Must a company pursue AGI because technology hype is increasing? No. Indonesian companies should start with an important workflow, trusted data, and decisions that people still own.
Digital systems, data, AI, and agentic AI can exist at the same time in 1 company. AGI is a research context, not a company readiness level. This article does not say that 1 capability must finish before another begins.
Also read how to connect separate data to an agentic workflow.
Do not make AGI a shopping list
Digital systems record work. Data makes work facts available. AI helps create or assess outputs.
Agentic AI can choose steps and tools within stated limits. AGI is a research term for general capability across tasks. Its capability level and autonomy remain under discussion.
A company does not need to place all data in 1 physical database. Use consistent identities, a source of truth for each fact, access rights, and recorded data exchange. Sales can be ready for AI summaries while the warehouse still needs to fix stock codes.
What the current evidence says
On 6 August 2026, AWS reported that 40% of Indonesian businesses used AI. It also said 38% had heard of agentic AI, 56% of adopters were experimenting, and 12% had fully integrated AI.
Strand Partners surveyed 1,000 business leaders and 1,000 members of the public for that vendor report. This vendor survey is not a census and does not prove causation.
Microsoft Indonesia released a useful counterpoint in June 2026. The release describes advanced AI users and human review practices. It does not provide an Indonesian sample size that supports a national percentage.
Telkom also announced Agentic AI by BigBox in April 2026. A product announcement shows industry activity. It does not prove national adoption or a result for a specific customer.
AGI remains a measurement problem
Google DeepMind separates breadth of capability, level of capability, and autonomy. These dimensions can change independently. A system that performs well on 1 task is not automatically broad, autonomous, or safe for every task.
The March 2026 DeepMind framework discusses cognitive capability measurement against human capability. The framework is not an announcement that AGI has arrived. These 2 sources do not set 1 date when all companies enter the AGI era.
For this reason, do not wait for a stable term before making sound decisions. Assess the work available today. State which actions an AI can read, propose, receive approval for, or execute.
A readiness map for each division
This table is a Rama Digital diagnostic tool. It is not an industry standard or certification. Assess each division with reviewable work evidence, not with the number of AI accounts it has purchased.
| Position | Work evidence | Safe next action |
|---|---|---|
| Recorded | The process, owner, and result are recorded. | Choose 1 repeated problem and measure its starting work time. |
| Connected | Data has a source of truth, identity, and clear access. | Connect the data needed for the process, not all data. |
| AI-assisted | The team uses AI for drafts, search, or classification with review. | Keep input, output, and error examples for evaluation. |
| Agentic with controls | An agent runs limited steps with logs, access limits, and approval. | Start with read or proposal actions before execution actions. |
| Ready to learn further | An owner measures quality, risk, change cost, and the next decision. | Expand only when process evidence supports it. |
1 company can occupy several positions at once. Finance may be recorded. Customer service may already be AI-assisted.
Do not force a division with unclean data to use an agent with broad access. Prove readiness from work evidence, not from the term used.
Start at recorded when the process and owner cannot yet be shown. Move to connected after important facts have a source of truth and stated access. Use AI for drafts or classification before you let an agent take action.
Expand an agent only after an owner reviews logs and trial results.
Diagnostic questions before choosing a tool
- Which repeated process makes customers or staff wait?
- Who owns the process and who can approve a change?
- Which facts are the source of truth for that process?
- Where do duplicate, missing, or inconsistent identities appear?
- Which actions can an AI read or propose, and which actions require human approval?
- Which errors must never occur?
- What evidence shows that an AI output helps or harms the process?
- Who reviews logs and corrects the rules?
- When will the team decide to continue, improve, or stop the trial?
An unavailable answer is a diagnostic finding, not a failure. Record that gap before buying an integration or giving an agent access.
If you cannot yet name 1 starting process, begin with AI Diagnostic. This session maps bottlenecks, AI opportunities, data needs, and a written initial action.
Signs that the process is still stuck
Check 4 signs before you expand AI. Staff often enter the same data in 2 places. Reconciliation still happens at day-end or month-end.
The source of a number remains unknown when results differ.
Also check the dependency on 1 person. If 1 person is the only person who knows why an order status changed, the process is not ready for more autonomy. Write rules, an owner, and a return-to-manual path when an AI gives a wrong output.
A return-to-manual path lets a worker return to a known manual process when the system fails. Keep a record of the issue, event time, and repair decision. Do not use a new trial to hide a failure that you do not yet understand.
Illustrative example: a distributor with separate data
This example is illustrative. It is not a client story. A distributor stores stock in a warehouse system, orders in a sales application, quotations in spreadsheets, and invoices in accounting software.
The team starts with a read-only agent that lists orders at risk of delay. A supervisor checks the list before contacting customers.
The next step is not to give the agent rights to change stock or send invoices. The team initially checks product codes, order status, data ownership, and recommendation logs. After quality evidence exists, the team can test follow-up proposals that still need human approval.
If a core process is visible but its work evidence is scattered, use AI Workflow Audit. The audit helps select a use case from effort, risk, impact, and workflow ownership.
If staff use AI separately without data rules and review, direct the team to Pelatihan AI untuk Tim Perusahaan. The training sets division practice, data rules, and human review.
If owners have selected several use cases but cannot set their order, use AI Implementation Roadmap. The roadmap sets priorities, resources, governance, and 30–90 day milestones.
Set a 90-day decision, not an AGI promise
If cross-division conditions remain unclear or the need does not fit 1 route, start with AI Consulting. We help choose a work route after we review the problem, process, and current readiness.
This chapter helps you choose a starting point. Continue to the 90-day company AI transformation plan to set owners, sequence, and decision criteria.
Return to the chapter on separate data and agentic workflows if your main issue is data sources and approval.
Frequently asked questions
Must a company wait for AGI before it uses AI?
No. Start with a process that you can review today. Limit access, use human review, and assess quality evidence before you expand use.
Must all data enter 1 database before the company uses AI?
No. Set a source of truth, consistent identities, access, and data exchange for the chosen process. 1 physical database is not a general requirement.
Which division should start first?
Start with a division that has a repeated process, a clear owner, reviewable data, and controlled risk. Do not start with the broadest access.
