Engineering work has two halves: the analysis and the paperwork around it. AI is already useful for the second (reports, specs, RFI responses, meeting notes, test plans) and useful with supervision for the first: setting up a calculation, scripting the test-data reduction, checking a unit conversion, drafting a failure-mode list you then complete.
Current chat models reason step by step, write and run Python, read a photo of a sketch, and summarize a 200-page manual. They also make arithmetic mistakes, quote clauses from the wrong edition, and invent material properties with total confidence. Treat them like any unverified source: a starting point that gets checked.
One line does not move. A licensed engineer's seal means a professional in responsible charge has reviewed the work and takes responsibility for it. AI output is never a sealed deliverable, never the calculation of record, and never an excuse.
Quick wins this week
- Paste a hand calculation and ask the model to check it step by step, stating every assumption and unit. Where it disagrees, one of you is wrong; find out which.
- Ask for a Python script that reads your test-rig CSV, converts units, plots the curve, and flags outliers. Run it on a file where you know the answer.
- Describe a subsystem and ask for a first-pass FMEA table (failure mode, effect, cause, detection) to correct and complete; editing a list beats starting one.
- Paste an RFI and the relevant spec section and ask for a draft response citing it, then confirm the citation is the section you pasted, not one it remembered.
What AI can do for engineers, task by task
First-pass calculations and hand-calc checks
State the problem completely: givens, units, assumptions, expected method, and governing standard. Ask for a step-by-step solution with the arithmetic done in code. Verify with your own calculation or an independent method, and take material properties from the datasheet, never from memory.
Scripts for data reduction and analysis
Ask for Python, MATLAB, or VBA that reads the raw file, converts units explicitly, applies your filters, and produces the plot and summary table. Test it on a dataset with a known result, read the unit handling line by line, and keep the script in version control with its report.
Reports, specs, and submittal responses
Paste your notes, the calc summary, and the required section headings, and ask for a draft in the firm's style. Then edit for precision: every value must match the calculation of record, every claim must be defensible, and the review-and-approval block stays human.
Standards and code lookups, as a study aid
Ask what a clause requires and where to find it, then confirm in the edition your jurisdiction or contract adopted; models mix ASCE 7-16 with 7-22 and NEC 2020 with 2023 without noticing. Loading your own licensed PDFs into NotebookLM keeps answers grounded in the actual text.
Failure analysis and design reviews
Describe the component, its loads, environment, and history, and ask for a failure-mode list, a fishbone of candidate causes, and a design-review checklist tailored to it. Use the output to catch omissions; the physics, the test plan, and the conclusion are yours.
Industrial engineering: SOPs, time studies, and line balancing
Turn a video transcript or shop-floor notes into a draft SOP with steps, safety callouts, and quality checks, and ask for a first-pass line balance or takt-time calculation in code. Validate against measured cycle times; the model has never seen your floor.
Prompts for engineers
Replace the bracketed placeholders, paste into any chat assistant, and iterate on the result.
Check my hand calculation step by step
You are a senior [DISCIPLINE] engineer reviewing a junior engineer's calculation. Problem: [DESCRIBE THE PROBLEM]. Givens with units: [LIST GIVENS]. Governing standard or method: [STANDARD OR METHOD]. My calculation: [PASTE THE STEPS AND RESULT]. Re-derive the result independently, showing every step with units and using code for the arithmetic. Then list where our results differ, which assumptions you questioned, and any limit or safety factor I missed. Ask for material properties rather than assuming them.
Tip: If it agrees instantly, ask for a different method; agreement is cheap.
Script to reduce test data with explicit units
You are writing a [PYTHON / MATLAB / VBA] script for an engineer. Input: a CSV with columns [LIST COLUMNS AND UNITS]. Output: [DESCRIBE PLOTS AND SUMMARY VALUES]. Rules: convert units explicitly with named constants, never silently drop rows, flag and list outliers using [METHOD OR THRESHOLD], and print the row count at each step. Include a test on this sample that checks the known result: [PASTE SAMPLE ROWS AND EXPECTED RESULT]. Comment every formula with its source.
Tip: A script with no known-answer check is not done.
First-pass FMEA for a subsystem
You are a reliability engineer facilitating an FMEA. Subsystem: [DESCRIBE THE SUBSYSTEM AND ITS FUNCTION]. Operating environment and duty cycle: [DESCRIBE]. Known history: [FIELD FAILURES OR NONE]. Produce a table with failure mode, effect, cause, current controls, and detection method for each component, leaving severity, occurrence, and detection scores blank for the team. Flag any mode where a single failure could cause injury. Ask clarifying questions first if the description is thin.
Tip: Score it as a team; the model's job is the blank table, not the ratings.
Draft an RFI response with a cited section
You are assisting an engineer of record with an RFI response. The RFI: [PASTE THE RFI]. The relevant contract documents, verbatim: [PASTE THE SPEC SECTION AND DRAWING NOTES]. Draft a response that answers the question directly, cites only the pasted sections by number, and says clearly if the documents do not answer it. Cite no code or standard I did not paste. End with anything the engineer must verify before signing.
