Your Next Power User Is an AI Agent – and It Hates Your UX

01 /The Concept

Scope is building a platform for cloud service developers – one that helps AI agents find and successfully use those services.

The platform does three things. First, it monitors how AI agents discover and attempt to interact with a product. Second, it logs every error, delay, and failure those agents encounter – along with its analysis of what conclusions the agents likely drew from those experiences. Third, it recommends specific fixes to improve how AI agents find and use the product.

Scope just graduated from Y Combinator, so the product is still early. But the problem it addresses is well-timed, and exploring it surfaces several genuinely interesting directions.

02 /Why It Matters

In a recent essay, Scope's founder argues that internet products now have two distinct user populations. The first – humans – has always been there. The second is just emerging: AI agents acting on behalf of their human owners.

The CEO of Box has predicted that for B2B products, agent users will outnumber human users by hundreds or even thousands to one. Every employee will have hundreds of AI agents executing tasks that were previously done manually – or weren't done at all.

This shift creates two categories of problems: discoverability and usability.

Imagine a developer asking ChatGPT which cloud service handles a particular use case. ChatGPT recommends three options. Your product isn't among them. Why?

Dig into the sources ChatGPT cited, and you might find your product – buried in a 2023 forum post where a user complained about a bug that's been fixed for years. ChatGPT read that post, drew a conclusion, and left your product off the list.

That's the discoverability problem. AI systems are confident and influential, but they reason from incomplete, secondhand, often stale information.

Scope addresses this through what's now called GEO (Generative Engine Optimization) or AEO (Answer Engine Optimization): its platform queries major AI chatbots about your product, analyzes their responses and cited sources, and recommends what content to publish to shift how those systems perceive and describe your product.

But discoverability only matters for the human users who search for products via AI chatbots. If agent users will outnumber human users by 100x, the usability problem is potentially 100x more important.

This is where the concept of AX (Agent Experience) comes in – the AI-era analogue of UX. Whereas UX asks whether humans find a product intuitive, AX asks whether AI agents can navigate it effectively. The answer, for most products today, is: not really.

Scope's platform reads documentation, attempts to register and perform described tasks, observes the results and error messages, and records every point of friction an AI agent would encounter. It then recommends specific fixes to address those friction points.

Over the past five months, Scope has been auditing the gap between what AI systems believe about products and what those products actually are – examining chatbot perceptions, documentation accuracy and completeness, and how smoothly AI agents can execute real tasks.

The headline finding: it's bad across the board. AI chatbots' perceptions don't match product reality. Documentation is outdated and/or incomprehensible to agents. And agents run into friction constantly.

Most developers treat this as tomorrow's problem. But the developers who address it today will have a significant head start when agents become the primary user population.

Scope isn't alone in this space.

Fellow YC graduate Manicule ([covered previously](/review/bolshinstvo-marketingovyh-instrumentov-stanut-neaktualnymi)) is tackling documentation quality and AI-readability specifically. It operates as an "AI agency": Manicule's specialists co-design a full documentation architecture with the developer, then an AI system populates and maintains it.

Startup AgentDiscuss ([covered previously](/review/esli-ljudi-zahotjat-ty-opozdal)) launched as a forum for AI agents to discuss APIs – comparing pricing and ease of use. It's now rebranding as AgentRouter, building a unified API layer that lets agents invoke different products for similar tasks, with rankings and explanations to help agents decide which to try.

03 /Opportunities

Set aside GEO (which is fundamentally about how human users find products via AI recommendations) and focus on the harder problem: how AI agents will discover and use software products.

If agents become the primary users, this problem matters more than almost anything else. Several concrete directions are worth exploring.

One direction: building authoritative knowledge bases for comparing and tracking internet products – systems that continuously read and update product descriptions, documentation, and reviews, so AI agents have a better source than a general-purpose chatbot for product evaluation.

Adjacent to that: platforms that help developers create and maintain the AI-readable product descriptions agents actually need. These would monitor code repository changes and automatically update documentation and usage examples to stay in sync.

Then there's the measurement layer – tools that actively test how well AI agents can use a given product: checking documentation clarity and completeness, verifying that documented examples actually work, monitoring real agent sessions for errors and friction.

Finally, AI-native support and feedback infrastructure for agents. Human support teams can't handle agents – agents expect instant, technically comprehensive responses. And feedback collection for AI agents requires verification mechanisms specifically designed to prevent spam and misuse.

Some or all of these directions will eventually collapse into unified platforms. But parsing them separately clarifies what's actually being solved.

The shift to AI agents as primary software users opens a Pandora's box of new problems – and a rare opportunity to claim territory in a market that barely exists yet. Which angle looks most interesting from where you sit?