Decentralization for SMEs: InvoiceNow & AI Agents

Part 2 of a two-part series on business philosophy for the age of platforms and AI. Part 1 covered the ideas: the kampong, the platform life cycle, Ostrom, Hayek, and subsidiarity. This part turns them into practice.

How can a small business benefit from decentralization without taking on crypto risk or running a cooperative? The practical answer: join open networks built on shared standards, keep your data unified and portable, own the tools that make your decisions (especially AI), and use a simple three-question test before depending on any single platform. Most of the useful decentralization happening today is quiet, boring, and has nothing to do with tokens.

In Part 1, we arrived at a principle: centralize the truth, decentralize the power. Keep one trustworthy version of your data and standards. Spread decisions, value, and the ability to exit across the people who do the work.

That sounds good on paper. This article asks whether it works in the real world, where it is already happening, and what it means for a Singapore SME deciding what to do next quarter.

Where does decentralization already work without anyone noticing?

The most successful decentralized systems are so ordinary that we forget they are decentralized.

Email. Nobody owns email. You can send a message from a Gmail address to an Outlook address to a company’s own mail server, and it simply arrives. That works because everyone agreed on a shared standard decades ago. Thousands of providers compete, and you can move your address if you are unhappy. The standard is centralized. The power is spread out.

The phone network. You can call any number from any carrier. The carriers compete fiercely, but they agreed to connect. Imagine if your phone could only call people on the same network. That is roughly what many digital platforms look like today.

The web itself. Any website can link to any other website, without permission. The shared rules for how pages are addressed and delivered made that possible.

Notice the pattern. None of these needed a blockchain. None needed a cooperative. None needed idealism. They needed one thing to be centralized: the standard. Once everyone speaks the same language, competition and choice take care of the rest.

This is the most important and least glamorous insight in this entire series. Open standards are the most proven form of decentralization humans have ever built.

What is InvoiceNow, and why is it a decentralized network?

Here is one that most Singapore business owners have heard of, but few think of in these terms.

InvoiceNow is Singapore’s nationwide e-invoicing network, run under IMDA and built on an international standard called Peppol. Instead of emailing PDF invoices, businesses send structured invoice data directly from one accounting system to another.

The design is what matters for this series. Peppol uses what is called a four-corner model:

The supplier (corner 1) sends through its chosen access point (corner 2).
That access point delivers to the buyer’s chosen access point (corner 3).
Which delivers to the buyer (corner 4).

There is no single company in the middle holding all the invoices. Many accredited providers compete to be your access point. You can switch providers. Your buyer can use a completely different provider from you. Singapore adds a fifth corner for tax reporting to IRAS, but the exchange itself remains spread across many providers.

That is textbook “centralize the truth, decentralize the power”:

Centralized: the invoice format, the business identifiers, the rules.
Decentralized: which provider you use, which software you run, who you trade with.

And this network is no longer optional. IRAS has announced that the GST InvoiceNow requirement will be extended in phases to all GST-registered businesses, beginning with smaller businesses on 1 April 2028 and reaching all remaining GST-registered businesses by 1 April 2031. New voluntary GST registrants are already required to use it. To ease the transition, eligible SMEs can apply for a S$1,000 grant to offset the cost of connecting, available from July 2026.

Think about what this means philosophically. A government is building a decentralized network and asking every GST-registered business to join it. Not a platform the state owns and operates for every transaction. A shared set of rules that lets a whole market of providers compete.

For SMEs, this could be much bigger than a compliance chore. Once every business in the country has a network identity and speaks the same invoice language, the door opens to other things: purchase orders, delivery notes, payment updates, supplier discovery. The road has been paved. What travels on it next is up to the market.

What happened when India built an open network for commerce and mobility?

India has become the world’s most ambitious laboratory for this idea.

It started with payments. India’s UPI system let any bank app pay any other bank app instantly, and it transformed daily life. Encouraged by that, the country then built the Open Network for Digital Commerce (ONDC), on an open protocol called Beckn.

The goal was to do for shopping what UPI did for payments. A customer can use one app to discover products from sellers registered on completely different apps. Discovery, payment, fulfilment, and logistics are unbundled, so a customer or seller can even choose their own delivery provider. According to figures shared by India’s Commerce Minister, ONDC has facilitated more than 350 million transactions so far.

The most striking story, though, came from mobility.

In 2022, an auto-rickshaw app called Namma Yatri launched in Bengaluru, built on the Beckn protocol. It was created with the Auto Rickshaw Drivers’ Union, and it made an unusual promise: no commission on rides. Instead, drivers pay a small flat daily subscription and keep the fare.

