Are engineers at risk from AI? A new study suggests it’s complicated

The study, carried out by Microsoft Research, is based on an analysis of nine months of anonymised interactions that employees had with Microsoft Copilot AI in 2024. It is one of the first extensive investigations of the actual interactions between people and AI technology. This study captures the reality of human-AI interaction across hundreds of jobs, which is in contrast to theoretical projections.

By dividing each interaction into two parts—the user’s goal and the AI’s action—the researchers were able to map what people meant to do and what the AI really performed in response.

The researchers created an “AI applicability score” in order to convert this into an impact on employment. This metric quantifies the degree of overlap between the tasks that comprise a certain profession and AI’s demonstrated capabilities. Simply put, it poses the question: How much of the labour performed by employees in a given job could be theoretically completed by today’s AI tools?

A job’s score would be greater if AI were highly competent at an activity that is essential to that occupation. The score would be lower if AI could only handle specialised or small portions of a work. The study quantifies this by ranking professions according to how much generative AI support they receive.

Which jobs are the most exposed to AI risk?

The findings clearly show a difference between manual and knowledge labour. Information-centric, “cushy desk jobs,” which require work involving language, data, or human communication, were the occupations with the highest AI applicability scores. However, jobs that need in-person labour or physical skills—things that current generative AI cannot do—have the lowest scores.

When ranked by AI applicability, these occupations sit at the very top and bottom of the list.

Top 10 most AI-exposed occupations

1. Interpreters and Translators: language and information translation tasks
2. Social Scientists: researching and writing about historical information
3. Transportation Attendants: providing information and assistance to travelers
4. Sales Representatives (Services): communicating information to clients
5. Writers and Authors: producing written content
6. Customer Service Representatives: answering inquiries, providing information
7. CNC Tool Programmers: programming machine instructions, a highly codifiable task
8. Telephone Operators: providing directory or information assistance
9. Ticket Agents and Travel Clerks: information lookup and booking tasks
10. Broadcast Announcers and Radio DJs: writing and delivering scripted content

Top 10 least AI-exposed occupations

1. Logging Equipment Operators: operating heavy machinery in forests
2. Motorboat Operators: piloting boats, a hands-on and physical task
3. Orderlies: hospital support work, patient assistance
4. Floor Sanders and Finishers: skilled manual craftsmanship
5. Pile Driver Operators: construction equipment operation
6. Rail-Track Laying & Maintenance Operators: railway construction work
7. Foundry Mold and Coremakers: metal casting, manual fabrication
8. Water Treatment Plant Operators: managing complex physical systems
9. Bridge and Lock Tenders: operating bridges/locks; on-site manual control
10. Dredge Operators: removing sediment; heavy equipment operation

With an applicability score of 0.49, which indicates that AI performs 49% of their tasks more effectively, interpreters and translators are the professions most affected by AI.

“The AI in language translation market size… will grow from $2.34 billion in 2024 to $2.94 billion in 2025 at a compound annual growth rate of 25.2%,” according to industry analysis. Yet rather than replacing human translators, the data reveals a more complex dynamic: over 70% of translation professionals now use AI tools to enhance their work, with AI handling routine translations while humans focus on cultural nuances and complex interpretation.

Behind translators, the occupations experiencing the highest AI impact include historians (0.48), passenger attendants (0.47), and sales representatives (0.46). The pattern that emerges challenges the narrative of AI primarily affecting blue-collar or routine work. Instead, knowledge workers and communication-intensive roles dominate the high-impact categories.

With a 0.44 applicability score, customer service agents—who employ around 2.9 million Americans—ranked sixth. The automation trend here is accelerating rapidly: Gartner predicts that chatbots will become the primary customer service channel for roughly a quarter of organizations by 2027, with AI already handling basic inquiries while routing complex issues to human agents.

However, the researchers caution that “applicability” does not equal immediate automation. Just because an AI can do parts of a job doesn’t mean an entire occupation can be handed to the machines. In fact, they explicitly note that their data “do not indicate that AI is performing all of the work activities of any one occupation.” This warning is particularly pertinent to engineering work, which frequently blends jobs that are easily automated with those that are still exclusively human.

