AI Agents vs Attention Economy Which Model Survives When Nobody Clicks

Video: AI Agents Break Web Ad Model | Lightspeed’s Harsha, Hemant Mohapatra| Corner Office Conversation

Nov 2, 2026 · based on the channel ET Digital

AI agents are the better bet for founders than another ad-funded consumer app, because agents remove the clicks that advertising depends on while creating a paid, task-based relationship with the user. That is the core argument Lightspeed India partners Harsha Kumar and Hemanth Mohapatra make in a Corner Office Conversation with The Economic Times, available on the ET Digital video hub. If software shops, books and decides on a consumer's behalf, the attention economy loses its inventory — and businesses have to earn money from outcomes rather than impressions.

Why AI agents break the ad-based internet

The ad model assumes a human sees a page, scans it and clicks; an agent skips all three steps. When a personal assistant compares prices, filters options and completes a booking, no banner is viewed and no sponsored result is clicked. The value shifts from capturing attention to executing a task correctly and cheaply. That is a structural change, not a formatting tweak: the entire free-content bargain — users trade attention for access — depends on a human in the loop. Kumar and Mohapatra frame this as the central strategic question for marketplaces, publishers and consumer brands alike. Large marketplaces that today sell placement to merchants would have to sell something else: verified inventory data, fulfilment reliability, or API access that agents can query. Brands that cannot be parsed by an agent risk becoming invisible regardless of their ad budget.

What agent economics mean for pricing and adoption

Running an agent costs real money per task, which makes unit economics the first thing to test. Every search, comparison and transaction consumes inference, and the more autonomous the agent, the more tokens it burns before the user sees a result. Kumar and Mohapatra point out that consumers are price-sensitive, so a subscription that looks trivial to a venture-funded startup can be a hard sell to a household comparing it with a free app. Practical implications for builders:

1. Measure cost per completed task, not cost per query — a cheap agent that fails twice is more expensive than a costly one that succeeds once.
2. Cap autonomy where it does not pay: a booking that needs one confirmation may not justify a fully autonomous loop.
3. Price against the value of the outcome (time saved, price found, error avoided), not against the cost of a chatbot subscription.
4. Expect large model providers to keep pushing prices down, which compresses margins for thin wrapper products.

The investor view is that cost curves improve, but not fast enough to rescue products with no defensible workflow.

Why vertical AI is the stronger startup bet in India

Vertical AI — agents built for one industry's workflow — is where Indian founders have an edge over general-purpose assistants. Kumar and Mohapatra argue that domain depth, local data and integration with existing back-office systems matter more than raw model quality, which is increasingly rented from a handful of labs. A general assistant has to be good at everything; a vertical agent only has to be better than a human clerk at one repeatable job, such as claims processing, logistics reconciliation or compliance filing. That narrower scope shortens the path to revenue and makes the product harder to copy. It also fits the Indian market's structure: fragmented businesses, high volumes of manual paperwork and owners who pay for measurable savings rather than novelty.

Where enterprise agents and personal agents collide

Enterprise agents and personal agents are converging on the same transaction, and whoever owns the interface captures the margin. A company's procurement agent negotiating with a consumer's shopping agent is a machine-to-machine negotiation with no room for a display ad. Kumar and Mohapatra describe this as the next contested layer: enterprises want agents that enforce policy and preferred vendors, while individuals want agents that optimise for their own interest. The winner is likely whoever holds trust and identity — the party that can prove an agent is authorised to act and accountable when it errs. For B2B marketers, the practical consequence is that machine-readable product data, clean pricing APIs and reliable availability signals become the new SEO.

Is this the dot-com era again

The current agent wave resembles the dot-com period: genuine structural change wrapped in a lot of companies that will not survive it. Kumar and Mohapatra draw the parallel without predicting a crash date — the internet did rewire commerce, but most of the first cohort of startups still disappeared. The lesson for founders is that being early to a real shift is not the same as having a business; distribution, cost discipline and a workflow somebody pays for decide who lasts. For incumbents, the risk is slower: ad revenue does not vanish overnight, it erodes as agent-mediated traffic grows and becomes harder to monetise per impression.

Common doubts about the agent boom

Scepticism about agent monetisation is widespread, and much of it is fair. The strongest objections cluster around three points: few ordinary consumers use agents today, model companies are burning cash on training and inference, and differentiation between assistants is thin. Each has a counter-argument worth weighing. Adoption of agents is concentrated among power users and developers, but the same was true of e-commerce in its early years, and the interfaces are improving faster than the underlying economics. On costs, inference prices have fallen sharply and continue to fall, though not enough to make every use case viable. On moats, the defensible position is rarely the model — it is proprietary workflow data, integrations and trust. The honest conclusion is that the category is real and the casualty rate will be high, which is exactly the pattern the Lightspeed partners describe.

Sources and further viewing

• ET Digital TV — the full Corner Office Conversation with Harsha Kumar and Hemanth Mohapatra
• EconomicTimes.com — business and markets coverage
• ET app — news on mobile
• Facebook
• Twitter

The bottom line for founders and marketers

The agent transition rewards businesses that can be transacted with by software, not just seen by humans. Kumar and Mohapatra's message is that the ad-funded default is weakening while paid, outcome-based agent services are still expensive and unproven at scale. The pragmatic response is to build vertical agents with tight cost control, publish machine-readable product data, and treat agent traffic as a channel with its own economics. Watch the full discussion on the ET Digital channel for the partners' own framing of how they use AI assistants day to day, and visit ET Digital's video section for more conversations on where the internet economy goes next.

Source: AI Agents Break Web Ad Model | Lightspeed’s Harsha, Hemant Mohapatra| Corner Office Conversation · Markdown version