The results were bigger than the company. Namma Yatri crossed 150 million trips, and competitors responded structurally: one major rival switched from commissions to subscriptions, and another dropped commissions for auto drivers entirely. A small app on an open protocol changed the pricing logic of an entire national market.

It has since gone further. In September 2026, Namma Yatri launched a store that lets its hundreds of thousands of drivers buy together at below-market prices, with the platform taking no margin. That is gotong royong in digital form: not just earning together, but spending together.

The philosophical lesson: you do not always have to beat a dominant platform to change it. Sometimes you only have to prove another design works. Once the market sees it, even the incumbents move.

Do cooperatives actually work in the digital economy?

The idea of businesses owned by the people who use them is almost two hundred years old. In 1844, a group of weavers in Rochdale, England, opened a small shop they owned together. Their rules became the foundation of the modern cooperative movement.

Today, a newer wave called platform cooperatives tries to apply that model to apps.

Stocksy United, a Canadian stock photography platform, is owned by its contributing photographers, who receive a large share of each sale and vote on how the business is run.
CoopCycle provides open-source delivery software to a federation of more than 60 courier cooperatives in Western Europe.
The Drivers Cooperative in New York launched as a driver-owned alternative to Uber and Lyft, and its website says members earn more per trip, with profits returned to drivers.

The honest picture is mixed. Many platform co-ops struggle with the “empty room” problem described in Part 1: without venture capital, filling both sides of a marketplace is hard. It is telling that The Drivers Cooperative now describes itself as specialising in paratransit and non-emergency medical transport, work with steady, contracted demand, rather than competing only for app downloads.

Academic research points to where cooperatives thrive. Professional cooperatives, where members are similar, earn decent incomes, and need only simple matching technology, tend to scale with little difficulty. Some European examples have thousands of members.

The lesson for business owners: cooperative ownership works best when the members are alike, the demand is steady, and the technology is simple. It struggles when it tries to out-market a venture-funded giant in a consumer app store.

Is blockchain the answer?

Blockchain and crypto tokens are the loudest part of the decentralization conversation, so they deserve a clear-eyed look.

The most serious work here is in what is called DePIN, decentralized physical infrastructure networks. The idea: instead of one company building all the infrastructure, many individuals contribute hardware (wireless hotspots, dashcams, spare computing power) and are rewarded for it. Some of these networks now earn real revenue from paying customers.

But a 2026 analysis of the sector summarised the catch in its own title: the networks profit, the individual nodes often do not. Network-level revenue is real, but profits concentrate among a small group of well-positioned operators. Another analysis described the trap plainly: if token rewards run ahead of real customer revenue, you get inflation and people leave; if rewards drop before demand arrives, contributors leave anyway.

Even the DIN whitepaper we mentioned in Part 1, written by a team deeply inside the crypto world, is cautious. It describes starting in a “federated” phase with traditional invoicing, and says the team is wary of decentralizing too early before proving a sustainable business model. It deliberately avoided launching a token with a floating price.

That is a remarkable admission from a crypto project. It echoes a principle that software engineers know as Gall’s Law, named after systems theorist John Gall: a complex system that works is invariably found to have evolved from a simple system that worked. You do not start with a fully decentralized, token-governed global network. You start with something simple that works, and decentralize it step by step.

The lesson: technology can make decentralization cheaper to run. It cannot make it succeed on its own. The design principles from Part 1, clear boundaries, fair rules, real voice, honest monitoring, still do the heavy lifting.

Will AI agents be your staff, or your landlord?

Now we arrive at the question that makes this philosophy urgent rather than academic.

AI agents are software that can take actions on your behalf: answering customers, drafting quotations, reconciling accounts, placing orders, chasing invoices. In the next few years, many SMEs will hand a meaningful share of their daily operations to agents. (We explored the commercial side of this shift in the rise of Agents as a Service, and what to watch for in Odoo 20’s agentic AI features.)

That raises exactly the question from the kampong. Who owns the thing you depend on?

There are two possible futures, and both are technically realistic.

Future A: The agent landlord. A handful of giant platforms run the agents. Your business rents them. Your customer conversations, your pricing logic, your supplier relationships, and your operational know-how all flow through systems you do not control. The platform learns from every business on it. Prices rise when you are dependent. This is the platform life cycle from Part 1, repeated one level deeper, not just controlling your sales channel, but your decision-making.

Future B: The agent workforce. Your business owns or controls its agents, the same way it owns its staff relationships and its ERP data. The agents run on your data, follow your rules, and can be moved to a different provider if needed. They talk to other businesses’ agents through open standards, the way your email talks to anyone’s email.

