"Send My AI"

Prediction:

AI user agents and legacy applications will become adversaries in the short term, competing for direct access to end users.

In the long term, legacy applications will maintain direct access to end users by getting out of their way.


Who Is The Customer?

Conversational AI interfaces like Chat GPT are displacing traditional search engines. In many cases, Chat GPT will give a useful answer to your query faster than Google without showing you ads.

When applications become successful and widely used, they tend to implement measures that reduce the quality of service they offer you to extract value from your usage – namely ads, engagement, and retention optimization.

Sign up, sign in, OAuth, restrictive APIs, decisions about what’s above vs below the fold, the number of clicks – these are purely negative impositions that reduce your quality of experience and accrue enterprise value to the app maker.

AI User Agents For Humans

AI interfaces threaten to disrupt such heavily encumbered applications by intermediating them and offering better service, free of all the crap, to end users.

AI user agents – programs that perform actions on behalf of users inside of other applications – are now emerging, using AI to emulate user interactions at the web browser or operating system level.

You might prefer to use an AI user agent to interact with an application for the same reasons you’d prefer Chat GPT over Google:

  • It’s faster and easier: You just express what you want. The reading, sifting, scrolling, and clicking are done for you.
  • It’s better service: There are no ads or dark patterns designed to waste your time.

AI User Agents For App Integrations

It turns out that the extractive measures legacy applications impose on their users extend beyond direct in-app use.

If you want to access data from one application inside another, for example, accessing your credit card transactions from your financial planning application, or accessing your Twitter or Reddit activity from your preferred third-party client application, you’re probably out of luck.

If legacy applications were incentivized to offer you the best service, these things would be table stakes. But they’re not.

With AI user agents, applications can offer you integrations with other upstream applications you use, even if they don’t have a permissive API.

Instead of using an API, the downstream application can use an AI user agent that emulates end-user activity in the upstream application to access data and take actions.

This is similar to how Plaid and Mint.com each worked in their early days, though their automations were driven by scripts, not AI.

Conflicts And Interests

Legacy applications and emerging AI user agent applications alike both want the same thing: direct access to users. To exist only behind the scenes is to risk being commoditized.

Users want the best service. Simply put, users just want whatever they enjoy using and feel is best.

Weaknesses Of Abusing GUIs

Emulating user interactions in a GUI is a hack with fundamental weaknesses.

First, it’s slow. GUIs offer mostly useless overhead to AI user agents. They’re packed with ceremony intended for humans that is at best useless and at worst distracting to AI user agents.

While the AI user agent is slowly navigating the GUI, the user is waiting and becoming impatient.

Beyond slowness, it’s riddled with incidental complexity. The method involves abusing a GUI, which may accidentally or intentionally contain content that misleads the AI agent and causes it to fail.

What Will Legacy Applications Do?

Many legacy applications will be increasingly intermediated by AI user agents and will find themselves on their heels, trying to defend their access to and relationship with the end user.

They will need to battle on two frontlines:

  • Defensive: Interfering with AI user agents. Legacy applications will implement measures to detect and interfere with AI user agents. AI user agents will implement measures to prevent detection and interference.

  • Offensive: Offering better service. Legacy applications will exploit the weakness of AI user agents to their advantage: abusing GUIs is slow and complicated.

This means they can beat AI user agents by backing off the extractive measures that caused the user experiences they offered to become unfavorable.

Permissive, interoperable APIs and fast GUIs designed to serve rather than exploit the end user are capable of offering a faster and better end-user experience than GUI interaction-emulating AI user agents possibly can.

Where This Is Going

In the short term, AI user agents will offer end users a preferable, albeit not perfect, experience by intermediating legacy applications and acting on users’ behalf.

Legacy applications will engage in a defensive battle against AI user agents, each party will try to remain a step ahead of the other.

Then, applications will attempt to displace their intermediaries by encouraging users to prefer interacting directly, rather than via an AI user agent.

Effectively, this will mean evolving beyond the use of obstructive and extractive measures and instead embracing a more open, user-centric model of service.