Use of AI for the Creation of FMEAs - Part 3 Potential Consequences
A blog post by Dr. Uwe-Klaus Jarosch, July 2026
What We’ve Covered So Far:
In Part 1, I described three FMEA tools and their capabilities for accessing AI models and deriving suggestions for the content of FMEA reports.
In Part 2, I walked through the 7 steps according to AIAG-VDA. What do the integrated AI applications—or those available independently of any tool—contribute in terms of content to each of these steps? The area of “FMEA maintenance” has not yet been addressed.
Here in Part 3, I’d like to share a few thoughts on the broader implications of the massive development of AI tools and how this might specifically affect FMEA.
New tools open up new possibilities.
But they also shift the cost-benefit ratio.
And in my view, powerful tools like AI assistants also change behavior
—perhaps not for everyone, but certainly for many.
Economical Aspects:
AI applications are being developed and operated worldwide with financial resources that are unprecedented in history. For 2026 alone, the American tech giants OpenAI, Anthropic, Google, Microsoft, Apple, and XAI have announced investments in new models and data centers totaling approximately 650 billion US$[1].
[1] The planned investments in 2026 are: OpenAI: 110 $Billion , Microsoft: 190 $Billion , Anthropic: 30 $Billion, Google (Alphabet): 185 $Billion, Meta: 135 $Billion, Apple: >40 $Billion , XAI: at the lower end 40 $Billion, Amazon : 200 $Billion ,
Not included are investments of NVIDIA and of many other smaller tech companies. Some of the invests are including other topics than development and computing centres. Furthermore, some of the investments are circular businesses.
These amounts of money can be interpreted in at least two ways:
- Investors have a vital interest in ensuring that AI is used and paid for in practically every area so they can recoup their investments. Every means is being employed to urge decision-makers to adopt AI. The promise is: a competitive advantage and increased profits for AI users.
- Competitors are vying for a unique market position—and, if possible, a monopoly. However, current approaches are very similar, leading experts to speak of “commoditization”[2] with intense price pressure. Internal pressure to justify the number of tokens[3] used (the end of “tokenmaxxing”[4] as a strategic shift in the AI market) also points to cost pressures and a focus on results. So far, sustainable economic benefits from AI use have been reported only in rare, isolated cases.
[2] commoditization ~ A product category offered by a group of competing providers facing price pressure in the market.
[3] Token = The AI’s unit of output. Tokens can be converted into computational effort, which in turn can be converted into costs for processors and energy.
[4] Over the past two years, “tokenmaxxing” has been promoted with the argument that “more tokens = more learning, more experimentation, more productive output.” Executives were evaluated based on how many tokens they used, in part to cut staff elsewhere and replace them with AI. However, tokens cost money. In early summer 2026, the question of cost versus benefit was therefore reevaluated, and a temporary halt was placed on the unwarranted consumption of AI tokens.
It remains to be seen whether this will lead to essential improvements in the models or even to unique selling points.
I do not expect applications such as FMEA support or, in the broadest sense, engineering capabilities to become the focus of development.
For applications in the technical field, it would be desirable to see enhanced “reasoning” that not only uses statistics as the basis for recommendations for the next tokens but can also increasingly incorporate scientific and technical rules.
Geoplitics
Technology companies classify artificial intelligence as THE economic driver of the future.
For nations—primarily the U.S. and China, but also for nations at war such as Israel, Russia, Ukraine, and Iran—AI is viewed as a factor in military power and is promoted and shielded under the guise of national security.
For economic powers such as Germany, the question of technological dependence and vulnerability to blackmail arises [5]. Both U.S. and Chinese policies in recent years leave little doubt that the focus is on “America First” and “China First” at any cost, with political vassals welcome (see Venezuela, Greenland, the Belt and Road Initiative, etc.) and old alliances no longer mattering.
[5] In a scenario „Europe2031.AI“ AI scientists from several european countries and the US have derived a potential development of AI driven geopolitics for the next 5 years. It looks on a Europe of 27 Nations between the interests of US and China.
Management Expectations on AI Technology
Tech trusts bundle immense power. At the very least, they have very powerful tools for shaping public opinion, whether through cross-ownership ties to the press and television media or through social media (X, Facebook, Instagram, LinkedIn). They have a great deal of money, which they use intensively not only for technology development but also for marketing (e.g., Super Bowl commercials, high-impact social media posts, podcasts, TV appearances, B2B sales activities).
The goal of this advertising is to generate hype around the new AI technology. Even though much of it is still under development: “Those who don’t use AI today will be left behind tomorrow.”
In contrast, there are also voices of caution, both from industry insiders [6] and from scientists in various disciplines [7].
[6] E.g. Mustafa Suleyman (founder oft he AI pioneer Deepmind, since 2024 he is CEO of the consumer AI of Microsoft– The Coming Wave (2023)
[7] E.g. Yuval Noah Harari, „AI has hacked the code of human civilzation“ , 30.6.2026, The 2026 Tanner Lecture on Human Values, Oxford, University UK, Youtube
AI companies themselves issue media-friendly warnings, accompanied by the assurance that their own ethical principles take this into account. Unfortunately, such principles are soon set aside due to financial considerations [8].
