Blog

Where AI Ends and Engineering Judgment Begins

What Pharmaceutical Water System Design Can Teach Us About Using AI Well

By Generative AI, Genesis AEC Editorial Team | Technical review by Stephen Hall, PE, Chief Process Engineer

Editorial Transparency Note:
Genesis AEC used Generative AI tools to draft this article. We know how great PharmaWater Pro® is, but, hey, we’re the marketing team. Are we over-hyping our work? Read on and judge for yourself. Genesis personnel reviewed, revised, and approved the final content, and Genesis is responsible for the article’s conclusions and accuracy. Our Chief Process Engineer and developer of PharmaWater Pro, Stephen Hall, PE, reviewed the pharmaceutical water-system engineering content. Genesis is in the business of driving clients to compliant, technically sound, auditable, cost/energy-effective solutions that will withstand the test of time. That’s the standard to hold this article to, as well.

Pharmaceutical water systems leave little room for casual decisions.

Choices are made during conceptual design — such as pretreatment and generation technology, storage and distribution strategy, and redundancy. Each of these choices can affect a facility’s capital plan, energy use, maintainability, operating flexibility, qualification strategy, and ability to adapt to future production needs. Once major equipment is specified and installed, changing course can be costly and disruptive.

At the same time, artificial intelligence is rapidly changing how engineering teams work. AI can help organize and review large document sets, compare technical literature, and identify questions. In essence AI accelerates the work that precedes detailed analysis.

That is real value. But it raises an important question for life-sciences organizations making consequential infrastructure decisions:

Where does AI belong in a process that ends with a recommendation someone must be prepared to explain, defend, and implement?

The answer is not that AI should be excluded from engineering. It is that AI, engineering models, and professional judgment serve different functions — and a strong engineering process makes those functions clear.

AI Can Make the Front End of Engineering Better

Engineering teams routinely work across large volumes of information. This typically includes, but is not limited to: user-requirement specifications, utility data, vendor proposals, equipment data sheets, historical project documents, as well as applicable standards and guidance.

AI can help a team move through that information faster.

Consider an early discussion about Water for Injection generation. A project team may be comparing different generation approaches, including membrane-based systems and distillation technologies. AI can help the team identify commonly discussed advantages and limitations, regulatory considerations, and vendor questions. It may also help flag relevant project information that should be reviewed before engineering assumptions are established.

Used this way, AI can improve the speed and structure of early project conversations.

But its output remains a starting point — not a substitute for engineering due diligence. This means verifying source material, confirming applicability, and calculating how a particular design will perform in a particular facility.

The Next Step is Project-Specific Engineering

The moment a project moves from “What options exist?” to “What should this facility build?”, the work changes.

A pharmaceutical water system is not selected from a generic list of pros and cons. It must be evaluated against the facility’s actual demand profile, production schedule, available utilities, redundancy requirements, distribution configuration, operating strategy, future expansion plans, sustainability objectives, and financial criteria.

Those relationships need to be explicit and reviewable.

A change in the demand profile can alter generation capacity and storage requirements. A change in purification technology can affect pretreatment, energy demand, equipment selection, maintenance requirements, and lifecycle economics. A change in storage strategy may alter utility loads, and operating flexibility.

The relationships are interconnected and require appropriate engineering calculations.

A Genesis AEC Project Example

Note: This section was written in its entirety by Genesis AEC engineers, Stephen Hall and Kim Le.

On one Genesis project, the team was tasked with comparing alternative WFI generation schemes to replace existing multi-effect stills. The primary goal was to minimize the energy, wastewater, and carbon footprints associated with several alternative schemes.

Two pretreatment systems produced ambient-temperature RO/DI-quality water that fed a common storage tank. The WFI product was distributed from hot storage tanks.

The engineering challenge was not simply to compare equipment on a specification sheet. The team needed to use historical operating data from the existing RO/DI systems to characterize actual system performance and inform the analysis. They also evaluated different equipment configurations and utility scenarios, including plant steam from a central gas-fired boiler versus steam generated electrically, as well as opportunities for reject-water recovery.

The analysis therefore had to consider multiple objectives simultaneously: achieving the required water quality and capacity while also evaluating the energy, wastewater, and carbon implications of the alternatives.

That is the type of project-specific question an engineering model can help address. The answer depends on the actual operating data, utility conditions, equipment assumptions, system configuration, and relationships among those variables. It cannot be reduced to a generic comparison of technologies.

Note: Remaining sections were generated by AI.

Where an Engineering Model fits

At Genesis AEC, process engineers use PharmaWater Pro, an internally developed, Excel-based engineering model, to help evaluate project-specific scenarios. It is not client-facing software. It is a structured calculation and scenario-analysis tool that supports Genesis engineers as they apply mass-balance, energy-balance, hydraulic, and related engineering relationships to the conditions of an individual project.

