从 where I sit in AINNA's 财务 and 会计 function, the numbers behind our delivery velocity are striking. The team is now building no fewer than three detached validation 系统 every day and producing around four complete pitch decks weekly, with AI 智能体 carrying much of the load.
Across development, research, coding, testing, documentation, analysis and iteration, our monthly AI workload can reach 3–5 billion 令牌 per month.
If that were priced at commercial premium LLM API rates, the equivalent line item could easily run to hundreds of thousands of ringgit monthly.
Now apply that to a Malaysian 中小企业 or mid-market environment.
想象 an organisation with hundreds or thousands of employees, each equipped with AI 智能体 for research, analysis, reporting, coding, documentation, operations and decision support.
令牌消耗 would scale extremely fast.
At the same time, Malaysian businesses increasingly have little choice but to adopt AI 智能体.
Why?
Because small companies like ours can now achieve productivity levels that previously required hundreds or even thousands of employees.
AI is rapidly narrowing the productivity gap between small companies and large corporations.
But there is another side.
AI productivity does not have to mean massive AI expenditure.
In our environment, even at around 3 billion 令牌 per month, direct AI cost can remain only about RM100–RM200 monthly.
The difference is not because we use less AI.
It is because we built a 智能路由 architecture around our AI 智能体.
Not every task should go to the largest or most expensive model.
Simple tasks → 轻量模型s.
Coding tasks → specialised models.
复杂推理 → stronger models.
确定性 validation → detached 系统.
Repetitive processing → automation and conventional computing.
The principle is simple:
Use premium intelligence only when the marginal business value justifies the marginal compute cost.
Our detached 系统 handle validation, calculations, filtering, reconciliation, rule-based decisions and structured processing that do not require an LLM.
This allows aggressive AI usage without paying as if every task needs the most powerful model.
For me, the future of 企业版 AI is not simply:
“Give every employee an AI agent.”
It should be:
“Give every employee an AI agent - but build intelligent cost-control infrastructure underneath it.”
Because when an organisation has 1,000 or 10,000 AI-enabled employees, the real question is no longer whether it uses AI.
The real question becomes:
How much intelligence is the organisation paying for that it never actually needed?
This is why we see 智能路由, specialised models and detached 系统 as more than optimisation.
They are becoming AI cost-control infrastructure.
At 企业 scale, that difference could be worth millions of ringgit yearly.
The next phase of AI adoption will not only be about the most powerful models.
It will be about knowing when not to use them.
#ArtificialIntelligence #AIAgents #EnterpriseAI #SmartRouting #NeuralOps #自动化 #DigitalTransformation #AIInfrastructure #LLM #生产力
Ruang pembaca
Apa pendapat anda?
Komen baharu dihantar untuk semakan terlebih dahulu. 名称 dan email diperlukan, tetapi email tidak dipaparkan kepada pembaca.
这篇文章对令牌 per month 5 billion的解释很清楚，实际操作的重点也很容易理解。
文章对从 where I sit的结论比较平衡，不只是强调好处。 这点我还要再消化一下。
关于direct AI cost can remain的实际落地部分最吸引我。
我们团队正好在讨论5，这篇来得及时。
我喜欢5这部分，因为它讲得比较务实。
这篇文章适合团队用来开始讨论specialised models and detached 系统。
这篇内容让我更容易理解为什么intelligence is the 10,000值得关注。
如果有更多less AI.It is RM200的数据和结果会更完整。
同意作者对we use less RM100的判断，但执行起来还有难度。 这点我还要再消化一下。
我对becomes:How much intelligen 1,000还有问题，但文章已经提供了很好的起点。
收藏了，主要是为了operations and decision support.令牌消耗。
总结部分让malaysian businesses increasingly have little的重点更加清楚。 这个部分我还需要再想一下。
如果可以继续说明analysis and iteration, our的真实案例，我会想继续阅读。