Inside XerpaAI’s Vision: CTO Bob Ng on Building the World’s First AI Growth Agent

Trusted Editorial content, reviewed by leading industry experts and seasoned editors. Ad Disclosure

1. Please introduce the founding background of XerpaAI. As part of the UXLINK ecosystem, how does XerpaAI position itself as the “world’s first AI Growth Agent”, and what is its core mission? In the Web3 field, what pain points exist in traditional growth models (such as manual marketing and KOL collaborations), and how does XerpaAI solve these problems through AI?

A: The establishment of XerpaAI originated from the UXLINK ecosystem. We observed that Web3 startups face significant challenges in terms of growth, such as high-cost manual marketing, inefficient collaborations relying on KOLs, and fragmented user acquisition. As the world’s first AI Growth Agent (AGA), our core mission is intelligent growth, helping WEB3 startups shift from manual operations to an intelligent and self-driven expansion model. The pain points of traditional growth models include: high marketing budgets (global technology companies spend 600 billion to 1 trillion US dollars annually on growth), subjective and time-consuming KOL matching, and difficulty in scaling community interactions. XerpaAI addresses these issues through AI-driven content generation, intelligent distribution, and real-time optimization. For example, it automatically generates multilingual content and distributes it through a network of over 100K KOCs/KOLs on platforms such as X, Telegram, and TikTok, achieving a 3x increase in conversion rates and a 70% reduction in costs.

2. XerpaAI’s core concept is the “intelligent growth engine”. Does this mean it can completely replace human growth teams? Considering 2025 AI trends, such as the autonomous agent model of agentic AI, how do you view XerpaAI’s role in helping startups transition from “manual expansion” to “intelligent self-drive”?

A: Yes, our core concept is to build an “intelligent growth engine” that can significantly reduce reliance on human growth teams, but not completely replace them — instead, it serves as an enhancer, allowing teams to focus on strategy rather than execution. In 2025, the rise of agentic AI endows AI agents with stronger autonomy, and XerpaAI is a manifestation of this trend: it acts like an intelligent Sherpa guide, autonomously handling user behavior analysis, incentive triggering, and campaign adjustments, helping startups transition from “manual expansion” to “intelligent self-drive”.

3. What is XerpaAI’s technical architecture? How does it integrate AI models (such as content generation and real-time optimization) with Web3 native elements (such as link-to-earn mechanisms and social graphs) to support project growth?

A: XerpaAI’s technical architecture is a highly modular multi-AI Agents system designed to handle complex tasks in Web3 growth, such as automated user acquisition, community expansion, and KOL/KOC matching. We have built the entire system as a collaborative agent network, where each agent focuses on specific subtasks but collaborates seamlessly through shared states and communication protocols (such as blockchain-based smart contract verification). This is a form of multi-agent agentic workflows, where agents can autonomously plan, execute, and optimize action paths, thereby achieving an end-to-end intelligent growth engine.

At its core, XerpaAI’s architecture revolves around a central AGA (AI Growth Agent) coordinator that oversees the interactions of multiple dedicated agents, forming a dynamic decision-tree structure. The following is a detailed breakdown from the perspective of multi-AI Agents:

Composition of the agent network:

– Planning Agent: This is the entry point, responsible for decomposing high-level growth goals (such as “increasing user conversion rates for a DeFi project”) into executable subtasks. It adopts the Plan-and-Solve prompting strategy, an advanced zero-shot reasoning method that first formulates a comprehensive plan (for example, dividing tasks into content generation, KOL matching, and performance optimization) and then solves each subtask step by step. This method addresses the missing steps issue of traditional Zero-Shot Chain-of-Thought (CoT), ensuring that the agent does not skip key reasoning links. For example, when handling a WEB3 viral marketing task, the planning agent will first plan:

“Step 1: Analyze the target audience;

Step 2: Generate multimodal content;

Step 3: Match platform-specific KOLs;

Step 4: Monitor real-time feedback.”

