Every few weeks I see another prediction that AI will replace energy modelers. Having spent hundreds of hours using AI tools alongside eQUEST, EnergyPlus, OpenStudio, spreadsheets, and custom workflows, my conclusion is more nuanced.

AI is already useful in energy modeling. In some areas it is remarkably effective. In others it is unreliable. And in a few areas it introduces entirely new risks.

The key question isn't whether AI will impact energy modeling. It already has. The more important question is where it adds value and where human judgment remains essential.

GUI Navigation and Software Support

Many modelers use AI as a substitute for software documentation. This can work for simple questions, but it has limitations.

Large language models will confidently tell you which menu item to click, even when the menu item does not exist. I have personally spent hours searching for fictional fields that AI repeatedly assured me were present.

This can be improved through careful prompting and by providing screenshots or documentation. Even so, GUI assistance remains one of the weaker applications of AI because software interfaces change frequently and hallucinations are difficult to detect until time has already been wasted.

When possible, always verify software-specific instructions against documentation or a known workflow.

Geometry Creation

AI still struggles with geometry.

While image recognition continues to improve, AI has difficulty accurately interpreting complex CAD drawings, floor plans, and architectural documents. It performs much better when geometry can be described through text, dimensions, coordinates, and structured inputs.

Even when geometry is accurately defined, human judgment remains essential. Determining appropriate thermal zoning, identifying perimeter versus core spaces, understanding occupancy patterns, and simplifying geometry for simulation are often more important than reproducing every architectural detail.

The challenge in energy modeling is rarely drawing the building. The challenge is understanding how the building actually behaves.

Creating Components and Inputs

This is one area where AI performs surprisingly well.

If provided with a sample, AI can generate schedules, constructions, equipment definitions, and other text-based inputs for EnergyPlus, eQUEST, and similar tools.

For example, AI can convert an Excel schedule table into an EnergyPlus schedule object or an eQUEST schedule definition within seconds. It can create repetitive objects, generate template inputs, and help translate information between formats.

However, attention to detail remains important. AI frequently introduces extra line breaks, formatting artifacts, or subtle syntax errors that can cause import failures. Verification is still required.

The good news is that AI is usually much faster at creating these components than a human. The bad news is that the last 5% still matters.

Bulk Editing and Automation

In my opinion, this is one of the most underrated uses of AI.

Energy models often require repetitive edits across hundreds or thousands of objects. AI is very effective at helping create regular expressions, search-and-replace routines, and custom scripts.

Even modelers with limited programming experience can use AI to generate Python scripts that make large-scale modifications to models. Tasks that previously required hours of manual editing can often be automated in minutes.

I've found AI particularly useful for:

• Editing large eQUEST input files
• Creating regular expressions for bulk text replacement
• Converting spreadsheets into simulation inputs
• Writing Python scripts for model transformations
• Automating repetitive QA/QC checks

For many practitioners, this may be the highest-return application of AI available today.

Results Analysis

AI excels at processing large amounts of tabular data.

It can quickly identify trends, summarize utility data, compare alternatives, and highlight anomalies in simulation outputs. For large spreadsheets containing thousands of rows, AI often spots patterns that would take significant time to identify manually.

This is especially valuable when reviewing annual simulation outputs, calibration results, utility bills, or ECM comparisons.

That said, important findings should always be spot-checked. AI can occasionally misinterpret units, labels, or relationships within complex datasets.

Trust, but verify.

Calibration Support

With sufficient context, AI can provide useful calibration suggestions.

It is particularly good at recognizing patterns such as seasonal biases, load profile mismatches, occupancy assumptions, and operational inconsistencies.

For example, if a building consistently overpredicts winter gas consumption while matching summer electricity usage, AI can often suggest several plausible causes worth investigating.

The challenge is that calibration problems are frequently highly specific to individual buildings. AI tends to generate plausible explanations, some of which may be completely wrong.

Outliers, unusual operating conditions, maintenance issues, control sequences, and occupant behavior still require engineering judgment and field knowledge.

Calibration remains as much an investigative process as a technical one.

Results Interpretation

AI can be a useful sounding board for interpreting results.

It can help summarize findings, identify unusual relationships, explain trends, and suggest additional analyses worth exploring.

What it cannot reliably do is determine whether a result is physically reasonable.

Experienced modelers develop intuition over time. They know when an energy use intensity seems unrealistic. They recognize suspicious end-use breakdowns. They understand when a savings claim appears too good to be true.

AI can assist with interpretation, but it cannot replace engineering skepticism.

ECM Identification

AI can be an effective brainstorming partner for energy conservation measures.

Given enough information about a building, it can generate a broad list of potential improvements and identify interactions between systems.

However, the quality of recommendations depends heavily on the information provided.

If local labor availability, contractor capabilities, maintenance requirements, owner preferences, and budget constraints are not included, AI often proposes solutions that are technically feasible but impractical in the real world.

The best ECMs are not necessarily the most efficient. They are the measures that can actually be implemented.

Reporting

Reporting may be the area where AI provides the greatest immediate value.

AI can quickly organize findings, summarize results, draft narratives, create report structures, and help communicate technical information to non-technical audiences.

Tasks that previously consumed hours can often be completed in minutes.

The concern is that AI has become so good at generating professional-looking reports and filling out templates that it may become difficult to distinguish between a report based on actual analysis and one generated with very little underlying work.

In other words, AI can make a mediocre analysis look impressive.

As AI adoption increases, technical reviewers may need to spend less time evaluating how well a report is written and more time evaluating whether the underlying analysis is credible.

The Real Risk

The greatest risk is not that AI will replace energy modelers.

The greater risk is that AI will make it easier for inexperienced practitioners to appear competent.

Energy modeling has always depended on engineering judgment, understanding building systems, identifying bad inputs, recognizing questionable assumptions, and knowing when results do not make sense.

AI can accelerate many tasks.

It cannot replace the experience required to know when a model is wrong.

The modeler of the future will likely spend less time entering data and more time evaluating assumptions, validating results, and making decisions.

That's where the real value has always been.

What's Next?

This article has focused primarily on general-purpose AI tools such as ChatGPT, Claude, Gemini, and Copilot.

However, a new generation of products is emerging that combines AI, machine learning, and energy modeling in ways that were difficult to imagine just a few years ago.

Over the past year, I've had the opportunity to evaluate several of these technologies. Some are AI-assisted energy modeling platforms designed to accelerate model creation and analysis. Others leverage machine learning trained on thousands—or even millions—of simulations to predict building performance, identify optimization opportunities, or create building-specific energy prediction engines.

In one case, I worked with a firm that generated thousands of simulations for a particular building type and used the results to develop an AI-driven prediction engine capable of rapidly evaluating design alternatives. The technology was impressive, but it also raised important questions about transparency, training data, accuracy, and where engineering judgment still fits into the process.

Over the coming months, I'll be exploring these technologies in greater detail, including:

• AI-assisted energy modeling platforms
• Machine learning for building performance prediction
• AI tools for calibration and analytics
• AI-assisted commissioning and fault detection
• AI workflows for eQUEST, EnergyPlus, and OpenStudio users
• Product reviews and interviews with developers
• Training opportunities for practitioners looking to leverage AI effectively

I've also launched an AI & Energy Analysis interest survey to identify the most promising use cases, products, and training opportunities.

If you're using AI in your workflow, developing a product, interested in AI-related training, or simply want to follow the series, I invite you to join the community:

https://energy-models.com/ai-and-energy-analysis-interest-survey

Registration takes less than 3 minutes.

The future of energy modeling may not be AI versus engineers.

It may be engineers who know how to leverage AI versus those who don't.