AgentSea: Singapore Healthcare AI Agents, Explained

On 24 August 2026, at the HIMSS Global Health Conference at Marina Bay Sands, Senior Minister of State Tan Kiat How announced that every professional in Singapore's public healthcare sector can now create their own AI agents. The tool is called AgentSea, built by Synapxe, Singapore's national healthtech agency, together with Amazon Web Services.

An AI agent is a step beyond the chatbots most of us know. A chatbot answers questions. An agent carries out a task: summarising records, drafting a roster, reviewing a procurement document. AgentSea lets staff describe the task in plain English, and the platform builds a working agent around it, with no coding required.

In just over two months since the platform launched, healthcare workers have already created more than 12,000 agents. That adoption rate is the real story, and our NDR 2026 AI explainer explains why the government is betting on AI agents like these.

Think of AgentSea as a builder for small automations, except each one is a chatbot wired to a specific job. A nurse opens the platform and types something like "build me an agent that turns shift notes into a handover summary". The system asks a few questions, attaches the right tools, and proposes a working agent. The user tests it on sample inputs, tweaks the instructions, and only then puts it to use.

None of this requires programming. The skills needed are the same ones you use to write a careful email: being clear about what you want, and checking the output before you trust it. Synapxe calls the review step "Test Agent Response", and it is exactly the habit every AI beginner should build, whether they work in a hospital or anywhere else.

AgentSea is open to the more than 80,000 professionals in Singapore's public health sector, across the three healthcare clusters. It is not a public app. But the idea behind it, that a person with no technical training can build a small automation for their own job, is something anyone can try at home with a free chatbot.

Two examples show the range. At KK Women's and Children's Hospital, staff built an agent that generates daily rosters. It cut down data entry and removed the transcription errors that came with typing schedules by hand.

At the National Heart Centre Singapore, Dr Kenneth Michael Chew, a cardiology consultant, built a pre-clerking agent with no coding experience. Pre-clerking is the preparation before a consultation: reviewing patient histories, flagging relevant clinical details, pulling records together. It is slow, careful work. His agent does the first pass, and he checks the output before anything clinical. As he put it, the agent has "not replaced the need for verification or my own clinical assessment", but it made his preparation more efficient.

That last part is the pattern to notice. Everywhere in this rollout, the human stays in the loop. The agent drafts, the professional verifies.

Healthcare data is the most sensitive data there is, and Synapxe designed AgentSea around that. Sensitive information stays inside the public healthcare network and is not shared externally. The platform screens what users feed in: it can detect identification numbers such as NRIC and financial credentials, and block them from being registered into an agent in the first place. Requests flagged by the safety controls can be refused outright. Everything runs with audit logging, aligned with Singapore's national agentic AI governance standards.

For the rest of us, this is a reminder worth keeping. When you paste documents into a free AI tool, that data leaves your device. Before you feed anything containing an NRIC, a bank account number, or a child's name into any AI tool, scrub it first. The same instinct that protects patients should protect you.

Healthcare is one of the four national AI missions PM Wong set out this year, and AgentSea shows what that mission looks like in practice. Tan Kiat How put the logic simply: AI should make the health system more productive and more resilient, so care teams can intervene earlier and spend time on patients instead of paperwork.

The numbers suggest staff agree. More than 12,000 agents in two months is fast adoption by any measure, and over 300 of them have been shared so colleagues can reuse and adapt them. For patients, the payoff is indirect but real: less time spent on rosters and record-pulling means more time on care.

To see who runs AI in Singapore, our AI agencies map covers Synapxe and every other government body. The State of AI in Singapore report tracks these rollouts quarter by quarter.

You do not need AgentSea to start building small automations. A free chatbot can handle simplified versions of the same jobs, as long as you write clear instructions and check the output. These prompts follow the same pattern as the hospital agents.

💼 Report and roster help

I want a reusable "mini agent" for my job. My recurring task is [describe the task, e.g. compiling a weekly sales report from five spreadsheets]. Write me a set of step-by-step instructions I can paste at the start of each session: what to ask me, what to do with the data, and what format to output. Keep it under 200 words so I can reuse it every week.

📋 Summarise, then verify

Summarise this document in 5 bullet points: [paste document]. Then list 3 things I must double-check against the original, such as numbers, dates or names. Do not invent anything that is not in the source.

🛡️ Scrub personal data first

Here is a draft document: [paste]. Replace any NRIC numbers, phone numbers, bank account numbers, home addresses and full names of private individuals with placeholders like [NRIC] or [NAME]. Keep everything else exactly the same. Show me what you changed.

✅ Test before you trust

You produced this output: [paste output]. Now act as my quality checker. Compare it against the source material [paste source] and list every place where it is wrong, incomplete or invented. Be blunt. Do not defend your earlier answer.

The hospital version runs on stricter rails: approved tools, blocked identifiers, audit logs. The habit is the same. Describe the task clearly, test the output, keep the human in charge. That is the whole trick, and it works with a free chatbot today.