Zooming in: How exposed are engineers?

Where do engineers, including software developers, mechanical, electrical, and civil engineers, and their kin, fit into this spectrum? It’s interesting to note that while engineers do not rank among the top ten “at-risk” occupations in the study, they do not rank among the relatively safe occupations either. Engineering, in many respects, lies somewhere in the middle. Engineers perform a combination of knowledge-based (calculations, design, analysis, documentation) and practical (prototyping, site supervision, hands-on issue resolution) roles, with generative AI being significantly more proficient than with the latter.

Occupation	AI Applicability Score	Coverage Rate	Task Completion Rate	Impact Scope	Employment
CNC Tool Programmers	0.44	0.90	0.87	0.53	28,030
Mathematicians	0.39	0.91	0.74	0.54	2,220
Technical Writers	0.38	0.83	0.82	0.54	47,970
Data Scientists	0.36	0.77	0.86	0.51	192,710
Web Developers	0.35	0.73	0.86	0.51	85,350
Statistical Assistants	0.36	0.85	0.84	0.49	7,200
Management Analysts	0.35	0.68	0.90	0.54	838,140
Market Research Analysts	0.35	0.71	0.90	0.52	846,370
Geographers	0.35	0.77	0.83	0.48	1,460
Economics Teachers	0.35	0.68	0.90	0.51	12,210

The conclusions are striking. The study found that computer and mathematical occupations emerged as the major occupational group with the highest AI applicability scores, with an overall score of 0.30. But drilling down to specific roles reveals even more dramatic differences.

CNC tool programmers topped the individual engineering rankings with an AI applicability score of 0.44, suggesting that nearly half of their core work activities could potentially be assisted or performed by AI systems. This finding challenges conventional wisdom—while much attention focuses on software developers, it’s actually the engineers programming computer-controlled manufacturing equipment who show the highest AI adoption potential.

Data scientists, despite working directly with data and algorithms, scored 0.36—high, but not the highest. This suggests that even roles we consider “AI-native” face limitations in how AI can assist with complex analytical tasks. The study notes that data analysis activities received some of the worst user feedback, indicating that current AI systems struggle with sophisticated analytical work.

Web developers and technical writers scored 0.35 and 0.38, respectively, reflecting AI’s particular strength in code generation and content creation—areas where engineers reported high satisfaction and task completion rates.

However, the AI applicability score is more modest for traditional engineering fields, including mechanical, civil, electrical, and aerospace engineering. The study gave the broad category of “Architecture and Engineering occupations” a mid-range score (~0.49 on a 0-to-1 scale), which is about half as exposed as the top-ranked computer-focused jobs. These engineers undoubtedly use software and deal with data, but a significant amount of their work involves the physical world, an area in which current AI is limited.

In the Windows Central analysis of the study, the author noted that jobs requiring “a physical human touch” — citing “builders, roofers, engineers, and surgeons” — seem safe from AI for now. While “engineers” as a general category is too broad to declare entirely safe, the point stands: the more an engineering role leans into physical or highly complex real-world tasks, the less current AI is able to replicate it.

In conclusion, software engineers and other purely digital engineers seem to be highly exposed to AI, and it seems likely that generative AI will have the biggest impact on their processes. The situation faced by mechanical, civil, and electrical engineers are mixed: while AI can complement many aspects of their work, particularly those involving computing, documentation, or simple design, other aspects of their jobs are protected from automation because of their intricate and physical nature. Consequently, while engineers are not at the top of the “AI risk” rankings, they are also not at the bottom. They are among the occupations that AI will significantly enhance, if not completely replace.

Automation versus augmentation

Perhaps the study’s most important insight challenges the binary thinking that dominates AI discourse. Rather than simply asking whether AI will “replace” engineers, the researchers distinguished between two types of AI impact: tasks where AI assists human engineers versus tasks where AI performs work that might otherwise be done by a third party.

