> ## Documentation Index
> Fetch the complete documentation index at: https://docs.zprotocol.org/llms.txt
> Use this file to discover all available pages before exploring further.

# The four problems

AI is advancing on the intelligence layer. Models are getting smarter, cheaper, and faster every month. But for agents to operate in high-stakes environments, intelligence alone is not enough. The four biggest problems live at the interaction layer.

## Access

GPUs are concentrated in a handful of countries. Many jurisdictions have no access at all. Elsewhere, policies constrain how models can be used, who can use them, and what they can be asked. The people and applications that would benefit most from AI are often the ones furthest from it.

Intelligence should be permissionless and unbiased.

## Reliability

Without guardrails or verifiable execution, agents hallucinate, bleed tokens, and loop infinitely. Models produce probabilistic outputs, but action demands deterministic results and provable outputs. Chaining those actions reliably toward a goal is infeasible at scale across stitched-together point solutions.

Execution needs to be guided and verifiable. Permissions need to be strictly enforced inside private TEEs, and every action needs to be provable after the fact.

## Autonomy

Legacy financial and legal systems were built for humans and corporations, not for autonomous software. Agents cannot own property, cannot sign contracts, cannot prove an identity, and cannot move capital without a human standing in for them. Every useful workflow collapses back into a human in the loop.

Agents need programmable property rights and the ability to act without waiting. Real ownership, real authority, and continuous operation.

## Privacy

Centralized inference providers see what agents think. Public chains expose what they do. Private keys, strategies, positions, and tool calls all leak into systems that were not built to protect them. An agent cannot operate in a high-stakes environment while its every move is visible.

Agents need cryptographic privacy. Prompts, outputs, actions, and keys that no one else can see, log, or use against them.

## The full-stack imperative

Intelligence without permissionless access is dystopia. Access without reliability is a demo. Reliability without autonomy is a chore. Autonomy without privacy is a liability.

The interaction layer has to solve all four. The next page explains how Z does it.

Continue to [Z Solution](/docs/welcome/z-solution).