Generative AI’s ability to create or improve products and services means it has to potential to commoditize some businesses, while greatly enhancing the competitive advantage of others. In what follows I will discuss the characteristics that make products or services susceptible to disruption by generative AI, and those that make them more resilient. To illustrate, I will then compare Chegg and Duolingo, two businesses which have been impacted very differently by the emergence of generative AI.

Products and services at risk of commoditization

Generative AI is well-suited for tasks that rely on vast datasets, pattern recognition, and the automation of routine processes. Consequently, sectors relying on such tasks are highly vulnerable to AI disruption. The most immediate examples include the following:

- Content Generation: Tasks such as writing articles, creating social media posts, or producing videos are easily automated by AI tools like ChatGPT, which can generate coherent and persuasive content in seconds. Businesses providing content as a primary service may struggle to retain a competitive edge as generative AI becomes increasingly accessible.

- Code Generation: Platforms like GitHub Copilot and OpenAI’s Codex are already capable of generating code based on user prompts. Basic programming and debugging tasks are now more accessible to non-experts, making certain coding roles more susceptible to replacement.

- Product Design: In creative sectors like gaming, clothing, and furniture, generative AI can design visual assets, gameplay ideas, and product prototypes. While human designers bring unique creativity, AI’s speed and cost efficiency could pressure companies to adopt AI for cost-effective initial designs.

- Advertising and Sales Campaigns: AI tools that generate advertisements, sales copy, or branding materials are increasingly effective at analyzing target audiences and creating tailored content. This reduces the need for human copywriters in some marketing functions.

- Customer Service: AI-powered chatbots can now handle most customer inquiries without human intervention. While some complex issues still require human assistance, AI has significantly reduced demand for human-led customer service.

- Education and Tutoring: Generative AI’s ability to answer questions, provide learning pathways, and tailor responses based on user needs has placed traditional educational services at risk. Tutoring platforms offering basic subject help could lose market share as students turn to generative AI for instant assistance.

Obviously, not all human workers in these industries will be replaced, but the scope and demand for human involvement in many of the relevant tasks will diminish as AI-generated alternatives become more affordable.

Factors that make products resilient to AI disruption

While the list of vulnerable sectors above provides a good start, it is difficult to extrapolate from those examples the fundamental characteristics that make products more or less likely to be disrupted by generative AI. This is why I found it useful to follow Charlie Munger’s advice of inverting problems, and ask the opposite question: what intrinsic aspects of a product are harder to replace with AI and therefore make the product more resilient to disruption by (generative) AI?

What follows is not an exhaustive list, but it arguably contains the most important factors that make products resistant to AI disruption. This framing also has the advantage of pointing out what businesses can and should do in order to improve their resilience to AI disruption.

Physical interactions

Products and services requiring physical presence or real-world interactions are naturally resistant to disruption by AI, which remains constrained to digital realms.

For example, spa treatments, haircuts, massage therapy, and products requiring sensory experiences (e.g., perfume, high-quality textiles, travel) cannot be digitally replaced. Industries such as construction and mining, which depend on physical labor and interaction with real-world materials, are also unlikely to see complete AI replacement.

AI can certainly assist in decision-making and improve worker productivity in these fields, but the fundamental physical components ensure continued demand for human expertise.

Network effects

Products where a large part of the value resides in network effects stemming from social interactions, community or user-generated content are difficult for AI to replace.

For example, platforms like Reddit, LinkedIn, and X rely on unique contributions from a diverse user base, which AI cannot replicate. An AI-based “social” network like https://socialai.co/, where each user gets customized AI followers, is obviously a very poor substitute for human social networking.

Similarly, online games like Fortnite and Roblox are resilient to AI displacement not so much because of their actual game content, but because they involve player-to-player interactions and entertainment.

Human connection and empathy

Services requiring empathy, cultural sensitivity, and interpersonal connections remain less vulnerable to AI disruption. AI may offer personalized content but cannot replicate the nuanced human touch that builds trust and engagement.

For example, therapy, personal training, and live teaching rely on the nuances of human communication. Generative AI may be able to generate exercises or provide advice, but it lacks the ability to form genuine emotional connections with clients.

Creativity and cultural sensitivity

Products that are highly creative or require a deep understanding of cultural trends are less likely to be replaced by AI, which tends to operate on learned patterns rather than original insights.