This distinction reveals a fascinating asymmetry. In 40% of the conversations analyzed, the work activities that users sought help with were completely different from the activities the AI actually performed. For example, when an engineer asks AI to help troubleshoot a technical problem (user goal), the AI might respond by providing technical support and explanation (AI action)—two different types of work activities that impact different parts of the labor market.

The implications are profound. AI isn’t simply automating existing engineering work—it’s creating new forms of human-AI collaboration that didn’t exist before. Engineers are using AI to access capabilities traditionally provided by other roles (like technical writing, research assistance, or customer support), while AI helps them execute their core engineering tasks more efficiently.

So, what are engineers actually using AI for?

The study’s examination of work activities captures how engineers are adopting AI tools into their everyday professional routines. The most common user goals—what engineers request from AI—can be classified into three use cases.

Information gathering is AI’s most common use case for engineers. AI is being used to research technical specifications, gather data about products and services, and keep engineers informed about new developments. This reflects AI’s strength as an advanced search and synthesis tool that can distill large volumes of technical information.

Writing and content creation are another critical use case. Engineers are using AI to edit technical documents, prepare design materials, and develop informational content. These tasks received the most positive evaluations, which suggests AI is particularly effective in supporting technical communication.

Problem-solving and explanation round out the top uses, with engineers seeking AI assistance to understand regulations, troubleshoot technical issues, and explain complex concepts.

The absence of core design and analysis tasks within the high-usage categories is particularly striking. Engineers do make use of AI to retrieve relevant information for their designs and to provide explanatory breakdowns of the work executed, but they do not seem to employ AI tools for fundamental engineering computations or critical design decisions.

The skills that still matter

While AI transforms many aspects of engineering work, the Microsoft study and supporting research consistently point to areas where human expertise remains irreplaceable. Data analysis activities—core to many engineering roles—received some of the lowest user satisfaction scores in the study. Engineers reported particular frustration with AI’s ability to “process data, calculate financial data, and analyze scientific data.”

Similarly, visual design and creative tasks showed poor AI performance. Activities like “creating visual designs,” “arranging displays,” and “developing models of systems” consistently ranked at the bottom of user feedback scores. This suggests that spatial reasoning, aesthetic judgment, and complex visual problem-solving remain distinctly human capabilities.

The study also revealed that AI performed worse when trying to directly provide support or advice compared to helping engineers provide that support themselves. This finding highlights the importance of human judgment in stakeholder interactions, ethical decision-making, and contextual problem-solving.

Pattern Category	Key Examples	Interpretation
Most Assisted by AI (vs Performed)	Purchase goods (118.4x), Execute transactions (58.8x)	AI helps with physical/transactional tasks humans must do
Most Performed by AI (vs Assisted)	Train procedures (17.9x), Train equipment use (16.0x)	AI naturally takes on teaching/advisory roles
Highest User Satisfaction	Research healthcare, Edit documents, Purchase goods	AI excels at research, writing, and evaluation tasks
Lowest User Satisfaction	Visual designs, Scientific analysis, Calculate data	AI struggles with creative and complex analytical work
Asymmetric Activities	Research (user) vs Provide info (AI), Troubleshoot (user) vs Support (AI)	AI provides complementary rather than identical services
Conversation Overlap	40% of conversations have completely different user goals vs AI actions	Human-AI collaboration is highly asymmetric and complementary

Even in the use cases discussed in the previous section, AI is automating slices of the work (e.g. drafting a CAD outline or solving a formula) while assisting the human in achieving the larger goal (creating a viable product design). The Microsoft study’s methodology — categorizing user intentions versus AI actions — underscores this dynamic: users enlist AI to play specific roles in service of their projects. Sometimes the AI is the “writer”, sometimes the “tutor”, sometimes the “researcher” on the team. But the user is orchestrating these roles.

The Microsoft study’s methodology, which categorises user intentions versus AI actions, highlights this dynamic: users enlist AI to play specific roles in the service of their projects. Sometimes the AI is the “writer,” sometimes the “tutor,” and sometimes the “researcher” on the team, but the user is orchestrating these roles. Even in the use cases discussed above, AI is automating slices of the work (e.g., drafting a CAD outline or solving a formula) while helping the human achieve the larger goal (creating a viable product design).

