adrian uses clinical trial targeting as an example: AI analysis of large datasets is already enabling GSK and others to identify the precise patient populations most likely to respond to a given drug, improving trial success rates not by gaming the system but by finding the right drug for the right people. Extending that logic – from patient data back into manufacturing parameters, from manufacturing intelligence forward into development decisions – would require a new kind of digital backbone connecting parts of the industry that currently operate largely independently.
Continuous manufacturing, revisited
The conversation returns to continuous manufacturing – a concept that has been discussed in the pharmaceutical industry for 20 to 25 years without achieving significant commercial adoption. Martin's diagnosis is direct: it failed to take off largely because manufacturing was not ready for the drugs being developed, and because the people-centric, batch-by-batch paradigm was never seriously challenged.
AI changes the conditions for continuous manufacturing in a fundamental way. If you have real-time process understanding, you no longer need the isolation steps and batch boundaries that currently define the architecture of pharmaceutical facilities. Adrian goes further: at the extreme end, highly potent drugs produced in tens of kilos per year could be manufactured at laboratory scale, distributed closer to patient populations, and the boundary between development and manufacturing could begin to dissolve entirely.
What the event is for
Martin closes with a characteristic observation: the history of advanced nuclear power – where dispersed investment across hundreds of different approaches diluted the collective effort – offers a cautionary parallel for AI in pharmaceuticals. The risk is not that the ideas are wrong, but that the field fragments before it coalesces around a sufficiently coherent direction to make real progress.
The Accelerate event brought together big tech, big pharma, biotech AI companies, and academic researchers – people who do not normally occupy the same room – to begin that process of shared critical thinking. This podcast is the conversation that preceded it: a setting out of the questions, not a delivery of the answers.
Martin Wood is co-founder and Director at Bryden Wood. Adrian La Porta is Technical Director.
Watch the full episode on YouTube.