Tip: Paste the section text; given only a number, it quotes from memory.
Site-visit report from field notes
You are drafting a site-visit report for a [DISCIPLINE] engineer. Project: [PROJECT NAME AND LOCATION]. Visit date and purpose: [DATE AND PURPOSE]. My dictated field notes with photo references: [PASTE THE NOTES]. Produce sections for attendees, numbered observations with photo references, deficiencies with recommended action and priority, follow-up items, and limitations of the visit. Keep every measurement exactly as noted, mark anything ambiguous with [CONFIRM], and add no observations I did not record.
Tip: Dictate measurements with units on site; it cannot recover a number you never said.
Want a prompt for something else? Use the Prompt Builder.
Skills to build
Stating the problem completely
Why: The model solves the problem you describe, not the one you have; a missing boundary condition produces a confident wrong answer.
How: Write givens, units, assumptions, method, and governing standard before you prompt, as you would set up a calc sheet. If you cannot state it, you are not ready to ask.
Verifying independently
Why: A second method, a bounding case, and a unit check catch most of what a language model gets wrong in engineering arithmetic.
How: For every AI-assisted calc, do a hand check, a different method, or a known limit case, and record which you did.
Reading generated code well enough to trust its numbers
Why: A script that runs is not a script that is right; unit errors and silent row drops hide in clean output.
How: Read every unit conversion and filter line, run the script on a dataset with a known result, and keep it in version control with the report.
Edition and jurisdiction discipline
Why: Codes change by edition, adoption varies by state and city, and models blend editions freely.
How: Keep a list of adopted editions per active project and confirm every AI-cited clause against that edition before it appears in a document.
Confidentiality and export control
Why: Drawings, specs, and test data are often under NDA, and defense or aerospace technical data can be ITAR or EAR controlled; consumer AI tools cannot handle either.
How: Learn which projects carry export-control or NDA restrictions, use only the firm's approved tools for them, and never photograph controlled drawings into a chat.
Tools worth knowing
ChatGPT
The general-purpose AI assistant most of your coworkers already use.
Claude
A careful writing and analysis assistant that shines on long documents.
Gemini
Google's assistant, strongest when your work already lives in Google Workspace.
Microsoft Copilot
AI inside Word, Excel, Outlook, and Teams, with your company's data protections.
NotebookLM
A research notebook that only answers from the sources you give it, with citations.
Perplexity
An answer engine that cites its sources, built for research rather than chat.
Cautions for engineers
A professional engineer's seal certifies that a licensed engineer in responsible charge reviewed the work and takes responsibility for it, and state boards do not recognize a model as a reviewer. AI output can inform a calculation, but the calculation of record is the one you performed or independently verified, and your review must be documented.
Models make arithmetic errors, drop units, misapply safety factors, and state material properties, load factors, and allowables from memory, often wrong. Have them compute in code, verify with a second method or a bounding case, and take every property from the datasheet or the standard. A result that looks reasonable is the dangerous kind.
Models cite code sections that moved between editions, quote superseded requirements, and occasionally invent section numbers. Confirm every citation in the edition adopted by the jurisdiction or contract, and treat AI summaries of standards as a reading guide, not the requirement.
Technical data on defense, aerospace, or dual-use projects may be ITAR or EAR controlled, and client drawings are usually under NDA; a consumer AI tool is not an approved destination for either, and a photo of a drawing counts. Separately, the model has never seen the site or the as-built condition; field judgment, safety calls, and the final design decision stay with the engineer.
Your 30-day plan
- Week 1: Confirm which tools your firm allows and which projects carry export-control or NDA restrictions. Put your own verification rule in your calc template.
- Week 1: Use AI only for reports, meeting notes, and RFI drafts, checking every citation against the adopted edition.
- Week 2: Run five calculations you have already verified through the check-my-calc prompt; log where it went wrong and how.
- Week 3: Build one data-reduction script for a recurring test or survey task and validate it against a known result.
- Week 4: Add FMEA and design-review checklists to a live project, note in each deliverable what AI touched and who verified it, and share the log.
Frequently asked questions
Will AI replace engineers?
Can I use ChatGPT for engineering calculations?
Can AI-generated work be stamped by a PE?
Is it safe to upload drawings to AI tools?
Terms used on this page
Related roles
- AI for Software DevelopersCoding assistants write, refactor, test, and explain code, and agentic tools run the whole loop. Treat every generated line as a pull request from a fast, overconfident junior: useful, never merged unread.
- AI for Construction ManagersConstruction managers drown in paperwork that AI can draft in minutes: daily logs, RFIs, change order narratives, meeting minutes, toolbox talks. Here is how to use it on a real job, and why nothing safety-critical ever gets delegated to a model.
- AI for Project ManagersAI cannot run your project, but it can draft the status report, turn a messy meeting into an action list, and pressure-test your risk register in minutes. Here is how project managers use it without losing the plot.
- AI for Operations ManagersOperations managers live in SOPs, incident reports, staffing plans and spreadsheets, which is the material AI handles best. Here is how to use it to get your week back, and where the line is for safety and people decisions.