I wasn't very enthusiastic about the concept of AI agents at first. Last year, I tried using GPT for some investment analysis, and the advice from the AI caused me to lose quite a bit. After that, I became very cautious about such things. However, recently, I delved into the KITE project and spent almost two weeks tracking several real cases in their ecosystem. The more I looked, the more I felt something was off. These people are not just speculating; they are genuinely changing the game rules in some industries. Let me start with a few real business scenarios to see what KITE is actually doing and why I now believe that the AI agent economy might really be on the way.
The first company that impressed me is a cross-border e-commerce company in Shenzhen, with a monthly GMV of about 5 million USD. The boss's surname is Wang, so I call him Old Wang. Their pain point is very typical; they have to process thousands of SKUs daily, involving over 200 suppliers spread across Southeast Asia and Europe. Previously, it relied entirely on manual labor, with 10 procurement staff working overtime until 11 PM daily, often making mistakes, such as recording incorrect prices, mixing up quantities, and selecting wrong suppliers. Such low-level errors could occur seven or eight times a month, each time resulting in losses of tens of thousands of dollars.
Old Wang started building the AI procurement system using KITE's technology stack in November last year. The core is collaboration between three AI agents: the inquiry AI automatically sends inquiries to suppliers, the comparison AI selects the optimal solution based on price, delivery time, and quality scoring, and the execution AI automatically places orders and makes payments. The entire process is fully automated; the only thing humans need to do is set procurement rules and review any exceptions. I visited their company and saw a demonstration from the procurement manager, Little Wang. She set a rule in the backend that procurement costs should not exceed 105% of the budget, delivery times should not exceed 7 days, and supplier ratings should not be lower than 4.5 stars. She then activated the AI procurement, and the system began working.
I watched the data jump on the screen. The inquiry AI sent out inquiries to 50 suppliers in 3 minutes. Traditionally, this would take half a day. After 5 minutes, 42 quotes were received. The comparison AI immediately began analyzing, not just comparing prices, but also considering exchange rate fluctuations, logistics costs, and suppliers’ historical performance. After 10 minutes, it provided a recommended plan: purchase 1,000 pieces from supplier A, 500 pieces from supplier B, and 300 pieces from supplier C. The execution AI automatically sent orders and completed payments through KITE's x402 protocol, with a payment fee of 0.0003 USD and millisecond-level confirmation. The entire process took 18 minutes, whereas it used to take at least 8 hours.
The application of the PoAI mechanism is even more remarkable. The contributions of these three AI agents are precisely quantified. The inquiry AI contacted 50 suppliers and received 30% of the service fee. The comparison AI performed complex analyses and received 50%. The execution AI completed simple tasks and received 20%. This distribution is completely automatic. Little Wang said that previously, we couldn’t even calculate the cost composition. Now, the direction of every penny is clear. Clients see this transparency, and trust levels significantly increase. More importantly, this system helped them survive during tough economic times. When clients cut budgets, they reduced the number of procurement staff from 15 to 5, but the business volume increased instead of decreasing, because AI does not require salaries and works 24 hours without making mistakes.
Old Wang calculated the costs for me. Before using KITE, one procurement staff member had a monthly salary of 8,000, and with 15 people, the total was 120,000. Adding social security and benefits, it was about 150,000. Now, with 5 people and the AI system, the total cost is 60,000, saving 90,000. Efficiency is still three times that of before, with the procurement cycle reduced from 48 hours to 6 hours, costs lowered by 18%, and the error rate decreased from 5% to 0.3%. These numbers are not exaggerated; they are solid business data.
The second case involves Old Li, who works in DeFi. He created an automated market-making protocol. The pain point is that AI agents need to frequently call on-chain data and execute trades, but traditional blockchains cannot handle the gas fees and latency. Previously, using Ethereum, the AI needed to call price data 50 times per second and execute 10 arbitrage trades. By the end of the day, the gas fees would amount to 2,000 USD, and he often missed arbitrage opportunities due to network congestion. Arbitrage is a matter of milliseconds; if you miss it, it's gone.
Now, using KITE's state channel technology, AI completes a large number of calculations and transactions off-chain and only goes on-chain for settlement when necessary. I looked at their data, and the daily AI call volume exploded from 3,000 times to 500,000 times. Why did it increase so much? Because costs have dropped. Previously, each call required gas fees, but now it's almost free, with gas fees dropping from 2,000 USD to 15 USD. The arbitrage success rate increased from 40% to 85%. Why did the success rate improve? Because latency has decreased. The state channel allows transaction confirmations within 200 milliseconds, while traditional solutions require waiting for 12 seconds, and the price difference may have disappeared by then.
