The Spectrum of Building Environments

Why Simulation Modeling Tools Still Haven't Cracked the Code

Building in a virtual space – let it be a simple calculator app or complex 3D models – has always been about finding the right balance between power and accessibility. This tension has shaped how we build software, design interfaces, and create models for better understanding real-world systems. This tension becomes most apparent in simulation modeling, where tools must serve both technical experts and business users while handling increasingly complex real-world scenarios. Despite decades of evolution, no simulation platform has successfully resolved this trade-off.

The Evolution of Accessibility

Back in “ye olde” days, everything was based on commands or writing code. Without a more user-friendly visual medium, if you wanted to use a computer, you had to learn its language.

When 'Ready.' really meant 'you'd better be.'

Once graphical user interfaces and the computer mouse started entering the scene, the door of accessibility began to crack open. Although, any sort of customization still required learning how to translate from intention to code. Even as recently as the late 2000s, I'm sure many people, like me, first got a taste of what's possible through writing custom HTML/CSS for their MySpace profile.

As Moore’s Law continued driving capabilities forward, the possibilities for increasingly complex visual interfaces grew. Drag-and-drop website builders exemplified this trend, promising that anyone could create professional websites without writing code.

However, this accessibility came with hidden costs.

The Drag-and-Drop Trap

Despite what the marketing claims suggest, there are limitations to visual interface builders; they are not possible without making fundamental assumptions about user needs.  Developers must decide which features matter most and bake those decisions into their platform's design. This creates a constrained environment where deviation from the expected use case becomes difficult or impossible. Unfortunately, these constraints are often not noticed until projects are put into motion and soon find themselves hitting a wall.

The limitations become especially problematic in simulation modeling, where problems share surprising commonalities yet remain entirely unique. Consider these scenarios:

• A person withdrawing money from an ATM

• Surgeons performing an operation

• A forklift operator unloading boxes

• A machine processing raw materials

All can be modeled using a simple "delay" process - the fundamental building block of discrete event simulation. This commonality explains why every simulation tool includes delay mechanisms.

But the devil lives in the details. What happens when the ATM breaks down? When a surgeon gets called to an emergency? When unloading time depends on box weight? These variations require customization that visual interfaces struggle to accommodate elegantly. Providers typically use one of three routes (with downsides given):

1.       Adding more widgets to the UI (risks making the UI cluttered)

2.       Forcing the user to write code (difficult for business-oriented users)

3.       Omit that feature from the interface (forces clunky workarounds or, at worst, prevents the model from being accurately built.

The Current Landscape of Imperfect Compromises

The balance that most simulation tools provide is something in between pure coding and pure visual interfaces. They each occupy different points on the power-accessibility spectrum.

Arena and Simio emphasize visual modeling with minimal coding requirements. They excel at standard scenarios but become unwieldy or unusable when models need significant customization.

FlexSim strikes a middle ground, offering visual tools backed by scripting capabilities when needed.

AnyLogic assumes high technical proficiency, providing powerful coding features alongside visual elements.

Notably, the amount of coding required doesn't correlate with visual feature richness. Each tool makes distinct assumptions about user technical skills, model complexity, and industry focus. This makes direct comparisons difficult and forces users to align their needs with a tool's built-in assumptions.

Of course, you still have the traditional, pure-code options as well. To avoid implementing everything from scratch, there are many simulation related Python libraries – Simpy or Salabim (discrete event), PySD (System Dynamics), and Python State Machine (agent-based). These options provide the ultimate flexibility, at the cost of being the most technical and with little-to-no handholding. The visual aspect also plays a substantiative role for humans to maintain and understand what’s going on in the model. The lack of any visual interface means more cognitive effort needs to be put into maintaining it.

Even in a simple model, it’s easier to get a sense of what’s going on in a visual medium

The AI Wild Card: Promise and Peril

Large Language Models (LLMs) have introduced a new dynamic to this landscape. The prevalence of LLMs for coding has introduced the notion of "vibe coding" where non-technical users, who have zero understanding of programming, can just ask an LLM to build what they want and have a (hopefully) working prototype in under 20 minutes.

Unlike visual builders designed specifically for non-technical users, “vibe coding” is like giving unskilled people all the power tools and equipment needed to build a house and an instruction manual and then expecting a well-built house (with no injuries)!

The problem compounds in simulation modeling due to the field's niche nature. LLMs lack sufficient training data to understand the nuances of building accurate, optimized simulation models. Creating a model that's both technically correct and performant requires expertise that current AI cannot reliably provide.

The Elusive Solution

The ideal simulation platform would offer business users an interface that's simultaneously intuitive, deeply customizable, and capable of handling complex technical requirements automatically. This platform would need to understand both the business context and the technical implementation details.

Such a solution requires more than better interfaces or smarter AI. It demands deep domain expertise built into the platform itself - knowledge about simulation modeling best practices, optimization techniques, and industry-specific requirements.

Looking Forward

The simulation modeling industry stands at an inflection point. Traditional visual tools have reached their flexibility limits, while AI-powered “vibe coding” introduces new risks alongside its promises. The solution likely lies not in choosing between these approaches, but in thoughtfully combining them with deep simulation expertise.

Organizations evaluating simulation tools should consider whether their needs align with a platform's core assumptions, assess their team's technical capabilities, and honestly evaluate their customization requirements. The perfect tool doesn't exist, but understanding these trade-offs helps identify the best fit for specific use cases.

The quest for the ideal balance between power and accessibility continues, but the next breakthrough may require rethinking the entire approach rather than simply improving existing paradigms.