Former Meta and TikTok Executive Wei Xu Builds Leverage AI to Give Businesses Greater Control Over AI Agents

Leverage AI’s Locus Platform Targets a New Challenge in Artificial Intelligence: Who Owns and Controls the Memory, Permissions and Operational History of Autonomous AI?

 

As artificial intelligence moves beyond answering questions and begins performing meaningful business tasks, technology companies are confronting a new challenge: how should businesses control AI systems that can remember information, access company data and act autonomously?

For technology entrepreneur Wei Xu, the answer requires rethinking some of the basic infrastructure behind AI Agents.

Xu, a former Meta technical leader and TikTok/ByteDance product manager, is the co-founder and CEO of Leverage AI Inc., a U.S.-based artificial intelligence company developing Locus, an AI Agent infrastructure platform designed around customer ownership, security, auditability and data portability.

The company says Locus has completed development of its core architecture and is now entering commercialization in the United States.

Rather than competing primarily on the intelligence of AI models, Leverage AI is focusing on the infrastructure surrounding those models—including memory, permissions, identity, synchronization and operational records.

“AI is moving from answering questions to actually doing work,” Xu said. “Once an AI system can send an email, modify a business record, access sensitive information, or initiate a workflow, transparency and control become fundamental infrastructure requirements.”

From Scaling Consumer Technology to Building AI Infrastructure

Xu’s interest in AI infrastructure follows a career spanning engineering, consumer technology, product development, monetization and entrepreneurship.

Before founding Leverage AI, Xu held technical leadership roles at Meta, where he worked on advertising monetization and consumer growth.

According to information provided by Leverage AI, Xu led a 15-person team whose advertising products expanded from approximately $400,000 in daily revenue to more than $40 million per day.

He also contributed to major consumer growth systems, including work related to Facebook’s People You May Know recommendation infrastructure and initiatives associated with approximately 20 million daily active users.

During his time at Meta, teams led by Xu reached the finals of the company’s internal hackathons three times, with product prototypes receiving senior-level review.

Xu later moved from large technology companies into entrepreneurship.

In 2021, he co-founded LiveIn, also known as Livehouse, and served as CEO. According to company figures, the social application grew from zero to more than two million daily active users within approximately four months of its U.S. launch without a conventional paid marketing campaign.

LiveIn subsequently reached No. 1 overall on the U.S. App Store.

Xu later advised consumer technology companies on U.S. product growth. Leverage AI says applications he supported included eight products that entered the top five of the U.S. App Store overall rankings, including LiveIn at No. 1 and Clapper at No. 2.

From 2023 through 2025, Xu worked at TikTok/ByteDance as a product manager focusing on video and e-commerce products and participated in the development and relaunch of TikTok Now.

Those experiences ultimately shaped his view of the emerging AI Agent economy.

“The first generation of generative AI was primarily about creating content,” Xu said. “The next generation is about AI participating in operations.”

“That changes the engineering problem. Intelligence alone isn’t enough. Businesses need to know who controls the Agent, what it knows, what it is allowed to do, and whether they can audit it afterward.”

A New Problem Emerges as AI Becomes Autonomous

Generative AI initially gained widespread adoption through systems capable of producing text, images, software code and other content.

AI Agents represent a potentially more consequential stage of that evolution.

Instead of merely generating information, an Agent may be authorized to interact with business systems, communicate with customers or vendors, update records, schedule employees, prepare documents and execute workflows.

That increased authority creates new questions around security and governance.

A company may need to understand what information an Agent remembers, what systems it can access, who authorized an action and how that action can later be reconstructed.

Leverage AI believes these issues will become increasingly important as AI Agents are deployed across real-world organizations.

The company’s Locus platform is designed to address those challenges by giving businesses greater visibility into and control over the operational state surrounding their AI systems.

Making AI State an Asset Controlled by the Business

A central idea behind Locus is what Leverage AI calls “customer-owned AI state.”

The company argues that an AI Agent’s accumulated memory and operational history may eventually become an important form of business infrastructure.

An Agent operating for several years could accumulate conversations, institutional knowledge, workflow history, permissions, configuration information and relationships between employees, customers and business systems.

If that information exists only within a proprietary vendor platform, changing providers could become increasingly difficult.

Locus is being developed around a “zero-lock-in” principle under which organizations should be able to inspect, export, back up and migrate important AI state.