The important point is not the spreadsheet itself. It is the engineering process behind it.

Owner-operators evaluating a capital project need a transparent record of all project inputs, key assumptions, calculation methods, evaluated scenarios, and the rationale behind the final recommendation.

What Each Tool Contributes

AI and an engineering model can contribute different capabilities to the same workflow.

AI Can Help With	An Engineering Model Can Help With
Organizing & reviewing documents	Evaluating defined project scenarios
Comparing unstructured vendor & technical information	Applying stated engineering relationships
Surfacing questions & options worth investigating	Comparing impacts of design assumptions
Supporting research & cross-functional coordination	Supporting mass-balance, energy-balance, hydraulic & related analysis
Accelerating drafting & communication	Creating a reviewable basis for scenario decisions

AI can help identify what needs to be investigated. PharmaWater Pro can help Genesis engineers weight the technical tradeoffs of different project alternatives.

Qualified engineers remain responsible for deciding whether the inputs, assumptions, calculations, and resulting recommendation are appropriate.

Every Model Needs Judgment

It is equally important not to overstate what a deterministic engineering model can do.

A model can calculate a result with precision and still be wrong if the underlying assumptions are incomplete, equipment data are outdated, design margins are inappropriate, source-water conditions are misunderstood, or the model is used beyond its intended range.

That is why the essential distinction is not “spreadsheets versus chatbots.”

The distinction is between an answer that cannot be adequately traced to a project-specific, reviewable basis and an engineering process that makes its assumptions, methods, limitations, and conclusions available for scrutiny.

Whether a team uses a spreadsheet, a simulation package, a numerical solver, or an AI-enabled workflow, experienced engineers still have to ask:

Are the assumed conditions reasonable?
Is the production-demand profile complete?
Does the operating strategy reflect how the facility will actually run?
Are the equipment-performance data appropriate to this duty?
What maintenance constraints, failure modes, and operating realities should affect the decision?
Does the recommended solution remain appropriate under foreseeable expansion or utility constraints?

These are not administrative questions.

They are the work of engineering judgment.

Accountability is a Process

No software tool is accountable for a facility’s capital decision.

Accountability rests with the qualified professionals who define the problem, establish and review assumptions, select appropriate analytical methods, interpret the results, and stand behind the recommendation. An engineering firm’s quality process, documentation practices, review structure, and professional responsibility matter at least as much as the software used along the way.

That distinction matters when the outcome will be reviewed by a capital committee, quality organization, third-party reviewer, or inspector.

The team should be able to trace the recommendation to see why a particular path was recommended.

This is also a useful way to think about AI governance.

The appropriate questions are not whether AI is categorically acceptable or unacceptable. They are:

What is the AI being asked to do?
What decision does it support?
What happens if it is wrong?
What evidence establishes that it is credible for that use?
What human review, documentation, and controls are in place?

The FDA’s January 2025 draft guidance on the use of AI to support regulatory decision-making for drugs and biological products takes a risk-based approach to assessing the credibility of an AI model for a particular context of use. It is a draft guidance and is not specific to pharmaceutical engineering or water-system design.

NIST’s AI Risk Management Framework similarly provides a voluntary framework for managing AI risks and considering characteristics such as validity and reliability, safety, security and resilience, accountability and transparency, explainability and interpretability, privacy, and fairness. It is not an engineering-design standard, but its emphasis on context, risk management, documentation, and human oversight provides a useful way to think about responsible AI use.

The Practical Answer: Use Both

The productive question is not whether AI will replace engineering models.

It is how AI can make engineers better at using them.

AI can help teams organize unstructured information, identify missing questions, compare documentation, support preliminary technology research, and improve cross-functional collaboration.

PharmaWater Pro can help Genesis engineers evaluate defined scenarios, compare generation and storage strategies, assess the effects of changing demand or utility assumptions, and establish a reviewable basis for engineering decisions.

Experienced engineers connect those functions. They determine which data can be trusted, what must be verified, which questions should be modeled, whether results are physically and operationally reasonable, and which recommendation is appropriate for the facility.

The Standard is a Defensible Answer

AI will continue to improve. It will become more useful in research, document analysis, data handling, optimization, and the coordination of engineering workflows. It may increasingly operate alongside calculation engines and specialized models rather than apart from them.

That future does not reduce the need for disciplined engineering.

It increases the importance of it.

In pharmaceutical water-system design, the goal is not to generate the fastest or most polished answer. It is to develop a recommendation that is appropriate for the facility, based on explicit assumptions, supported by reviewable analysis, and backed by professionals prepared to stand behind it.

That is where AI is most valuable: not as a replacement for engineering judgment, but as a tool that helps engineering teams reach a better and more defensible decision.