Future B is not a fantasy. The standards are already being built.

An open protocol called A2A (Agent2Agent), originally developed by Google and donated to the Linux Foundation, lets AI agents built on different frameworks discover each other, exchange tasks, and collaborate without exposing their internal workings. By its first anniversary in April 2026, it counted more than 150 supporting organisations, with production use inside major cloud platforms from Microsoft, Amazon, and Salesforce.

That matters for the same reason email matters. If agents can talk across vendors, then no single vendor needs to own the conversation. Your purchasing agent can negotiate with a supplier’s sales agent, even if the two were built by completely different companies.

What does Conway’s Law predict about AI agents?

In 1967, programmer Melvin Conway made an observation that became famous in software circles: organisations design systems that mirror their own communication structure. A company with four teams that do not talk to each other will build software with four parts that do not talk to each other.

Turn that around and it becomes a warning for the AI era. The systems you adopt will reshape your organisation to match them.

If all your agents live inside one vendor’s platform, your business will gradually organise itself around that vendor’s categories, limits, and pricing tiers. Your processes will bend to fit the tool.

If your agents are yours, built around your processes, speaking open standards, your organisation stays shaped by your own judgement.

This connects straight back to Hayek and subsidiarity. The promise of AI agents is that they can push good decisions down to the level closest to the work: the agent that notices a supplier is late, the agent that spots a customer about to churn. That is subsidiarity, automated. But if the agent is really controlled from a distant platform, then the decision has not moved closer to the work at all. It has moved further away, and just looks local.

The question to ask about any AI tool is not only “what can it do?” but “whose judgement does it carry, and whose data does it learn from?”

How can a business decide what to centralize and what to decentralize?

Here is the practical tool this series has been building towards. Before depending on any platform, system, or network, ask three questions.

Question 1: How much does the middle take, and how much does it lock you in?

Look beyond the fee. Ask:

If this provider raised prices by 30% tomorrow, could we leave?
Do we own our customer list, our data, and our history here, or does the platform?
Can we export everything in a usable format?

If the middle takes a lot and locks you in tightly, the case for an alternative, or at least a backup route, is strong.

Question 2: Are the participants similar enough to agree on rules?

Ostrom’s commons worked because members shared a situation. Cooperatives thrive among similar professionals.

Are the other participants businesses like yours, with similar needs?
Could you realistically agree on shared rules, standards, or purchasing?

If yes, a shared network, a buying group, or a cooperative arrangement may work. If participants are wildly different, a neutral central operator is often more practical.

Question 3: Is there a neutral standard, or a regulator, pushing everyone the same way?

Decentralization succeeds fastest when the language is already agreed.

Is there an open standard (like Peppol for invoices, or A2A for agents) that many providers support?
Is a regulator requiring participation?

If yes, join the open network early. You get the benefits of competition among providers without having to build anything yourself.

Reading the results

Answers	What it suggests
High lock-in + similar participants + open standard	Strong candidate to decentralize. Join or build a shared network.
High lock-in + different participants + no standard	Use the platform, but protect your data and build an exit route.
Low lock-in + any	Centralized platform is probably fine. Revisit yearly.
Open standard exists + regulator push	Adopt early. The road is being paved for you.

Apply this to InvoiceNow and it passes all three questions. Apply it to consumer ride-hailing and it struggles on the second. Apply it to AI agents and the honest answer today is: lock-in risk is high, standards are emerging, and now is the time to choose tools that keep your options open.

What should a Singapore SME actually do next?

Philosophy is only useful if it changes something on Monday. Here are practical moves, roughly in order of effort.

1. Get your own truth in order. Before anything else, have one trustworthy version of your customers, stock, orders, and finances. You cannot participate in any network, open or closed, with messy data. This is the “centralize the truth” half, and it is the foundation.

2. Treat InvoiceNow as an opportunity, not only a deadline. Check your implementation date with IRAS, confirm your system is InvoiceNow-ready, and consider the transition grant. Then ask what else you could exchange with trading partners once you are connected.

3. Audit your platform dependence. List every platform your revenue travels through. For each, apply the three questions. Where lock-in is high, start building a second route: your own customer list, your own website orders, your own delivery arrangements for key accounts.

4. Own your customer relationships. Platforms may own the transaction. Make sure you own the relationship: permission to contact, purchase history, preferences.

5. Choose AI tools that keep you in control. (If staff are already using AI tools you have not approved, start with our guide to shadow AI governance.) Prefer agents that run on your own data, follow your rules, can be exported or moved, and support open standards. Be cautious of any AI tool that becomes impossible to leave.

