Notes

When AI connected what experts could not

Maninder Singh Juneja opens by pointing to a simple but disorienting fact: over a billion people are now using AI across every generation, language, and medium, making it the fastest technology adoption in human history. He argues that something shifts the first time anyone uses AI to check a professional's advice, whether a doctor, lawyer, or salesperson, even if the person doesn't consciously register the shift. Customers are already doing this routinely.

To ground the idea, he shares a personal anecdote. His wife had meibomian gland dysfunction (MGD), a condition where tiny oil-secreting glands in the eyelids stop functioning properly. The diagnosis and medication were correct, but progress was slow, so the family checked the diagnosis against AI, which confirmed it was right. Weeks later, while shopping for cosmetics, she fed photos of her products into the same AI tool. Instead of shade advice, the AI flagged that one of her products, a waterproof eyeliner, was effectively "pouring wax into a drain" and blocking her pores.

Maninder draws out the structural lesson from this story:

"The ophthalmologist saw the eye, the cosmetic counter saw the product, AI saw the person. The AI did not know more about ophthalmology. It knew enough context to connect the two domains. No institution was designed to connect."

None of the individual experts were wrong, he stresses, but something permanently changed at the customer's end: belief in experts as a category came into question, and that questioning extends across financial services, legal advice, design, and beyond.

The two foundations of trust and the market for lemons

Maninder lays out a framework for why trust has worked the way it has. Trust rests on two foundations. The first is genuine expertise: the doctor knows medicine, the banker knows credit, the insurance agent knows policy structure. This foundation, he says, survives and is more valuable than ever.

The second foundation is the customer's practical inability to verify, not through ignorance but through lack of access and lack of time. Getting a second medical opinion means finding a specialist, booking an appointment, and spending money and time. Comparing insurance policies across four companies means wading through different terms, exclusions, and inclusions. Comparing loan structures means parsing foreclosure charges, floating versus fixed rates, and layered fees. Historically, customers simply couldn't do this work, so they either trusted the expert or gave up and bought anyway.

This is where institutions like regulators, ombudsmen, and endorsements entered, not to give customers the truth directly, but to give them a reason to trust without needing to verify it themselves. Maninder references the underlying economic idea:

"George Akerlof named it in nineteen seventy, the market for lemons. When the buyer cannot verify quality, the market builds institutions to verify on her behalf. None of them gave the customer the truth, but they gave her a reason to trust without it."

AI has broken this second foundation. The customer can now do for herself what institutions once did on her behalf.

From customer to counterparty

Because the customer can now verify independently, Maninder argues her relationship to the institution has fundamentally changed. She is no longer arriving to buy a product; she is arriving to evaluate one. He frames this shift sharply:

"She's no longer coming to buy a product. She's coming to evaluate. She's no longer a customer, she's a counterparty now."

Effectively, she now arrives with an AI-powered analyst sitting beside her. This erodes the advantages institutions held for decades: knowing more than the customer, comparing options better, and framing the customer's choices for her. Since the cost of information has dropped to near zero and become instant, institutions are, for the first time, no longer the customer's primary source of intelligence.

Maninder distinguishes between how institutions and customers are each using AI. Institutions use it mainly to cut costs and increase efficiency, which is important for competitive survival. Customers use it to increase their own agency: the ability to act on knowledge they've just received, comparing and evaluating things that used to be too complex to untangle. Because customers can now act across the whole of their financial or personal lives at once, and institutions are typically organized around narrow verticals like products, risk, and service, there is a structural mismatch: institutions optimize fragments of a customer's life, while AI optimizes the whole of it.

Three fronts of exposure: products, frontline staff, and brand

Maninder identifies three areas where this shift lands hardest. First, products: many were designed with information and support systems built for a time before AI existed, and that era is over. Customers now arrive understanding fine print that used to be opaque, much like MGD becoming simply "blocked oil glands" once explained clearly. Products that survive this scrutiny will need a genuine moat, and manufacturers need to start designing for that reality.

Second, frontline staff. Since customers arrive prepared, what will matter from advisors is judgment: reading the human being in front of them, not just answering a scripted query. He is direct about what no longer adds value:

"The RM who recites the product terms to the counterparty is not adding value, he's consuming trust."

Third, brand. Maninder splits brand into two components. The "promise" part, things like complaint volumes and how issues were handled, is now auditable within roughly sixty seconds by asking an AI to summarize public information. The "signal" part, tied to identity, tribe, and aspiration, remains something AI cannot evaluate because it lives in the customer's mind. He argues institutions need to know which half of their brand they are actually built on, since the promise half is no longer protected by information scarcity.

How trust breaks quietly

Maninder describes two specific scenarios where trust ruptures, occurring at different points in time. The first happens live: a customer arrives already knowing the competition's products, the fine print, and the weakest clauses in what she's being sold. If the relationship manager doesn't realize she has come prepared and pushes a product dishonestly, trust breaks immediately.