For example, art, fashion design, and film production are industries where cultural sensitivity, originality, and responsiveness to emotional or social cues are critical. Generative AI can produce impressive creative outputs but lacks true innovation and a deep understanding of cultural trends.

Regulatory and safety concerns

Industries governed by strict regulations, high accountability, and ethical concerns are naturally resistant to AI automation, given the complexity of compliance.

For example, healthcare, finance, and legal services require rigorous standards and often rely on trust. AI alone cannot fully satisfy the need for accountability and regulatory compliance in these fields.

These industries rely on trusted relationships, professional accountability, and regulatory frameworks that are challenging for AI to navigate independently.

Chegg vs. Duolingo

To illustrate some of the points above, it is useful to compare two purely digital businesses, Chegg and Duolingo, that face the threat of disruption by AI.

Chegg provides students with step-by-step textbook solutions, writing assistance, practice exams and live Q&A with online tutors. Duolingo offers language lessons structured as interactive and gamified activities (streaks, leaderboards), and also includes virtual events.

In principle, generative AI can be quite disruptive for both Chegg and Duolingo. Rather than signing up for Chegg, students can simply ask LLMs like ChatGPT or Claude for help with their homework. And rather than using Duolingo, language learners can ask LLMs to create customized language exercises, simulate real-world conversations, and adapt to a learner's unique pace and needs.

And both companies seem to have embraced -- or at least tried to embrace -- generative AI. In early 2023 Chegg redesigned its services around generative AI tools, which are offering its student subscribers personalized learning solutions. Similarly, in March 2023, Duolingo launched "Duolingo Max," a premium subscription tier powered by OpenAI's GPT-4. This tier introduced features like "Explain My Answer" and "Roleplay," offering users in-depth explanations and interactive conversation practice.

Despite these similarities, the rise of generative AI over the past two years (ChatGPT was launched in November 2022) has impacted the two businesses in very different ways, as reflected in the evolution of their respective stock prices. As of November 8, 2024, Chegg Inc. (CHGG) is trading at approximately $1.70 per share, down from around $28 per share two years ago. This represents a decline of about 94 percent over the past two years. This stands in sharp contrast to Duolingo. As of November 8, 2024, Duolingo Inc. (DUOL) is trading at approximately $320 per share, up from $70 per share two years ago. This represents an increase of about 350 percent over the past two years.

So how can we make sense of this huge difference in the eyes of investors? Is it mainly that Chegg’s offerings are inherently more exposed to commoditization by AI than Duolingo’s or is it because Chegg was slower and less effective in its efforts to adapt?

The answer is: a combination of both. Indeed, one could argue that Chegg’s product is more transactional and straightforward (asking for help with specific homework issues), which therefore means it’s easier to replace by AI. By contrast, creating an engaging language learning plan is more complicated and therefore allows more scope for human creativity and ingenuity, which would make it less vulnerable to AI.

That being said, it is also quite apparent that Duolingo’s product strategy has been much better suited to deal with disruption by generative AI than Chegg’s. Indeed, a big driver of Duolingo’s success are its social features, such as leaderboards, streaks, and peer competition, which create a sense of community and engagement that goes beyond the core language-learning content. In other words, Duolingo has created community-based network effects that make the service more appealing than solo learning with an AI-powered system. By contrast, Chegg has not built any such community-based network effects, which has left its service a lot more vulnerable to replacement by AI.

Concluding thoughts

The discussion of factors that can render products resilient to AI disruption and the comparison between Duolingo and Chegg provide some valuable lessons for all businesses. First, every business should add or double-down on features that are hardest for AI to disrupt, such as for example enhancing physical, social, or empathetic elements. Second, rather than viewing AI as a threat, businesses should embrace AI where it is clearly better than humans (e.g., repetitive and data-intensive tasks) and leverage it to create unique data feedback loops that continuously improve products. This approach positions AI as a tool to boost competitive advantage rather than merely a cost-cutting substitute.

The same principles apply to individual workers. Instead of waiting in fear of being replaced by AI or opposing its adoption in one’s organization or industry, better to treat it as an opportunity. Every worker should reflect on which parts of one’s job are best outsourced to AI and which are worth developing deep expertise in -- where human judgment and intuition will continue to be necessary.

Andrei Hagiu

Andrei Hagiu is an associate professor of information systems at Boston University. The views expressed here are the writer’s own. — Ed.