The AI Paradox: Spectacular Failures Alongside Revolutionary Successes
Two recent developments in the artificial intelligence landscape reveal a fundamental truth about where this technology stands today. These contrasting narratives highlight that AI's impact across different sectors is far from uniform, with some areas experiencing transformative breakthroughs while others face significant setbacks.
Salesforce's AI Customer Service Experiment Ends in Retreat
The first story involves Salesforce, the enterprise software giant that aggressively pursued AI implementation for customer service operations. CEO Marc Benioff proudly announced that AI deployment had enabled the company to dramatically reduce support staff from approximately 9,000 employees to roughly 5,000. This move was presented as a triumph of technological efficiency and cost reduction.
However, by late 2025, reality had intervened with sobering force. Reports emerged indicating that Salesforce was withdrawing from its AI customer service initiative due to widespread system failures. The AI agents demonstrated several critical shortcomings: they confidently provided incorrect answers to customer queries, dropped instructions when presented with more than eight sequential steps, and lost focus entirely when users asked unexpected or complex questions.
Customers reported that interacting with AI support systems actually took longer than using traditional search functions. The company is now retreating to rigid, rule-based scripting systems—essentially admitting they were, in their own words, "more confident" than the technology warranted. This represents a significant setback for AI implementation in customer-facing business functions.
The Coding Revolution: AI Transforms Software Development
The second story represents a complete zeitgeist shift in how we perceive AI's capabilities. Over recent months, the conversation around AI and software development has transformed dramatically. Senior developers who were skeptical just six months ago—professionals who actually write code for a living—are now declaring that the age of human beings writing code is ending.
This transformation isn't projected for some distant future but is happening imminently. Entire software features are being shipped by AI systems with minimal human intervention. The productivity gains are no longer incremental improvements but represent structural changes to how software is developed. This revolution suggests AI has found particularly fertile ground in the coding domain.
Resolving the Paradox: Understanding AI's Divergent Paths
How can both realities coexist? How can AI fail comprehensively in seemingly straightforward customer service applications while simultaneously revolutionizing the far more complex field of software development?
The answer lies in how we've been conceptualizing artificial intelligence. We've treated AI as a single phenomenon that will sweep through the economy at roughly the same pace across all sectors. However, AI in business isn't one unified story but many parallel narratives moving at wildly different speeds. This distinction has almost nothing to do with how intelligent the AI systems are.
Consider what makes coding such fertile ground for AI advancement:
- Code is formally structured and machine-verifiable—it either runs and passes tests or it doesn't
- The feedback loop is immediate and precise
- When AI makes a mistake, developers (or other AI agents) can notice, fix it, and move forward
- Errors remain private and are generally reversible
Now contrast this with customer service applications:
- Customers don't communicate in data schemas or formal structures
- Emotional context, sarcasm, and cultural nuances matter enormously
- A single wrong answer can escalate to social media outrage or regulatory complaints
- Failures become public and are often irreversible in their impact
The crucial difference isn't about intelligence levels but what might be called error economics. AI thrives in environments where mistakes are cheap, private, and correctable. It struggles where mistakes become expensive, public, and permanent.
The Executive Disconnect: Bajaj Finance's AI Implementation
We received a clear illustration of this executive-technology disconnect just recently. During Bajaj Finance's third-quarter earnings call, CEO Rajeev Jain announced that AI systems had listened to 2 crore customer calls and generated 100,000 new customer offers. He proudly projected, "We'll be able to listen to 100 million calls next year."
The social media response was predictably hilarious. As essentially the entire country except apparently Mr. Jain knows, Bajaj Finance's incessant spam calls have become the butt of countless jokes. Here was a CEO using sophisticated technology to optimize something customers actively despise. The machine learning worked perfectly from a technical standpoint, but the learning about actual customer preferences was completely absent.
Investment Implications and Future Trajectories
For investors and business leaders, these developments carry significant implications. When you hear "AI" attached to a business function, the critical question becomes: what happens when the system is wrong? If the answer involves customer dissatisfaction, regulatory scrutiny, or reputational damage, progress will inevitably be slower than vendor presentations claim.
If the answer is "someone notices and fixes it quietly," that represents an entirely different world of implementation possibilities. The story of AI in business isn't one of universal acceleration but of selective escape velocities. Coding has left the atmosphere and entered orbit, while customer service is still fighting gravity. Most other business functions lie somewhere between these extremes—often mistakenly assumed to be closer to the rocket than they actually are.
The many worlds of artificial intelligence are not converging toward a single destination. They're diverging along separate paths determined by their error economics. This divergence will ultimately determine which AI investments succeed spectacularly and which disappoint profoundly.
Dhirendra Kumar is Founder and CEO of Value Research with over three decades of experience in stocks, funds, and personal finance.
