When Agents Become the Customer (Part 1)
Why conversational interfaces and simple affordances are set to close the gap between AI capability and consumer adoption
We’ve long thought about AI as a tale of two curves.
The first describes capability: what AI systems are technically capable of doing. That curve has gone nearly vertical. Models can reason, browse, write code, author legal defenses, use tools, make purchases, and increasingly execute complex work autonomously.
The second describes adoption: how much of that capability has actually made its way into the lives of normal people. In comparison to the capability curve, this curve is stunted. For engineers, researchers, and early adopters, AI already feels transformative. But spend time with almost anyone outside of that bubble and the gap is obvious. Most people still use AI as a better Google search box, if they use it all.
Pat Grady at Sequoia has a useful phrase for the distance between these two curves: the diffusion gap. Most folks refer to this diffusion gap in the context of enterprise, but the consumer diffusion gap may be even larger. And increasingly, one of the simplest ways to close it may be hiding in plain sight: texting.
You already know how to text. You already do it dozens of times a day, there is no blank-canvas anxiety (especially as agents become more proactive), no prompt-engineering learning curve, and no new behavior to learn. For a lot of ‘normies,’ the path to the first magical agent experience might therefore look much less like downloading increasingly sophisticated AI software and much more like adding a new contact.
The Wrapper is the Product
Most agents are commoditized. Claude Cowork, OpenClaw, and other ‘agents’ behave similarly. The real innovation here is UX.
The best historical consumer technologies frequently worked this way. Winning products did not necessarily introduce the underlying technological primitive. They made an existing primitive accessible enough that dramatically more people could use it. We wrote a bit about this here.
Today, Instinct, Muse, Dots, Fo, Tab, and more are embracing ease of use as the core product.
In consumer AI, we increasingly think affordance may matter as much, if not more, than intelligence. This invokes questions like:
Can I immediately understand what this thing does?
Does it live somewhere where I already spend time?
Does it understand enough about me that I don't have to continually explain myself?
Can it actually finish the task?
That last question matters. Consumer agents become much more interesting when we move from answers to outcomes. Magic emerges when AI becomes situated enough in your life to know what happened, what matters now, and what should happen next (and hopefully take the initiative). Ashwin Gopinath describes it as this state.
The Consumer Agent Becomes the Demand Layer
Agents get much more interesting once they start transacting. Today, companies spend enormous amounts of money ($295B to be exact) trying to influence consumer decisions: search ads, recommendations, influencer marketing, retargeting, loyalty programs, affiliate marketing, etc.
All of these systems assume the consumer is the decision-making interface. Increasingly, purchases are beginning with:
“Find me the best flight to LA next Wednesday.”
“Reorder groceries for the week.”
“Find a good couch that fits this wall and keep it under $4,000.”
The agent now sits between consumer intent and the market. Most goods and services – travel, food, banking, home services, healthcare, shopping, logistics – could be purchased through a personal agent. The agent becomes the demand layer of the economy and brands stop competing only for human attention and increasingly for an agent’s recommendation and trust.
The consumer agent may become the new front door to demand. But a front door is only useful if there’s something on the other side. If agents are going to book our travel, buy our groceries, schedule our appointments, and negotiate on our behalf, they will need to interact with a much broader ecosystem of specialized agents, merchants, and services built to serve them. This raises a new set of questions. Do we each have one agent or fifty? How do agents discover one another? And what happens when the customer showing up to a business is no longer a person, but software?