“AI memory will become business infrastructure,” Xu said. “If years of institutional knowledge accumulate inside an Agent, that information should remain an asset of the company—not an asset that exists only because the company continues paying one software vendor.”

Human-Readable Infrastructure for AI Agents

One of the platform’s central architectural concepts is known as File-Is-State.

Instead of requiring important AI memory and operational history to exist exclusively inside a proprietary database, Locus is designed so that an Agent’s workspace can be represented through a human-readable file structure.

Conversations, memories, permissions, configurations and other important state can therefore be inspected and managed using conventional computing tools.

For technical users, familiar tools including version control, backups, file comparisons and command-line utilities can be used to inspect portions of the environment.

The larger objective is transparency and portability.

Leverage AI is also developing Context-Based Access Control, or CBAC, an authorization model intended for AI Agent environments.

Traditional access-control systems commonly focus on roles and attributes. AI Agents create another layer of complexity because whether information should be accessible may depend on the context in which the Agent is requesting it.

Locus is designed to incorporate elements of that context into authorization decisions.

The platform also uses synchronization technology drawing on Conflict-Free Replicated Data Type, or CRDT, principles, allowing AI state to operate across multiple devices and environments while supporting offline operation and conflict resolution.

Co-founder and CTO Wenjie Tan leads engineering for Locus, including its distributed state architecture, edge synchronization, CRDT implementation and security infrastructure.

Tan and Xu jointly developed the architecture and protocol design underlying the platform.

Building AI Infrastructure for Smaller Businesses

Although sophisticated AI infrastructure is increasingly important, many small and medium-sized businesses do not have dedicated AI, cybersecurity, infrastructure or compliance teams.

Leverage AI sees this as a significant adoption challenge.

Large corporations can invest substantial resources in AI security, access controls, internal integrations, audit systems and data governance. Smaller companies often cannot.

Yet those businesses could benefit significantly from AI automation.

“Our goal with Locus is to make advanced AI automation practical for businesses that may not have the engineering, security, or compliance resources of a Fortune 500 company,” Xu said.

Leverage AI is initially targeting digitally active U.S. small and medium-sized businesses operating across multiple applications and devices.

The company estimates an initial target segment of approximately 330,000 digitally sophisticated U.S. SMBs that are comparatively well positioned to adopt AI-driven workflows.

Locus is expected to be offered through subscriptions beginning at approximately $50 to $100 per business per month.

From AI Application to AI Infrastructure

Leverage AI did not initially begin with Agent infrastructure.

The company’s earlier work applied artificial intelligence to digital advertising, developing technology intended to automate elements of video-advertisement creation, localization and optimization.

Its early commercial activity included work with social media platform Clapper.

Through those deployments, the founders began confronting a broader problem.

Generating content was becoming easier, but giving AI systems persistent responsibility for business workflows introduced much more difficult questions around memory, permissions, identity, synchronization, security and accountability.

Those questions ultimately led to the development of Locus.

The company is now advancing the platform into commercialization while pursuing U.S. venture investment to support continued research and development, security engineering, product commercialization and team growth.

Leverage AI says its hiring strategy prioritizes U.S.-based engineering and technical talent.

Who Controls the AI?

For Xu, the long-term question surrounding AI may ultimately extend beyond model intelligence.

As autonomous systems become more capable, businesses will increasingly need to determine who owns their accumulated memory, who controls their permissions and whether their decisions and actions can be audited.

“When software was passive, vendor lock-in was primarily a technology and economic issue,” Xu said. “When software can act autonomously on behalf of a company, control of that software becomes a governance issue.”

“Our principle is straightforward: your AI should work for your business, its memory should belong to your business, and its actions should be accountable to your business.”

As AI Agents take on increasingly significant responsibilities inside organizations, that principle could become an important part of how businesses evaluate the next generation of artificial intelligence infrastructure.

About Leverage AI Inc.

Leverage AI Inc. is a U.S.-based Delaware C corporation developing artificial intelligence infrastructure for small and medium-sized businesses.

Its Locus platform is designed around customer-owned AI state, human-readable data, contextual access control, distributed synchronization, edge-first infrastructure, auditability and data portability.

The company is advancing Locus through commercialization in the United States while expanding its engineering, security and product-development efforts.