– Data Collection Agent: Responsible for real-time collection and preprocessing of multi-source data from the Web3 ecosystem (such as blockchain transactions, social graphs, cross-platform user interactions). Data sources include X, Telegram, on-chain activities (such as smart contract interactions), and the social graph of the UXLINK ecosystem. As the input layer of the multi-agent system, the data collection agent provides real-time, structured data streams for other agents (planning, content generation, distribution, optimization, integration), ensuring that decisions are based on the latest insights. For example, it extracts interaction trends from over 110K communities for the planning agent to decompose tasks.

– Content Generation Agent: Focuses on creating multilingual, multimodal content (such as text, images, and videos). It utilizes Zero-Shot Chain-of-Thought prompting by adding “Let’s think step by step” to induce step-by-step reasoning, such as deriving personalized narratives from user data without the need for pre-trained examples. This allows the agent to generate high-quality content in a zero-shot setting, supporting cross-platform distribution (such as X, Telegram, and TikTok).

– Distribution & Matching Agent: Handles intelligent matching and content distribution within the 100K+ KOL/KOC network. It integrates Web3 native elements such as social graph analysis and link-to-earn mechanisms, using multi-agent collaboration to optimize paths — for example, decomposing the matching process through Plan-and-Solve into “planning a list of potential KOLs, then solving compatibility and incentive allocation”.

– Optimization & Feedback Agent: Monitors performance indicators (such as conversion rates and costs) in real-time and adjusts strategies through self-reflection loops. It运用 Zero-Shot CoT to analyze data biases, such as step-by-step reasoning “If the conversion rate is lower than expected, why? Step 1: Check content relevance; Step 2: Evaluate KOL influence; Step 3: Adjust incentives”, thereby achieving a 70% cost reduction and a 3x increase in conversions.

– Integration Agent: Bridges AI and Web3 components, ensuring decentralized verification (such as data privacy on the blockchain) and cross-track support (DeFi liquidity incentives, SocialFi community building).

Multi-agent collaboration mechanism:
Agent communication is achieved through a shared knowledge graph based on GraphRAG technology, allowing real-time data ingestion and reasoning. The central coordinator uses an A* search-inspired algorithm to navigate the action space, avoiding inefficient paths and ensuring efficient execution.

We have incorporated Plan-and-Solve as the core reasoning engine to overcome the limitations of Zero-Shot CoT (such as calculation errors or semantic misunderstandings). For example, in a SocialFi project, the planning agent first formulates a plan: “Subtask 1: Identify target communities; Subtask 2: Generate interactive content; Subtask 3: Distribute and optimize”, and then each agent uses Zero-Shot CoT to solve them step by step, avoiding reliance on manual examples.

This multi-agent system supports parallel processing and iterative learning: if one agent fails (such as the matching agent not finding a suitable KOL), the feedback agent triggers a reflection loop to re-plan the path. This design follows multi-agent trends, such as inter-agent teaching and optimization in simulated environments.

Memories support:

XerpaAI enhances the learning and adaptive capabilities of the multi-agent system through a Memories mechanism (based on long-term context storage), storing historical tasks, user preferences, and optimization results, similar to a “near-infinite memory” architecture. This enables agents to reuse knowledge across tasks and continuously improve.

Memories are stored in a distributed knowledge graph (based on GraphRAG) combined with a vector database (Milvus) to support efficient retrieval. Each agent (planning, content generation, distribution, optimization, data collection) stores key decisions and results in Memories, such as “A project’s KOL matching increased conversion rates by 3x, and high-interaction KOLs should be prioritized”.

As a shared resource, Memories promote collaboration between agents. The data collection agent stores new data in Memories, the content generation agent adjusts its creations accordingly, the distribution agent optimizes KOL matching, and the optimization agent evaluates performance, forming an adaptive loop.