The second scenario is delayed and quieter. At renewal time, a customer can simply ask AI to scan through all her emails and flag risks or issues across everything she has ever purchased. Once that surfaces, the customer becomes aware of problems she may not have noticed for months or years. This rarely becomes a formal complaint; instead, it shows up later as silence, specifically as a renewal that simply doesn't happen.

Being informed is not the same as being right

Maninder adds an important caveat to his own framework: verification does not mean correctness. AI substantially reduces information asymmetry, but it does not eliminate human bias or error. He draws a sharp distinction:

"AI makes the consumer undefeatable, not necessarily right. Those are not the same thing. An informed customer is defensive; a misinformed customer with AI is powerfully wrong. They arrive at the wrong conclusion with algorithmic certainty."

This means institutions' frontline response has to be prepared for multiple scenarios: customers who are correctly informed, and customers who are confidently wrong. Both must be handled carefully so that the underlying trust relationship isn't damaged either way.

Three questions every institution must ask itself

Maninder closes his framework by arguing that trust built on genuine expertise will survive scrutiny and grow stronger, while trust that depended on the customer's inability to verify will not. This means AI is not purely a technology decision; the winners will not just build better AI, they will design better businesses.

He proposes three questions that follow directly from this: What part of our value proposition exists only because the customer cannot verify it? If every customer walked in tomorrow with an AI advisor, what would still make them choose us? Are we using AI to reduce our own costs while our customers are using AI to increase their bargaining power?

He ends with a compact summary of the stakes:

"Every institution eventually gets redesigned around whatever has become abundant. Moats come from what is scarce. Verification is becoming abundant. Earned trust is now what's going to be scarce."

Tarun Chugh: from search engines to AI as the new trusted uncle

Tarun Chugh responds by tracing a generational shift in where trust comes from: from word of mouth (what an uncle, father, or friend told you), to Google over the last fifteen to twenty years, to AI now. He frames Google's earlier role vividly:

"Google became the most trustworthy uncle you had, or aunt."

He shares his own jolt into this reality: a large distributor, sitting in front of him, searched Tarun's own product on an AI engine, which surfaced information the company itself would have been "shocked" to see in the market. This incident revealed how much can go unnoticed by an organization until tested by a customer's own AI query, and it prompted real changes inside his company.

Tarun highlights why people trust AI more readily than older tools: interactions are conversational rather than jargon-heavy, and the AI retains context about the user over time. He describes seeing younger users share deeply personal queries, from relationship advice to interview preparation, with AI tools, treating them almost like a confidant who "knows their context" better than an older relative might.

His first instinct on discovering the gap in his own product's visibility was to invest in generative engine optimization (GEO), the AI-era analog of SEO. But reading Maninder's article pushed him further: companies cannot leave loose ends untied across the internet and social media, must stay agile in responding to what surfaces, and above all must stay transparent, since AI engines cross-reference many sources and will surface inconsistencies rather than reward gaming the system.

Building trust for advisors, personalization, and the appless world

Tarun connects these ideas specifically to life insurance, which he calls possibly the most complex product in financial services. He argues that the more complex a product, the more exposed it is to AI-driven scrutiny, and also the more potential upside there is if a company uses AI well, for instance in training advisors and preparing them for tougher customer queries.

Because life insurance is typically sold over four or five meetings rather than a single transaction, customers have repeated opportunities to verify what they've been told, making advisor preparation essential. Tarun notes his company has started using AI specifically for this kind of advisor coaching.

He also raises a structural point about how trust has traditionally been built in his industry. Companies have historically operated as "absentee landlords," with the brand manufacturing the product while the individual advisor carries the actual trust relationship with the customer. In the AI era, he argues, the individual advisor's own credibility may now need to be built and endorsed directly, not just the parent brand's, which represents new investment his company hadn't previously considered.

On personalization, Tarun sees a coming shift where AI engines themselves may start recommending financial products directly to customers, the way search engines already nudge behavior through suggested content. He raises an open question about when personalization tips into marketing, and stresses companies need to be positioned well before that shift fully arrives.

Finally, he introduces a concept not covered in the original article: MCP, or model context protocol, which he describes as an emerging shift toward an "appless" world where users interact through AI directly answering questions rather than through dedicated apps, requiring significant personalization work behind the scenes to make this possible.

Amrita Agarwal: questioning whether AI really changes the verification game

Amrita Agarwal opens by summarizing what she took from the article: verification has gone from being hard and costly to easy and cheap, enabling real-time claims verification and fine-print analysis; brand promises are becoming more auditable, though this doesn't replace the tribal or emotional connection built through human interaction; customers need to be equipped to ask better questions or the benefits of AI won't materialize; and the article calls for treating AI as a business-model-level strategic decision rather than a simple cost-saving tool.