From a risk perspective, this means that rather than being completely replaced by AI, engineers who use AI as a tool will probably perform better than those who don’t. Naturally, the downside is that companies may feel they need fewer engineers overall if one engineer with AI can perform the tasks of two engineers without AI. This efficiency advantage could result in job layoffs in a purely economic sense.

So, will AI take my job?

The impact of these AI adoption patterns extends far beyond individual productivity gains. The engineering profession is experiencing what researchers describe as a “jagged technological frontier”—AI excels at some tasks while failing completely at others, creating unpredictable patterns of disruption across different specialties.

Manufacturing engineering is seeing a particularly rapid transformation. AI-powered systems are revolutionizing everything from predictive maintenance to quality control. Computer vision systems can now detect defects smaller than a human hair. Generative design algorithms are helping engineers explore thousands of design alternatives automatically.

Software engineering remains the most visible AI adoption case, with tools like GitHub Copilot becoming standard in many development teams. However, adoption rates remain modest—even after a year, only about 60% of developers consistently use AI tools. This suggests significant room for growth as tools improve and engineers become more comfortable with AI assistance.

Civil and environmental engineering show lower AI applicability scores in the Microsoft study, reflecting the physical nature of much of this work. However, AI is making inroads through applications like smart infrastructure monitoring, automated design optimization, and predictive modeling for environmental systems.

The big question hovering over all of this: if AI can do some of an engineer’s work, does that mean companies will need fewer engineers? Or will engineers simply become more productive, tackling more ambitious projects with AI’s help?

The Microsoft study notably did not attempt to predict layoffs or wage effects. The authors explicitly warn against assuming that high AI applicability leads directly to job losses.

Despite dramatic predictions about AI-driven job displacement, current employment data for engineers tells a different story. According to the Bureau of Labor Statistics, overall employment in architecture and engineering occupations is projected to grow faster than the average for all occupations from 2023 to 2033.

Specific engineering fields are seeing even stronger growth. Data scientists face particularly robust demand, with employment projected to grow 36% from 2023 to 2033—far outpacing average job growth. Mechanical engineers are expected to see 11% growth, with about 19,800 annual openings.

This employment growth occurs alongside increasing AI adoption, suggesting that AI is augmenting rather than replacing engineering capabilities in most cases. The World Economic Forum projects that AI and data processing trends will create 11 million new jobs by 2030 while displacing about 9 million, yielding a net gain of 2 million positions globally.

The bottom line: What can engineers proactively do?

For individual engineers, the Microsoft study’s findings suggest several strategic imperatives. First, focus on developing skills that complement rather than compete with AI capabilities. The study shows AI excels at information processing and routine analysis but struggles with contextual judgment and creative problem-solving.

Second, actively experiment with AI tools in low-stakes environments. Engineers who develop fluency with AI assistants today will have significant advantages as these tools become more powerful and prevalent.

Third, cultivate uniquely human capabilities that the study shows AI cannot effectively replicate: stakeholder communication, ethical reasoning, interdisciplinary collaboration, and creative problem-solving.

For engineering organizations, the study suggests the need for thoughtful AI integration strategies that leverage AI strengths while preserving human expertise in areas where AI remains weak. This includes investing in training programs, updating work processes, and developing new quality assurance methods for AI-assisted work.

For engineering educators, the findings point to curriculum needs around AI literacy, human-AI collaboration, and the skills that remain distinctively human in an AI-augmented world.

The Microsoft study provides the clearest picture yet of how AI is actually transforming engineering work—not through wholesale automation, but through selective augmentation that varies dramatically across specialties and tasks. This isn’t a crisis to be managed but an opportunity to be seized. AI is proving to be a powerful amplifier of human engineering capability, enabling professionals to tackle larger problems, explore more design alternatives, and deliver higher-quality results. The engineers who master this collaboration will define the future of their profession.