Memories endow the system with “memory”, enabling agents to learn historical patterns and optimize future tasks. For example, after a failed viral marketing campaign for a WEB3 project, Memories record the reasons for failure (such as insufficient incentives), and the planning agent adjusts the incentive mechanism for new campaigns accordingly.

The essence of XerpaAI’s Memories is to build an external brain for XerpaAI’s users, transforming fragmented knowledge into reusable structured memories through hierarchical storage, dynamic indexing, and MCP protocols.

Overall, this architecture makes XerpaAI more than just a tool but an adaptive growth partner that has served over 110K communities. Through the collaboration of multi-AI Agents, coupled with advanced prompting technologies such as Plan-and-Solve and Zero-Shot Chain-of-Thought, we have achieved efficient, zero-shot automation of Web3 growth. If you have specific task examples, I can further demonstrate how these components are applied.

4. In the 2025 AI breakthroughs, small specialized models and inference time computing are becoming focal points. Has XerpaAI adopted similar technologies to handle massive amounts of data (such as 100K+ KOL matching and cross-platform distribution, including X, Telegram, and TikTok)? How does its data analysis engine ensure real-time feedback and self-optimization?

A: Yes, we have adopted small specialized models to handle specific tasks such as KOL matching and cross-platform distribution. These models are optimized for Web3 data to reduce inference time. In line with the 2025 trend of inference time computing, our engine uses efficient algorithms to process massive amounts of data, such as real-time matching from over 100K KOLs and distribution across X, Telegram, and TikTok. The data analysis engine ensures self-optimization through machine learning loops: collecting user interaction data, applying reinforcement learning to adjust strategies, and avoiding overfitting.

5. XerpaAI has served over 110K communities. How does it utilize multimodal AI (combining text, images, and social data) to automate user acquisition and community interaction? Compared with current AI trends such as near-infinite memory and custom silicon, what are XerpaAI’s innovations in edge computing or cloud integration?

A: XerpaAI utilizes multimodal AI to process text, images, and social data, such as generating image-enhanced content or analyzing social graphs to automate interactions, and has served over 110K communities. Compared with 2025 trends such as near-infinite memory, we have innovated in cloud integration by using distributed computing to process large-scale data; in terms of edge computing, we have optimized mobile agents to ensure low-latency interactions, such as real-time responses to user queries in Telegram groups.

6. XerpaAI has a network of over 100K KOLs/KOCs. How does it serve these influencer groups through AI tools (such as personalized content generation and incentive optimization) to help them improve monetization efficiency and community interaction, thereby establishing a mutually beneficial channel advantage? Considering 2025 AI trends such as personalized agents, how do you think this will amplify the viral spread of Web3 projects?

A: XerpaAI’s 100K+ KOL/KOC network is the core of our channel advantage. Through AI tools such as personalized content generation and incentive optimization, we provide tailored services to these influencers to help them improve monetization efficiency and community interaction. For example, our AGA engine uses multimodal AI to generate exclusive content (such as images, video scripts, or posts targeting specific audiences) and maximizes their income through real-time incentive optimization (such as dynamically adjusting revenue sharing ratios based on interaction data) — this can increase KOLs’ monetization efficiency by 2-3 times while enhancing community stickiness, such as automated replies and gamified interactions. The result is mutual benefit: influencers gain more exposure and revenue, while we expand our distribution channels through their networks. In the 2025 AI trends, personalized agents (such as custom AI assistants) are dominating the influencer economy, and XerpaAI is a pioneer in this application — our agents can autonomously learn KOL preferences and predict trends, thereby amplifying the viral spread of Web3 projects. For example, in a DeFi campaign, through KOCs’ micro-sharing chains, exponential user growth can be achieved, with conversion rates increasing by more than 5 times.

7. When serving KOLs/KOCs, what strategies has XerpaAI adopted to ensure data privacy and fair revenue sharing (such as through blockchain-verified link-to-earn mechanisms) to cultivate long-term loyalty? How does this channel advantage translate into a competitive barrier for startups, especially in multi-platform distribution (such as X, Telegram, and TikTok)?

