Most companies adopt AI the way they adopt any new gadget: one more tool on one more laptop. On October 12, Liu Run, a widely read Chinese business commentator, argued for a different approach in a post on his WeChat channel. He calls it a "five-star architecture". His subject is personal computing, but the problem he describes will be familiar to any Chinese company operating in Kazakhstan: knowledge lives in too many places, work stalls when a manager is on the road, and every new tool or rule change seems to force a fresh start. What follows is our own reading of his framework, not a translation of his text.
The idea in brief
Think of a company, Liu Run says. Production and records sit at headquarters. The owner directs things from a meeting room or an airport lounge. Customers deal with the firm at a shop front. Yet many people run all three roles on one laptop, which is asked to compute, display and face the public at once. His architecture pulls these jobs apart.
The hub is the first star: a capable machine that stays switched on and does the real work. What matters most, in his view, is not raw power but permanence. Connectors, data and tools pile up on a machine that is always there. You can buy tools; accumulated context only comes with time.
The touchpoints are stars two and three: a laptop and a phone. The laptop is for reviewing what the hub has produced and issuing the next instruction. The phone lets you hand the hub a task from a taxi or a departure gate, then get on with your day.
The interfaces are stars four and five: two modest cloud servers, one facing China and one facing the rest of the world. They are the public doors through which a small Q&A page or lookup tool can reach customers and colleagues.
His conclusion: once the structure is stable, a new model or tool is a spare part, not a reason to rebuild. He links this to the "productivity J-curve" described by Erik Brynjolfsson, Daniel Rock and Chad Syverson (American Economic Journal: Macroeconomics, 2021; earlier an NBER working paper). A general-purpose technology first demands investment in invisible complements such as processes, skills and organisation, so measured productivity can stall before it accelerates.
Why this fits doing business in Kazakhstan
Information here is fragmented by design. Project details sit with 17 regional and 3 city administrations and with companies, mostly in Russian or Kazakh. Laws are authoritative only in the official text on adilet.zan.kz and are amended often. Border, port and freight conditions change by the quarter, sometimes by the month. Head office is in China, the team is in Kazakhstan, the cargo is somewhere in between. If what the company knows lives in one manager's laptop and chat history, it leaves when that manager does.
The hub: knowledge base, editorial review, enquiry log
For a business, the hub is less a machine than a single place where knowledge is kept, owned and checked. Ours has three parts. A knowledge base whose policy answers cite the act, its number, article and effective date. Editorial review: entries drafted by editors or AI assistants go into a review queue and are published only after a person signs off. And a record of questions and enquiries: questions the Q&A cannot answer go onto a to-do list for the knowledge base, requests left through forms are followed up by the local team, and the outcome flows back as new entries.
A company can build the same thing: one repository for legal texts, contracts, customs documents and border and freight records, each item tagged with its source and date, with a named person responsible for review. Expiry matters. As our Middle Corridor guide noted, the Caspian leg in the Trans-Caspian route association's annual tariff table was marked valid only until March 31, 2026. A figure without an expiry condition quietly turns into a wrong figure.
Touchpoints: news, guides, WeChat Moments
Touchpoints are where people look at results and decide what to do next. For a platform, news and long-form guides are the screen on which the hub's work appears. Our news pages carry key points, FAQs and sources, plus a WeChat QR code so a story can be passed to colleagues or posted to Moments, and a head-office manager can grasp an issue on a phone.
Inside a company the rule is the same: the weekly report for head office and the phone summary for the owner should both draw on the one hub. Keep touchpoints thin. They are for seeing and steering; the evidence stays in the hub.
Interfaces: Q&A, service forms, agent entry points
Interfaces are where outsiders deal with the system. On our site there are three. The Q&A on the home page is answered by AI: policy answers come from a verified policy base; anything outside it gets a reference answer marked as unverified, and a person replies within 24 hours when needed. The "Tell us" and business-service forms take requests on company registration, visas and staffing, logistics and customs, site selection and more; services are delivered by providers vetted by the platform, which does not issue legal opinions in its own name. The project section lists 15 special economic zones and industrial parks, all currently marked "pending verification". The platform's Chinese name is 途马 (Tuma). According to the platform operator, the Jiangsu Center for Central Asia has added a Tuma AI agent to its mini-program as an enquiry entry point; the scope of actual services should be agreed in writing with the operating company.
Liu Run runs one server for China and one for abroad. A company can mirror that: a Chinese-language interface for head office and suppliers, and a Russian or Kazakh one for local customers, parks and partners, both fed by the same hub so the two never contradict each other. That is why our site serves four languages from a single body of content.