6. Look for your gotong royong. Are there businesses like yours that could buy together, share delivery capacity, share a warehouse, or share specialist staff? Some of the most effective “decentralization” is simply neighbours cooperating on purpose.

7. Start simple (Gall’s Law). Do not design the perfect network on day one. Start with something small that works between a few partners, and grow it.

A closing reflection: carrying the house together

Go back to the kampong one last time. When the neighbours lifted a house onto their shoulders, nobody was the platform. Nobody took a commission. But it was not chaos either. Someone called the count. Everyone knew where to stand. There was a shared understanding of how it was done.

That is the balance this series has been reaching for. Shared rules, shared truth, shared effort, and nobody holding the whole house alone.

The technologies have changed beyond recognition. Invoices travel across a national network in seconds. Rides are matched by protocol. Software agents negotiate with each other on our behalf. But the question underneath is the same one the kampong answered: who carries the load, who decides, and who benefits?

Businesses that answer it deliberately will be more resilient, more independent, and, in the long run, more profitable than those that simply rent whatever road is most convenient today.

Key takeaways from Part 2

Open standards are the most proven form of decentralization. Email, phone networks, and the web all work because the standard is shared and the providers compete.
InvoiceNow is a decentralized network that every GST-registered business in Singapore will be required to join by 2031. It is an opportunity as well as an obligation.
India’s ONDC and Namma Yatri show that an open protocol can shift the pricing logic of an entire market.
Cooperatives work best with similar members, steady demand, and simple technology.
Blockchain is a tool, not a strategy. Even serious crypto projects advise starting simple and decentralizing gradually.
AI agents raise the ownership question again, at the level of decision-making. Open agent standards like A2A make it possible to keep control.
Use the three-question test: lock-in, similarity of participants, and presence of a neutral standard.

Frequently asked questions

What is InvoiceNow in Singapore?

InvoiceNow is Singapore’s nationwide e-invoicing network, based on the international Peppol standard. Businesses send structured invoice data from one accounting system to another through accredited access points. IRAS is extending the GST InvoiceNow requirement in phases so that all GST-registered businesses must use it by April 2031.

How is InvoiceNow decentralized?

InvoiceNow uses the Peppol four-corner model. A supplier and a buyer each connect through their own chosen access point provider, so no single company sits in the middle of every invoice. The invoice format and rules are shared, while the choice of provider and software stays with each business.

What is the A2A protocol for AI agents?

A2A (Agent2Agent) is an open protocol, originally developed by Google and now hosted by the Linux Foundation, that lets AI agents built on different frameworks discover each other, exchange tasks, and collaborate without exposing their internal data or tools. It helps businesses avoid being locked into a single AI vendor.

What is Gall’s Law?

Gall’s Law, from systems theorist John Gall, states that a complex system that works has almost always evolved from a simpler system that worked. For businesses, it means starting any new network, partnership, or automation with a small, working version rather than designing a large, complex system from scratch.

How can a small business reduce dependence on platforms?

A small business can reduce platform dependence by keeping its own unified data, owning its customer relationships and contact permissions, building at least one alternative sales or delivery channel, choosing tools that allow data export, and joining open networks based on shared standards where available.

Mxgsoft has helped Singapore SMEs with digital transformation since 2004, from ERP and InvoiceNow readiness to private AI agents. If you want help deciding which systems your business should own and which it should rent, talk to us.

Missed the beginning? Read Part 1: The Kampong, the Platform, and Who Owns What We Depend On.

References and further reading

IRAS. GST InvoiceNow Requirement.
Thomson Reuters. Singapore regulatory update on GST InvoiceNow extension and grants.
LAFFAZ (1 Apr 2026). Namma Yatri takes its zero-commission model to Europe.
Business News This Week (Sep 2026). Namma Yatri launches Namma Santhe.
Linux Foundation (2025). Launch of the Agent2Agent Protocol Project.
Platform Cooperativism Consortium. When Co-op Principles Go Digital.
The Drivers Cooperative.
ScienceDirect (2022). The feasibility of platform cooperatives in the gig economy.
Infura and DIN Team (2024). Decentralized Infrastructure Network (DIN) Whitepaper [operational draft].
Gall, J. (1975). Systemantics.
Conway, M. (1968). “How Do Committees Invent?” Datamation.

Tags: A2A Protocol AI Agents Business Philosophy Decentralization InvoiceNow Peppol Singapore SMEs

Centralize the Truth, Decentralize the Power (Part 2): Networks, AI Agents, and the Three-Question Test