Discover how to use AI to best fit your welding needs
Technology has moved through the industrial trades in different ways. Sometimes as a promise, other times as a disruption. Occasionally, technological advances have changed what professionals can accomplish. When large language models (LLMs) first became publicly available, something shifted. Professionals now had access to a wealth of knowledge that lay beyond a simple internet search.
While LLMs, colloquially known as artificial intelligence (AI), have allowed me to push beyond what I already knew, they did not hand me expertise I hadn’t earned. AI expanded on what I already knew. That distinction is the lens through which every practical application and AI usage should be centered. This technology rewards mastery. In the hands of someone still building foundational knowledge, it can be a crutch that quietly undermines the very development they need most.
Talking to the Machine
Prompt engineering is the practice of refining the prompts fed into AI models to directly improve the quality of the responses. In other words, the quality of an AI response is determined by the quality of the request.
A question about preheat requirements means something different coming from a structural fabricator working to AWS D1.1, Structural Welding Code — Steel, than from a pipe welder working to ASME B31.3, Process Piping. AI won’t know the difference unless you tell it. That’s not a limitation of the technology — it is the point.
Providing context is half the battle. Tell the model who you are, what you are working on, and which constraints matter to you. Be specific about what you want back: a summary, a checklist, a comparison, or a plain-language explanation. Format matters, and AI will follow your lead if you give one.
Finally, treat it like a conversation. If the first response misses the mark, push back. Ask it to reconsider, add a constraint you forgot, or ask it to explain its reasoning. The output improves as the conversation develops.
The underlying requirement is domain knowledge. You must know your constraints well enough to state them clearly.
AI as a Knowledge Resource
The most immediate value an LLM delivers to a weld engineer, or lead technician, is compression of research time.
Interpreting AWS, ASME, or ISO code language, cross-referencing material compatibility, and working through preheat or postweld heat treatment parameters are tasks that have always required digging through standards, textbooks, and reference documents. AI doesn’t replace those sources, but it can quickly surface relevant information and translate dense, technical language into plain English that a technician can act on.
It can also serve as a starting point for defect diagnosis. Describe a porosity pattern, a cracking location, or an irregular bead profile, and an LLM can help you work through probable causes and contributing factors. It won’t replace the experienced eye standing at the part, but it can organize the diagnostic conversation and surface considerations worth checking.
The critical qualifier in every one of these applications is evaluation. AI output requires verification by someone who knows enough to recognize a wrong answer. An experienced engineer asking about preheat requirements for a specific base metal combination can assess whether the response is in the right territory. Someone without that background can’t, and that’s where AI stops being a resource and starts being a liability. Having experience isn’t optional. It is required to make the tool work.
AI as a Building Tool
Weld engineers and leads have always had ideas for tools they couldn’t build, such as tracking logs, inspection travelers, training assessments, and data dashboards. Building those tools required experience and financial input from people with different skill sets.
LLMs have changed that in a meaningful way. If you can clearly describe what you want, including the inputs, the outputs, the logic, and how you want to interact with it, you can build functional spreadsheet tools and HTML-based applications without manually writing a line of code.
The operative word is clearly. This is where domain knowledge does the work again. A weld engineer who understands exactly what data they need to capture and why, what calculations matter, and what the output needs to communicate, will build a better tool than someone with a vague idea and an optimistic prompt. AI is a capable collaborator, but it builds what you describe, not what you meant to describe.
This is a limitation of AI. If you can’t read or modify the code yourself, you will need AI assistance every time something breaks or needs to change. That’s not a reason to avoid these tools, but it is an operational reality worth planning for.
AI as a Visualization Tool
AI image generation, particularly from models trained or refined on technical imagery, can produce defect identification visuals, procedure illustrations, and training materials at a quality level that simply hasn’t been accessible to shops before. If you didn’t have a graphics budget or a technical illustrator on staff, you had to make do.
Proficiency carries significant weight here. An AI-generated image of porosity or incomplete fusion is only as accurate as the person directing its creation. The person directing its creation needs to understand what they’re depicting accurately enough to recognize when the output is wrong.
A plausible but subtly incorrect training image is worse than no image at all because it teaches the wrong thing with visual authority. The expert eye guiding the prompt is what makes the output trustworthy, not the technology producing it.
The Ethics of AI in a Skilled Trade
Drafting procedure summaries, organizing data, building basic tracking tools, and generating training visuals were all considered cornerstone entry-level responsibilities. The math isn’t hard to do, and some organizations are already doing it. Why hire someone when AI can handle it?
Here is why AI should not replace trade professionals:
If you stop hiring at the bottom, you stop developing the expertise that makes AI useful at the top. The weld engineer who learned to evaluate an AI response about preheat parameters did so somewhere. They learned it as a technician making mistakes under supervision, as a junior engineer reading procedures and asking questions, and as someone who spent years developing the foundational knowledge that now enables them to use AI productively. For AI to be used effectively and accurately, we need to keep this learning pathway for future welding professionals.
Someone must know when AI is wrong. There is a competence floor argument that organizations should consider carefully. That judgment isn’t innate, and it’s not developed by reading AI outputs. It is developed through years of doing the work. Eliminate entry-level positions, and you eventually eliminate the human capacity to audit the tool you depend on.
AI didn’t make the decision to replace entry-level workers. Companies that focused on cutting costs with spreadsheets did. The resentment that working professionals feel toward AI is often legitimate frustration directed at the wrong target. The technology isn’t the problem; the decision making around it is.
AI should expand what your team can do. It should not determine how small your team needs to be.
The Tool and the Hand
AI does not create expertise; rather, it multiplies it. The professionals who engage with it thoughtfully will find capabilities that weren’t previously accessible to them: faster research, better tools, stronger training materials, and deeper diagnostic conversations. The profession stays healthy when those professionals are developed at every level, not substituted in the name of efficiency.
I have spent my career watching adults get better at their jobs. The best of them understands their tools well enough to know what they can’t do. That has not changed. It has only become more important. The tool is only as good as the hand holding it.
Reprinted with permission: The AWS Welding Journal