MANIFESTO
Software 3.0: why AI agents change the very nature of ERP.
A thesis in four acts.
ACT I
Software 1.0 — Code written by humans.
For fifty years, enterprise software was a frozen artifact: lines of code written by engineers, configured by consultants, operated by business users with neither the language nor the permissions to modify it.
ERPs are the culmination of that era: ultra-powerful, ultra-rigid, ultra-costly to evolve. Every deviation from the standard is paid for in integrator man-days. Every regulatory change becomes a project.
That structural dependency between the business and its integrators is neither a bug nor a fatality. It is the business model of traditional enterprise software.
ACT II
Software 2.0 — Code written by statistical learning.
The arrival of machine learning shifted the line: some software behaviors were no longer explicitly written, but learned from data. The first gains came — recommendation engines, anomaly detection, OCR.
But those capabilities stayed siloed: each model served a narrow function, with no ability to reason about its own context or to act. ERPs gained only peripheral features.
ACT III
Software 3.0 — Systems written by agents interacting through natural language.
The latest generation of generative models introduces a break: an agent can now understand spoken instructions, reason about a complex business context, plan a sequence of actions, and execute those actions in existing information systems.
The unit of programming is no longer a line of code. It is a conversation. The business user becomes a programmer — without realising it. The ERP, once rigid, becomes an interlocutor to instruct.
This is what we call Software 3.0 — and it is the only framework that reconciles the rigor of large IT with the agility of small teams.
“The ERP stops being software to operate. It becomes an interlocutor to instruct.”
Richard YI · KHMERP
ACT IV
The AI Factory — Why orchestrating multiple agents beats a single monolithic one.
No generalist agent today is better than a well-orchestrated combination of five specialists. One excels at reasoning, another at desktop action, another at multimodal voice. The whole is greater than the sum.
That is why we operate as a factory: an industrial production workshop, where each agent is selected for its specialty, embedded in a repeatable process, and deployed under a six-phase methodology.
A factory, not a demo.