A few months ago, I started reviewing our AI expenditure the same way I review any other operating asset: if AI is now part of daily operations, why should it remain an external service line item rather than an internal capability on our own infrastructure? From a finance and accounting standpoint, I wanted AI under our direct control-inside our servers, aligned with our systems, and governed by our own policies.
That analysis led us to deploy AI agents directly on our own servers. In the current AINNA environment we operate three distinct AI agents, each configured around specific finance, operations, and system-development requirements. In practical terms, every server we run now has its own AI-powered Pengurus IT, Server Administrator, and Pembangun working continuously-monitoring, maintaining, troubleshooting, and extending systems without overtime or per-seat licences.
Together, these agents have delivered more than 250 detached systems to date. We are also building two parallel environments for Bahasa Melayu and Chinese-language domains, each with distinct knowledge bases, workflows, and compliance requirements that map to different customer segments.
The detached-system model matters financially because it converts recurring API calls into owned, amortised functionality. Once a workflow, parser, automation, or process has been properly designed, it can run independently, while the AI agen remains focused on higher-level development, supervision, exception handling, and decision support. This reclassifies part of our AI spend from a variable cost into a controlled, depreciable internal asset.
Our preferred architecture is Manusia → Peribadi Ejen AI → Sistem Berasingan → Servers / Databases / APIs / MCP / Aplikasi.
This stack gives us greater control over data residency, privacy, permissions, memory, model selection, workflows, deployment, operating cost, and system integration-each of which has a direct line-item impact on jumlah cost of ownership and audit readiness.
MCP, cloud AI, APIs, and third-party platforms still have a place in our toolset. We prefer them to be optional services our infrastructure can consume, rather than structural dependencies that define the infrastructure. For AINNA, the next stage of AI is not conversational novelty; it is about building AI that can operate, maintain, develop, and automate real systems continuously while progressively reducing unnecessary external dependency. That shift is especially relevant for Malaysian PKS that need to control operating costs and keep data within local boundaries.
#ArtificialIntelligence #AIAgents #NeuralOps #DetachedSystems #AIInfrastructure #LocalAI #Automasi #MCP #SystemDevelopment #DigitalTransformation