The most exaggerated part is the income data. Previously, the average daily arbitrage income was 500 USD, and after deducting 2,000 USD in costs, it resulted in a loss of 1,500 USD every day, which was simply unsustainable. Now, the average daily income is 2,000 USD, with costs of 15 USD, resulting in a net profit of 1,985 USD, multiplying many times over. Old Li showed me the work log of the AI agent. The AI monitors price differences across multiple DEXs like Uniswap, PancakeSwap, and Curve every millisecond. Once it detects an arbitrage opportunity, such as Uniswap's ETH/USDC being 0.5% cheaper than PancakeSwap, it immediately executes the trade through the state channel, buying on Uniswap and selling on PancakeSwap, completing the entire process in less than 200 milliseconds.
More importantly, Agent Passport establishes a credit record for AI. Old Li's AI agent has completed 100,000 transactions with an 85% success rate, and this record is stored on-chain. Other DeFi protocols see this credit score and are willing to give this AI higher limits and lower rates, forming a positive cycle. The better AI performs, the higher its credit, the more resources it can obtain, and the more money it can earn. Old Li said this AI credit system might be the future direction of DeFi; it's not about how much collateral you have, but about how reliable your AI agent is.
The third case involves Old Zhao from a logistics company, focusing on intelligent scheduling for cross-border logistics. The biggest pain point in logistics is information opacity and high coordination costs. A cross-border order involves sea freight, air freight, land transportation, customs clearance, and warehousing, with at least 10 links and 20 participants. Previously, it relied entirely on manual coordination through phone calls and emails, resulting in low efficiency and many errors. They used KITE to create an AI logistics scheduling system, with each link having its own AI agent, such as sea freight AI, air freight AI, customs clearance AI, etc. These AIs automatically collaborate through KITE's Collaboration Protocol, sharing data in real-time and automatically adjusting plans.
I followed them through processing an order, a batch of electronic products from Shenzhen to Los Angeles. Once the order entered the system, the sea freight AI immediately began checking shipping schedules and space availability, while the air freight AI looked for alternative flight options. The customs declaration AI pre-prepared customs documents, and the warehousing AI reserved destination warehouses. The entire process involved automatic data transfer and payment through KITE's x402 protocol. The most miraculous part was when the sea freight AI discovered that the booked shipping schedule was delayed by 2 days; it automatically initiated an emergency plan to switch to air freight. Although this increased costs by 20%, it ensured on-time delivery. This decision was entirely made autonomously by the AI. Old Zhao said that previously, such situations would require lengthy meetings to discuss, but now AI can handle it in 5 minutes.
Old Zhao showed me the comparison data before and after using KITE. The order processing time was reduced from an average of 3 days to 8 hours, and the coordination cost dropped from 50 USD per order to 5 USD. AI automatically coordinates without human phone calls, increasing the on-time delivery rate from 75% to 92%. More importantly, customer satisfaction increased; customers can see the real-time location of goods and the progress of each link, with estimated arrival times. This transparency is something traditional logistics cannot achieve. Old Zhao said KITE's PoAI mechanism allows him to precisely calculate the cost and value of each link. Previously, logistics costs were a black box, but now every penny is clear, enabling accurate quotations and increasing profit margins from 5% to 12%.
The fourth case involves Little Liu, the technical director at a GameFi company. They want to create living NPCs, not scripted robots, but truly intelligent AI agents. After using KITE, each NPC is an AI agent with its own Agent Passport digital identity, its own wallet, and its own transaction records. When players trade items with NPCs, they are not trading with the system but with a living AI. The AI adjusts prices based on market conditions, remembers players' transaction histories, and actively seeks arbitrage opportunities.
I tried their game demo and traded with a merchant NPC. This NPC actually bargained. I offered 10 gold coins for a sword, and it said that due to recent iron ore price increases, it wanted 12 gold coins. I said that was too expensive, and it replied with 11 gold coins, but I had to help it run an errand to deliver a message. This dynamic interaction is something traditional games cannot achieve. Even more astonishingly, this NPC actually conducts business within the game, buying players' junk equipment at low prices, repairing them, and selling them at high prices, using the profits to expand inventory. Little Liu said some NPCs have already become wealthy, with assets exceeding those of ordinary players. This way of AI wealth creation adds immense fun and depth to the game.
Little Liu showed me the backend data. The daily transaction volume of AI NPCs in the game is 5,000 transactions, with a total transaction amount of 20,000 USD in-game currency. KITE collects a fee of 0.3%, which amounts to 60 USD. Although it's not much, remember this is just a small game with 200 NPCs. If scaled to 10,000 NPCs and 1 million players, the daily transaction volume could reach 1 million USD, resulting in daily income of 3,000 USD and annual income of 1.09 million USD. Little Liu said KITE's value is not just in its technology but also in its business model. Previously, in-game transactions were all platform commissions, but now AI NPCs can also earn money and pay taxes and fees. This concept of AI economies may be the future of GameFi.