She then offers a more skeptical layer of analysis. First, she notes AI could actually work against companies by inserting itself between the company and the customer, since customers now go to ChatGPT or Gemini rather than the company directly, creating new opacity about what's being said and reducing the company's own avenues for building trust.

Second, she questions the reliability of AI's answers themselves, arguing that quality depends heavily on what underlying data exists. Domains with strong statistical foundations, like clinical medicine backed by randomized controlled trials, likely produce more reliable AI outputs than domains built on anecdote or individual experience. So whether AI gives the customer a genuinely correct answer, she argues, is uncertain and varies by domain.

She also raises the possibility that companies will adapt by feeding AI more favorable, though not necessarily false, information, and offers an analogy to describe the resulting uncertainty:

"Women have makeup, and they use makeup to make themselves look nicer, and from a distance you cannot tell, they just look nicer. But perhaps now the distance has reduced, and you can tell that the makeup is there, and it may not look as nice. But the companies may evolve, and women may evolve and use even nicer makeup."

Her point is that it remains unclear whether this dynamic genuinely improves trust or simply shifts the contest toward more sophisticated information-packaging on both sides.

On brand-promise auditability specifically, she is doubtful that AI represents a true step change. She notes that open information, auditors, regulators, and not-for-profit product testers already existed before AI, and questions whether AI-generated summaries of complaints or claims ratios carry real statistical validity, especially if underlying online information can itself be manipulated or seeded by competitors. She frames the open question directly:

"Is AI really a step change, or is it just an aggregation of a lot of transparency which existed?"

She also flags emerging research suggesting AI can simulate tribal or emotional affiliation, meaning the "signal" side of brand, which Maninder framed as safe from AI, may not be entirely immune. She suggests companies may need to think about building a coherent architecture of empathy and connection through AI itself, particularly for younger customer cohorts.

On the idea of equipping customers with better questions, she agrees this will be a differentiator, but notes it won't only come from not-for-profits and research groups, as suggested in the article. For-profit companies will also build libraries, skills, and plugins on top of AI models to make them more useful for specific industries and use cases, since AI providers themselves have an incentive to make their tools more relevant.

Labor, empathy, and the competitive pathway to change

Amrita turns to labor and productivity, noting existing literature shows AI improves productivity most clearly for low-skilled or novice workers on well-defined tasks like sales, writing, coding, and specific research. But she argues it remains an open question whether this holds for more complex tasks.

Statistic: she cites a study of entrepreneurs in Kenya that found no net effect of AI usage on outcomes overall, with a striking split underneath that headline number, higher-skilled entrepreneurs improved their results using AI, while lower-skilled entrepreneurs did not.

She connects this to a real example from a technology company she supports, where design, product, engineering, and data roles are collapsing into each other under AI's influence. She wonders whether only highly skilled workers who can integrate multiple disciplines will be able to leverage AI effectively for complex work, while lower-skilled workers struggle to keep pace, a pattern she notes is also being reported across Europe and the US.

On the assumption that AI will push human labor toward empathy and relationship-building, she offers a nuanced counterpoint. She cites a blinded study where patients rated AI responses as more empathetic than doctors' responses when they didn't know which side was which, but notes other studies found that once people knew they were talking to AI, they discounted that same empathy. Her conclusion is that this remains an unsettled area, and it need not be true that only humans can effectively support human emotional needs.

Finally, she broadens the discussion beyond customer-driven pressure. She argues that pressure for institutions to change flows not only from customers becoming more informed, but also, and perhaps faster, from competition, since AI makes it easier for rivals to innovate quickly. She illustrates the limits of customer-driven pressure with a thought experiment:

"Let's take an example of a monopolistic business where there is no competition. Even if the customer knows better, what's going to happen? Nothing. That's the only option."

Her view is that in low-competition sectors with limited value creation, customers may lack real alternatives regardless of how informed they are, whereas in highly competitive, high-value sectors, AI-enabled competitive innovation may force business model change faster than customer pressure alone. She closes by agreeing fully with the article's broader framing that AI adoption should be treated as a strategic lens for business model change, not simply a productivity or cost-management tool.

Supplementary Resources

• Trust in the Era of the AI-Informed Customer by Maninder Singh Juneja (Article)
• The Market for "Lemons": Quality Uncertainty and the Market Mechanism by George A. Akerlof (Paper)
• Generative AI at Work by Erik Brynjolfsson, Danielle Li and Lindsey Raymond (Paper)
• ChatGPT sets record for fastest-growing user base -- analyst note by Krystal Hu (Article)
• Annual Report 2025-26 by Reserve Bank of India (Report)

The complete transcript file is available to download below.