How a 20-minute email edit explains the AI scaling problem

This article summarizes an episode of Sequoia Capital’s video series featuring Ellias Torres, co-founder and CEO of Agency.

Elias Torres, co-founder and CEO of Agency / Photo credit: 20VC

Businesses are trapped between demanding AI’s benefits and rejecting its imperfections, warns Elias Torres, a former HubSpot executive and now co-founder and CEO of Agency. He cautions that this behavior prevents companies from achieving growth.

He also cautions that high revenue can often mask a failing product. Ultimately, his arguments on perfectionism and shallow products all point to one conclusion. To scale, companies must shift from AI assistants to fully autonomous systems.

The AI expectation mismatch

The current struggle with AI adoption is not about the technology itself but about the user’s mindset, argues Elias Torres. He sees a recurring pattern where the desire for new tools conflicts with a resistance to new ways of working.

Customers demand AI but reject its imperfections
Torres explains the paradox. “everybody says they want [AI] and when you come and bring it to them, they just reject it… people are demanding this level of perfection when they’re not even able to… use a chat prompt to ask a basic question.”

The influence of past technology
Torres observes that this demand for flawlessness comes from decades of using applications that were designed to be perfectly accurate and predictable. “[We are] used to pre-AI technology… that it was able to look up a string from a database and it was a perfect lookup and a perfect match. So I think they’re used to that level of perfection.”

How humans make AI inefficient

Torres argues that the biggest barrier to AI’s efficiency isn’t the technology, but a human tendency toward perfectionism. The core issue is an unproductive trade-off where users sacrifice AI’s primary benefits of speed and scale for tiny improvements in quality.

The 20-Minute Email Edit
Torres highlights this phenomenon with a specific, telling observation from watching user research recordings. He describes how users, given a nearly complete AI-generated email, get bogged down in minor details.

“I watch them on on recordings… ‘Wow, this person just spent 20 minutes and they really didn’t edit the email.’ Sometimes they’re just changing the salutation… And so I think that that’s fundamentally why businesses can’t scale today, because there’s humans in the middle.”

Accepting imperfection is necessary for speed
Torres advocates for a pragmatic approach. “My approach to AI is, if it’s doing 10% of a task that I wasn’t doing before, if it’s doing 20%, 30%, I’m extremely happy… But other people, if it’s not 100% perfect, I rather have none. And I think that’s a very dangerous take.”

High revenue can conceal a shallow product

This mindset trap isn’t the only delusion Torres cautions against. He also warns that high revenue can be a deceptive metric of a product’s true value. This insight comes from his experience at Drift, which he co-founded and later sold for over US$1 billion in 2021.

Drift’s initial success masked a product gap
Torres explains, “[the] product was shallow, it wasn’t too deeply penetrated because it was only solving one basic step in the customer journey. It only did the first connection”.

A shallow product creates a broken customer experience
Torres describes the problem this created for customers later on. “Everybody would be like, ‘I came back to the website. Why doesn’t it recognize [me]? Why doesn’t it pick up where it left off? I already had this conversation… How come when I’m inside the app, it doesn’t know what I said before?'”

Building for obsession creates sustainable value

The failure of human-led growth

AI must operate autonomously, not as a co-pilot