🧮 Using AI for Engineering Calculations: What Works, What Does Not, and How to Do It Safely

The Promise and the Problem

AI assistants can write calculation procedures, explain formulas, set up calculation templates, and check the logic of an approach in seconds. For engineers, this seems like an obvious productivity win. But AI tools also make arithmetic errors, misremember code values, hallucinate reference standards, and produce confidently wrong results — especially on calculations involving multiple steps, unit conversions, or lookup table values.

The engineers getting the most value from AI for calculations are not the ones blindly trusting AI output. They are the ones who understand where AI is genuinely helpful versus where it is dangerous, and who have developed workflows that capture the benefits while catching the errors.

Where AI Genuinely Helps with Calculations

Setting up the calculation procedure: "Walk me through the steps to calculate voltage drop on a 3-phase circuit per NEC" or "What is the NFPA 13 procedure for sizing a sprinkler system branch line?" — AI is excellent at generating the correct procedure, identifying the right formula, and explaining which code section or reference applies. The procedure is almost always right even when the arithmetic in a specific example is not.

Writing calculation templates and scripts: Ask AI to write a Python script that performs a specific calculation — load calculation, conduit fill check, pipe sizing — and it will produce working code that you can verify, run on your own numbers, and reuse. The code is easier to check than a step-by-step manual calculation because you can test it with known values.

Checking dimensional consistency: Paste a calculation and ask AI to verify that the units are consistent throughout. AI is quite good at catching unit mismatch errors that are easy to miss when you're deep in a calculation.

Explaining an unfamiliar calculation type: When you encounter a calculation method you haven't used recently — Hazen-Williams pipe sizing, Joukowsky water hammer, LMTD heat exchanger sizing — AI can give you a clear conceptual explanation and worked example that gets you oriented before you open the reference manual.

Checking your approach: Describe the problem and your proposed approach and ask AI to identify any flaws in the methodology. This is different from asking AI to do the calculation — you're using it as a technical sounding board, which it handles well.

Where AI Makes Dangerous Mistakes

Arithmetic on specific numbers: Large language models are not calculators. They predict likely token sequences, and while they often get arithmetic right, they make errors — especially on multi-step calculations, decimal placement, and unit conversions. Never use AI arithmetic output directly in a design document without verifying with a calculator or spreadsheet.

Code table lookup values: AI frequently confuses or misremembers specific values from NEC tables, NFPA standards, ASHRAE handbooks, and other references. It may give you the right table number but the wrong value, or the right value from the wrong edition of the code. Always verify table values against the actual code publication.

Jurisdiction-specific requirements: AI trains on general code language but often lacks specific knowledge of local amendments. A calculation that is correct per the model NEC may not comply with your state's adopted amendments. AI cannot reliably tell you about local requirements.

Recent code changes: AI has a training cutoff date. If you're asking about requirements from a recently adopted code edition, the AI may give you the answer from a prior edition without flagging that the requirement has changed.

Complex interdependent calculations: Multi-step calculations where an error in step 3 cascades through steps 4–10 are risky. AI may correctly execute most steps but make a small error early that compounds into a significantly wrong final answer.

A Safe Workflow for AI-Assisted Calculations

Step 1: Use AI to set up the procedure. Ask AI to walk you through the calculation methodology and identify all relevant code references, formulas, and table values you will need. Review this for completeness and accuracy before starting the calculation.

Step 2: Pull the actual code values yourself. Look up table values, correction factors, and specific requirements in the actual code publication. Do not rely on AI to accurately recall specific numbers from standards documents.

Step 3: Run the arithmetic independently. Use a spreadsheet, calculator, or verified software tool for the actual arithmetic. You can use AI-generated Python code as a calculation tool, but verify the code logic before trusting the output.

Step 4: Use AI to check your work. Present your completed calculation to AI and ask it to identify any logical errors, missed factors, or inconsistencies. AI is good at finding conceptual mistakes even when it shouldn't be trusted to do the arithmetic itself.

Step 5: Apply professional judgment. AI output is a tool, not a substitute for engineering judgment. The engineer of record is responsible for verifying that calculations are correct, code-compliant, and appropriate for the specific project conditions.

Prompting Strategies That Improve Accuracy

Ask for the formula before the numbers: "First give me the formula and explain each variable, then apply it to these values." This separates the conceptual setup (where AI is reliable) from the arithmetic (where errors occur) and makes the arithmetic step easier to verify.

Ask AI to show its work: "Show every step of the calculation." Errors are much easier to catch when you can see the intermediate steps rather than just the final answer.

Provide the code values yourself: "Using NEC Table 310.16, the ampacity of 4 AWG copper at 75°C is 85A. Now calculate..." — by providing the lookup value yourself, you eliminate the risk of AI misremembering the table value.

Ask for a sanity check: After getting a result, ask "Does this answer seem reasonable for a typical [circuit/load/system] of this size?" AI can often identify when an answer is orders-of-magnitude wrong, even when it can't reliably do the arithmetic that produced it.

The Liability Question

For licensed professional engineers, the use of AI tools in calculations raises a straightforward professional responsibility question: you are responsible for the work you stamp, regardless of what tools you used to produce it. AI-assisted calculations are no different from spreadsheet-assisted calculations — the engineer must understand the methodology, verify the inputs and outputs, and exercise independent professional judgment before applying a PE stamp. Using AI does not transfer or reduce that responsibility.