A: When serving KOLs/KOCs, we prioritize Web3-native strategies to ensure data privacy and fair revenue sharing: all interaction data is verified through the blockchain (such as using zero-knowledge proofs to store anonymized information) to prevent leakage; the link-to-earn mechanism automatically executes revenue sharing based on smart contracts, ensuring transparency and instant payments (such as token rewards based on interaction metrics), which cultivates long-term loyalty — our retention rate exceeds 85%. This channel advantage translates into a competitive barrier for startups: in multi-platform distribution (such as real-time tweets on X, group interactions on Telegram, and short videos on TikTok), our network forms a “moat”, providing exclusive access and optimized paths, helping enterprises bypass traditional advertising bottlenecks and achieve low-cost, high-efficiency growth. For example, a WEB3 project covered 5 million users in 3 weeks through our KOL/KOC channels, while competitors needed several months.

8. In 2025, with the rise of AI agents, data privacy and algorithmic bias are key challenges. As a Web3 & AI-native platform, how does XerpaAI ensure transparency and decentralization (such as through blockchain verification)? What are its considerations regarding AI ethics?

A: Data privacy and algorithmic bias are crucial. As a Web3 & AI-native platform, we ensure transparency through blockchain verification, such as using decentralized storage to protect user data and conducting fairness audits to avoid bias. Our AI ethical considerations include: anonymization of all model training data, user-controllable opt-out mechanisms, and regular third-party audits to comply with regulatory trends.

9. XerpaAI recently secured $6 million in seed funding, led by UFLY Capital. How will this funding be used for expansion? Please share a specific case, such as how it helped a Web3 startup achieve growth from scratch, highlighting its role in user acquisition and community building.

A: This $6 million seed funding will be used for product iteration, international expansion (such as team recruitment in Silicon Valley, Tokyo, and Singapore), and ecosystem integration. A typical case is our assistance to a Web3 startup: starting from scratch, our AGA generated multilingual content, distributed it through the KOL network, built a community graph, and ultimately acquired 100,000 users within one month, with community activity increasing by 2 times. This highlights our role in user acquisition and community building.

10. Looking to the future, how will XerpaAI integrate into broader AI trends such as personalized AI agents or automated investment? What are the company’s next technical iteration plans? What advice do you have for AI entrepreneurs to cope with the dynamic changes in Web3 growth?

A: In the future, XerpaAI will integrate into the trend of personalized AI agents, such as custom growth paths, and explore automated investment modules. The next iteration includes enhancing multimodal capabilities (such as video generation) and deeper Web3 integration. Advice for AI entrepreneurs: focus on pain points such as growth automation, embrace agentic AI, and build ecosystem partnerships to cope with the dynamic changes in Web3 — for example, monitor real-time trends and iterate quickly. XerpaAI’s service capabilities will also empower KOLs/KOCs, enabling this group to enhance their respective influence with the help of XerpaAI.

11. As CTO, what is your greatest expectation for the integration of AI and Web3? How does XerpaAI help more startups “connect, expand, and dominate the market”? Finally, what would you like to say to potential partners or users?

A: As CTO, my greatest expectation for the integration of AI and Web3 is to realize a truly decentralized intelligent economy, where AI Agents such as XerpaAI drive intelligent growth. XerpaAI will help more startups “connect, expand, and dominate the market” through our AGA engine, providing end-to-end support from content to optimization. Finally, to potential partners and users: join us to speed up your growth — welcome to visit xerpaai.com to try it out, or DM us to discuss cooperation!

Editorial Process for bitcoinist is centered on delivering thoroughly researched, accurate, and unbiased content. We uphold strict sourcing standards, and each page undergoes diligent review by our team of top technology experts and seasoned editors. This process ensures the integrity, relevance, and value of our content for our readers.