New practice · AI Plumbing

Fix the pipes first — then AI can flow.

Where the flow stops, it hurts.

The most common blocker in listed companies isn't the lack of an AI strategy — it's that data simply doesn't flow. AI Plumbing treats every department as a pipe: inspect each one, then open it up.

Run the diagnosis How we engage

The problem · data silos

Every department, an island of data

Inside one company: some teams still run on Excel and paper forms; some on a legacy ERP that can't export; some on modern SaaS; some on in-house builds. Different quality, different storage, different interfaces.

The result: you can't even imagine AI use cases — the strongest model does nothing if data can't reach it.

The metaphor · the company as a building

Four pipes, four states

Departments are pipes: floors the water can't reach, AI can't reach either.

The estate · one view of the building

Six pipes, six fates 诊

Click any pipe — or a chip below — to see its diagnosis and treatment.

The treatments · one state, one path

Three treatment paths

Path one · Rebuild

Blocked & aging pipes: replace

Applies to offline teams and legacy ERPs that can't export — the prescription is the same: give the data a new home first.

data-quality review → architecture & schema → migration → full-stack AI app · bespoke front-end

See it live: Paper records → structured ↗ P&L analysis ↗

Path two · Sidecar

ERPs with APIs: agents on the outside

The fastest path to value: the pipe itself is healthy — intelligence is attached outside it, and the ERP stays the single source of truth.

interface audit → API + MCP integration → write-back → zero intrusion

See it live: Third-party risk ↗ Sourcing agent ↗

Path three · Upgrade

In-house systems: build AI in

Code you own is the best pipe there is — AI capability grows inside the system: an evolution, not an add-on.

codebase review → upgrade plan → native integration → continuous evolution

See it live: Meeting → ticket → PR ↗ Definition extraction ↗

The end state · flow, then the hub

Silos opened 通 · centrally orchestrated 枢

Department agents exchange data freely; a centralised management dashboard monitors and orchestrates every agent — the hub that 灵枢 (Lingshu) is named for.

The catalog · what grows in each pipe

An agent for every function

The trunk is live — pick a function to see the agents it can grow. Tiles marked ↗ have a live demo.

Advisor · conversational analysis Autonomous · executes end-to-end

A sample map, not a fixed SKU list — every agent is designed bespoke for your business.

Why not just buy an ERP suite

Why not just roll out SAP?

Wholesale ERP replacement

Configuring a listed-company instance: ~18 months
The architecture is not AI-native — intelligence is bolted on
Processes are templates, not bespoke — the business bends to the software
AI features arrive on the vendor's roadmap, not yours

The AI Plumbing way

Start one pipe at a time — value in weeks
Every pipe is AI-native: designed for agents
Processes are fully bespoke — the software bends to the business
Capability evolves continuously, at your pace

Which pipe first

Three ways to choose the first pipe

digitalisation ↓

The least digitalised department

The biggest before/after contrast — the clearest way to show the organisation what AI is worth.

labour cost ↑

The most labour-intensive department

Where repetitive human effort is heaviest, agents pay back fastest.

key-person risk

The key-person-dependent department

The hidden risk of 'nothing moves without one person' — turn personal know-how into organisational capability.

Once one pipe flows, the next is faster — schemas, connectors and the agent framework are all reused.