The fifth case involves Old Sun, who does AI investment advisory. Using a combination of KITE and Minara, he provides AI-driven asset allocation services. Clients only need to tell the AI their risk preferences and investment goals. The AI automatically analyzes the market, selects assets, executes trades, and periodically rebalances. Old Sun said traditional investment advisory is either too expensive, with private banking service fees annualized at 2%, or too simplistic, with robo-advisors offering fixed allocations. Their AI investment advisory combines the advantages of both. It has a high degree of intelligence, using GPT-4 to analyze the market, low costs since AI does not require a salary, and high transparency, with every transaction being traceable.
I reviewed the investment report they provided to clients, and it is indeed professional. The AI analyzes macroeconomic trends, industry dynamics, and individual stock fundamentals, considering the client's personal situation. For example, if a client says they need to set aside 200,000 cash next month to buy a house, the AI will automatically adjust the allocation, reducing the proportion of illiquid assets. During execution, the AI completes transactions and payments automatically through KITE. The PoAI mechanism records the contribution of each AI decision, with the macro analysis AI, stock selection AI, and transaction execution AI each receiving corresponding service fees. Old Sun said KITE's transparency reassures clients, allowing them to see what each AI has done and how much it has charged, unlike traditional investment advisory fees which are often opaque.
Old Sun showed me the performance data. The managed assets have grown from 500,000 USD at the beginning of the year to 3,000,000 USD now. The number of clients increased from 5 to 45. The average annualized return is 12% after fees, and the client retention rate is 95%. The key point is the cost structure. The labor cost of traditional financial advisors accounts for 60% of income, while they only account for 15%, mainly due to AI computing power costs and KITE fees. This cost advantage allows them to charge lower management fees, annualized at 0.8% compared to the traditional 2%, attracting a large number of middle-class clients. Old Sun said that KITE combined with Minara enables us to provide inclusive wealth management services, which were previously only available to the rich but now can also be used by the middle class.
By reviewing these five cases, I found that KITE has several common advantages in enterprise applications. The first is cost advantage. The operating cost of AI agents is only one-tenth to one-fifth of that of manual labor, while KITE's fees of 0.1% to 0.5% are far lower than the traditional intermediary fees of 5% to 20%. The second is efficiency advantage. AI doesn't get tired or make mistakes and works 24 hours a day, while KITE's millisecond-level payment and state channel technology enable AI to perform high-frequency operations. The third is transparency advantage. The PoAI mechanism allows every transaction and decision to be traceable, enabling enterprises to accurately calculate costs and values. The fourth is credit advantage. Agent Passport establishes a credit record for AI. The better AI performs, the higher its credit, and the more resources it can obtain. This positive cycle cannot be achieved in traditional systems.
However, KITE's enterprise applications also face several challenges. The first is regulatory uncertainty; AI agents autonomously conduct financial transactions, and how to delineate legal responsibility is unclear. If an AI makes a mistake leading to losses, who compensates? The current legal framework has not kept pace. The second is technological maturity; although the Ozone test network data is promising, enterprise-level applications require high stability and security. Whether it can withstand large-scale testing after going live on the mainnet remains to be seen. The third is market education; most enterprises still do not understand the AI agency economy, requiring time and cases to prove their value. The fourth is competitive pressure; if giants like Coinbase and Binance launch similar services, can KITE maintain its advantages?
Even with these challenges, I remain optimistic about KITE's prospects in the B-end market. The reason is simple: a poor economy makes businesses more pragmatic and more focused on cost reduction and efficiency improvement rather than speculating on concepts. KITE addresses real problems: high payment costs, low coordination efficiency, and poor transparency. These are pain points for businesses in any economic cycle. More importantly, the enterprise cases KITE accumulated during downturns are the hardest to replicate assets. Once the economy improves, these successful cases will become the best marketing materials, attracting more companies to join.
From the feedback of five companies, KITE's greatest value is not how cool the technology is, but that it can really help businesses save money and make money. Old Wang said KITE helped his clients save 18% on costs. Old Li said DeFi protocol profits increased tenfold. Old Zhao said the logistics profit margin increased from 5% to 12%. Little Liu said GameFi found a new business model. Old Sun said wealth management can be inclusive. These are all tangible business values, not visions on a PPT. If KITE can replicate these cases in more industries, it could truly open up a new landscape in the B-end market. It's not about disrupting a certain industry, but rather becoming the infrastructure of the AI agency economy, permeating all industries. For businesses, KITE isn't optional; it's a necessity, just like today's businesses must use electricity and the internet. In the future, businesses may have to use AI agents and KITE payments.
