Cobo: How are we using AI to drive transformation?

Author: alexzuo4, Investment & Custody VP @Cobo

Since the end of 2024, besides our core crypto custody and stablecoin payment services, Cobo has been exploring the integration of AI and blockchain.

Our earliest insight was the potential of MCP to bring standardized skills. In theory, if skills are sufficiently standardized, AI can invoke capabilities like plugins, making blockchain the most natural infrastructure for AI-driven finance.

So we internally incubated an MCP application store. But it was quickly disproven.

At that time, AI barriers were still high, only mature engineers could proficiently invoke capabilities. MCP was not yet standardized; each integration was time-consuming, labor-intensive, costly, and progress was slow. The implementation far fell short of expectations.

But the AI team was still built. It was expensive, hard to recruit for, and not easily dismantled.

So we decided to pivot. Since we couldn’t transform our clients’ world yet, we would first transform ourselves.

The first issue: security

As an asset custody company, Cobo handles extremely sensitive data and internal technical workflows. We have strict data hierarchies internally. But without data and real business input, it’s impossible to develop our own Agent.

Initially, we considered deploying local models. But in reality, local models couldn’t meet the intelligence requirements. They could run, but were not user-friendly; they could answer questions, but weren’t smart enough.

In the end, we chose Claude and Gemini as main models (with the option to apply for ZDR—Zero Data Retention clauses—to achieve the highest level of isolation).

But large models are just the “brain” underlying our business. The real complexity lies in data and permissions.

We then developed a comprehensive internal knowledge base and Agent framework.

Internal Knowledge Base + Cobo’s self-developed Agent system

The knowledge base manages internal data layering, assigning read permissions based on employee roles.

When Agents access the knowledge base, they inherit employee permissions, not a “god’s eye view.”

Details include:

• How to isolate network environments
• How to restrict cross-layer data flow
• How to control log retention for auditability
• How to prevent sensitive information leaks

These may not be glamorous, but they determine whether this system can run long-term. AI must not become a security vulnerability.

After building the architecture, the next challenge: no one uses it

Even today, the company still faces a harsh reality: many front-office teams are dismissive of AI.

Encouragement alone won’t change workflows.

We later realized we had to start from company management.

The first breakthrough: OKR Agent

Our initial focus wasn’t customer service or coding.

It was OKRs.

We used AI to deconstruct company strategy, help set OKRs, track progress, and review milestones.

In other words, transforming management from human-driven to a co-governance model with silicon. This process is extremely uncomfortable for employees.

Previously, goals could be polished and processes explained convincingly. Now, weekly data is transparent, leaving fewer excuses.

From that moment, goals ceased to be just meeting discussions; they became continuously recorded in the system.

Strategy OKRs weekly monitor business progress

But it was only through performance management that employees truly became familiar with AI, because if they didn’t participate, it would directly impact their compensation.

From performance to business: full Agent integration

Once OKRs were running smoothly, we pushed for internal service Agent deployment. We enforced each department to establish and operate their own relevant Agents through competitions and bonuses.

Customer service teams built Customer Service Agents. Legal teams developed Contract Assistance Agents. Sales teams created CRM Agents.

We sought out the most eccentric client Agents.

In total, over 100 Agents went live.

We can’t precisely quantify the results of “silicon-carbon co-governance.”

But one clear change is:

Before, when problems arose, the first reaction was “should we hire more people?” Now, it’s “can we get the system involved first?”

This is our understanding of silicon-carbon co-governance. It’s not AI replacing humans, but humans getting used to working alongside systems.

The journey over the past year has yielded some practical insights:

First, maintain healthy cash flow.

If the company’s cash flow isn’t healthy, this transformation can’t reach the finish line. AI isn’t a cost-saving tool; it’s an upfront investment for long-term structural upgrades. Thanks to Cobo’s core business, we have healthy cash flow.

Second, push from the top down.

Organizations won’t change spontaneously. Without strong management push, this effort will naturally fail.

As is well known, Cobo’s founders are heavy AI enthusiasts. CTO Dr. Jiang started AI research during his postdoc at CMU in the early 2000s.

Third, enforce mandatory use.

Encouragement alone keeps AI at the level of writing emails. Real process change requires some “compulsion.”

Fourth, solve your own business first.

Many companies talk about AI + Web3. But if they haven’t internalized AI, they’re just talking concepts externally.

Looking back

We can’t fully quantify this transformation. The company has shifted from “people-driven processes” to “goal-driven systems.”

If a “smart organization” truly emerges in the future, it won’t be through natural evolution. It will be pushed out through uncomfortable iterations.

Because with everyone involved, the company can better understand the real needs in the AI era.

This is also a byproduct of our internal transformation.

Recently, we launched Cobo Waas Skill. Cobo WaaS Skill is an integrated operational layer designed specifically for AI Coding Agents. Through structured knowledge, executable examples, and scenario orchestration, it enables Agents to accurately invoke WaaS APIs. We are upgrading wallet APIs into financial capability modules directly callable by AI Agents. Development cycles have shortened from weeks to dialogue-level.

This isn’t the result of a single product idea. It’s the natural spillover of our internal silicon-carbon co-governance.

We are still exploring.

But at least, today’s Cobo is no longer the same company it was in 2024.