[8] A detailed look at this interplay between ethos, investors, and the public is explored in the Deutschlandfunk podcast series “The OpenAI Story.”
Furthermore, the hype is fueled by the fact that, with many technical advancements and achievements, it is reflexively assumed that they must be AI-driven[9].
[1] Example: Leap71’s success with computational engineering is continually touted as an AI method, even though—as the developers themselves have explicitly stated—it is purely algorithmic software, despite the fact that it has achieved results that were previously unimaginable.
Investors’ expectations on the stock markets are heavily influenced by expectations surrounding AI. Anthropic released Claude Code with the expectation that this model can write program code faster and better than programmers. Already, the stock prices of traditional software companies are plummeting[10].
[10] With the release of Anthropic’s Claude Code, the stock value of software companies has dropped by a total of approximately $300 billion, even though many of these companies are not even affected by Claude Code’s capabilities.
All of these developments are putting pressure on management teams across all industries to make decisions: Get to grips with AI before you fall behind, lag irretrievably, and lose your competitive edge. Serious consequences for the job market are predicted[11]—and are already being observed in some places[12].
[11] See the press coverage in The Guardian in response to statements made by Anthropic CEO Dario Amodei in January 2026
[12] Article in *Der Spiegel*, Issue 25/2026 dated June 15, 2026, on developments in the German labor market. It states that jobs in corporate administration, the creative and advertising industries, as well as entry-level positions—such as in software development—are already being significantly impacted today. For the labor market as a whole, no significant changes can yet be attributed solely to AI.
This question is also relevant to the use of AI for FMEA.
The expectations raised by the AI industry are enormous.
A meaningful assessment of how the costs of preparation, implementation, and operation compare to the resulting efficiency and/or quality gains is unclear and varies greatly from application to application.
For a labor-intensive development method like FMEA, management’s expectations are unlikely to be “better results with the same effort,” but rather range from “the same results with half the effort” to “everything automatic from now on.”
This poses a massive risk that the necessary resources will be withdrawn from the team to assist with development. Simply creating a document without the active involvement of the developers seems pointless to me.
Psychology
AI techniques will not only raise expectations among management but also change the ways in which those involved work.
In Part 2, I attempted to demonstrate that, at the beginning and end of the 7 steps, existing preliminary information can be processed and effectively integrated or derived using AI tools.
In the analyses—particularly regarding functions, errors, and the derivation of measures—some information is already known from templates, similar cases, or standards. But value creation in development arises from the creative work involved in addressing new situations.
Experienced, professional facilitators will draw on their many years of experience to allow AI to support them, but not to be dictated to by it. And they will focus their attention on the active participation of team members. Team-based work will only add value to the development process if everyone contributes their ideas, if everyone understands what the analysis reveals, and if they think creatively to find solutions.
Success-oriented, strong-willed individuals—such as experienced, professional FMEA facilitators—resist “cognitive nickel nurser”[13] and make full use of their cognitive abilities, their capacity for thought, and their ability to synthesize information.
For the average FMEA moderator, who occasionally struggles with this method in a project, there is a strong temptation to follow the AI’s appealing suggestions rather than critically questioning them. The FMEA is completed and linked formally and correctly. The team can follow along—provided they still consider their own participation necessary for this mode of operation.
In cognitive psychology, it is a well-known phenomenon that people take the easy way out. Mental effort is avoided as much as possible. This “cognitive aversion to effort” corresponds to the natural principle of “as much as necessary, as little as possible.”
With a powerful AI tool, it would likely be easy to delegate the mental work and engage in cognitive outsourcing [13]. The fact that these models are becoming more powerful has been evident for years and seems inevitable to me—given the amounts of money being invested.
[13] The terms “cognitive nickel nurser ” and “cognitive outsourcing” are taken from an article by Prof. Christian Stöcker on the impact of AI. Spiegel, June 21, 2026, “Why AI Has Changed the Way I Think About Digital Technology” (German)
The likely consequence could be that FMEA will continue to be performed formally for as long as standards or customers require it. The documents will be significantly better in terms of both form and content than “handmade” FMEAs are on average today.
But the method will no longer support developers in their thinking and thus will no longer offer any creative added value.
In my view, this would become a killer for the FMEA method.
Conclusions:
- With generative AI models since late 2022, machines have learned to process human language.
- Unprecedented amounts of money in industrial history are being poured into the development and availability of AI models.
- Expectations regarding the development of generative AI are being hyped worldwide and are being met, at least in part.
- Management is facing pressure to make decisions without reliable preliminary information or benchmarks. Lose competitiveness? Invest? Lay off staff? Enter into dependencies?
- Entry-level and creative jobs where routine tasks can be automated are already being affected.
- Risk management is a challenging aspect of development. Effective FMEA creation today relies on critically thinking FMEA moderators.
- The large pool of occasional users of FMEA method will likely accept AI assistance gratefully and uncritically.
- “Cognitive outsourcing” would reduce creativity in development and could spell the end of the FMEA method.
Stay curious
Yours
Uwe Jarosch