Rules related to this news
- Which foreign employees are exempt from work permits and quotas in Kazakhstan?Under Migration Law Art.Text and legal basis →
- China outbound investment ODI filing for KazakhstanA Chinese investor must complete approval/filing, information reporting and cross-border fund registration where requi…Text and legal basis →
- Investment incentivesFrom 1 January 2026 the former investment priority project regime is replaced by three instruments: (1) investment agr…Text and legal basis →
Three ways to apply it
Investment. When parks, zones and projects go into the hub, label each as "intention", "signed" or "delivered", with source and date of verification. Show only verified items externally and mark the rest clearly.
Trade. Turn recurring customer questions, on certification, duties or payment terms, into standard answers reviewed by legal or customs staff before they reach a Russian-language enquiry page or chatbot. New questions go to a backlog rather than being improvised.
Logistics. Data moves fast. Kazakhstan's transport minister reported on October 6 that truck queues on the Nur Zholy–Khorgos route had fallen from about 30 days to about 3, that more than 11,000 digital transport passports had been issued, and that the Smart Cargo platform draws on 58 data sources. Official figures like these belong in the hub; contracts and production plans should still leave room for border delays. External tracking platforms are exactly the kind of new part an interface can plug in.
Slow first, then faster
The J-curve is a reminder that building structure and fixing processes may show little return at first. Rather than switching methods every time a tool appears, separate hub, touchpoints and interfaces; test new tools in the hub and expose them only once they prove reliable.
A quick checklist
- Are legal texts, contracts and border records kept in one place, with a named reviewer?
- Does every item show its source, update date and expiry condition?
- Do head-office reports and customer replies come from the same hub?
- Do external channels separate verified from unverified content, with a route to a human?
- Are new tools tested in the hub before going public?
Editor's note
Liu Run is writing about personal AI hardware; applying his frame to a business has limits, since a company's hub also involves compliance, data security and delegated authority. The core point carries over: separate accumulation, judgement and the public window first, then worry about upgrading tools.
This is editorial commentary, not legal, tax or investment advice. Current Kazakh legislation and the competent authorities prevail.
Note: the cover is an AI-generated illustration and does not depict a real scene. AI tools helped organise material for this article, which was reviewed by an editor. This English version was written for English-language readers on the basis of the Chinese article, not translated sentence by sentence.
Sources:
Liu Run, "尽快，给自己搭建这套五星架构吧", WeChat channel "刘润", October 12 (inspiration only; not reproduced), https://mp.weixin.qq.com/s/9sJ_xO3M-ffvfDCTf7lEUQ
Brynjolfsson, Rock, Syverson, "The Productivity J-Curve", American Economic Journal: Macroeconomics, 2021, https://www.aeaweb.org/articles?id=10.1257/mac.20180386
NBER Working Paper w25148, https://www.nber.org/papers/w25148
tumah.kz home page (checked October 12), https://tumah.kz/?lang=en
tumah.kz business services (checked October 12), https://tumah.kz/services?lang=en
tumah.kz projects (checked October 12), https://tumah.kz/projects?lang=en
tumah.kz: Nur Zholy–Khorgos truck queues cut from 30 days to 3, https://tumah.kz/news/nur-zholy-khorgos-digital-freight-passport-queue-2026
tumah.kz: Middle Corridor routes, transit times and customs, https://tumah.kz/news/middle-corridor-kazakhstan-china-europe-routes-transit-times-customs
tumah.kz: Services of the Jiangsu Center for Central Asia, https://tumah.kz/news/jiangsu-center-central-asia-suhao-khorgos-services
Key facts
- The article divides enterprise operating systems into a hub, touchpoints, and an interface, responsible respectively for accumulating information and making judgments, viewing results and issuing instructions, and handling external interactions.
- Kazakhstan project information is distributed among government bodies and enterprises across 17 regions and 3 cities of republican significance.
- The official texts published on adilet.zan.kz are authoritative for Kazakhstan’s regulations and are frequently amended.
- The Q&A on the 途马 homepage distinguishes verified policy responses from unverified reference responses and provides a route to human support.
- The 途马 project database lists 15 special economic zones and industrial parks, all of which are currently marked as “Pending verification.”
- Kazakhstan’s Minister of Transport reported that the queue of freight trucks in the Nurlure–Khorgos direction has shortened from approximately 30 days to approximately 3 days.
FAQ
What three components are included in the “Five-Star Architecture” proposed in the article?
The article divides enterprise operating systems into a hub, touchpoints, and an interface: the hub is responsible for accumulating information and making judgments; touchpoints are used to view results and issue instructions; and the interface handles external interactions.
How should Chinese enterprises build the hub repository?
Original regulations, contracts, customs documents, border crossing records, and freight rate records can be stored centrally, with their sources, update dates, and expiry conditions clearly identified and reviewed by designated personnel.
How does 途马 answer questions that are not covered by its policy repository?
途马 provides reference answers labeled as unverified and, when necessary, can transfer the matter to a human representative.
How should enterprises upgrade their systems in response to new models and tools?
New tools should first be connected to the hub for trial use and made available through the interface only after they pass verification, avoiding the need to rebuild the entire system each time a tool is replaced.