← Back to Workshop Hub F3 · AGENTSEA IN ACTION

AgentSea in Action

Four demonstrations from the AWS Healthcare Day (14 Jul 2026), rebuilt here as interactive recreations you can play and replay. Across all four, one repeatable pattern: describe the outcome, give the agent the right skills and context, then verify the work.

🎬 Four live demos ▶ Play the simulations 🏥 Healthcare context L100 Foundational

🌊 One platform, one repeatable pattern

Before you build your own agent in the F3 lab, it helps to see what AgentSea looks like in real work. Across four live demonstrations, the platform itself fades into the background — what stays is a single pattern you can reuse for almost any task.

Click each step to explore it, or press auto-play to watch the whole arc.

01 · ORIENT 02 · BUILD 03 · APPLY 🧭 Find your way Terms · guided tour · My Agents vs. the shared Discover space 🛠️ Describe the goal Plain language becomes a system prompt, model & skills Do real work Restyle a keynote · analyse an RFP — then verify the result
🧭
Orient — find your way
Accept the terms, take the guided tour, then find your private agents and discover what colleagues have already shared. Two minutes, and you know where everything lives.

🎬 The four demonstrations behind this walkthrough

Each tab recreates one of four demonstrations given live on 14 July 2026 in the real AgentSea environment (agentsea.synapxe-ai.sg, labelled Beta) at the AWS Healthcare Day. Rather than projector photos, the interface is rebuilt in clean HTML and animated — press play on any tab to watch the interaction unfold.

🧭

First login

Terms, the 7-step tour, My Agents vs. Discover, and the built-in Lifeguard helper.

🛠️

Create an agent

A plain-language goal auto-builds the description, system prompt, model and skills.

📊

The keynote

Rough plain-text slides restyled to match an executive deck — about 80% done, then human polish.

📈

The RFP analysis

A procurement analyst's agent scores every supplier and produces a decision-ready report.

💡
These aren't four separate tricks. They share one pattern — goal + context + skills + human verification — which is exactly what you'll practise in the Guideline Assistant build.

🧭 Your first 60 seconds in AgentSea

On first login you accept the standard Terms of Use, then an optional seven-step tour points out where to create an agent and where to discover what colleagues have shared. Press play to walk the tour, then land on the Discover space.

agentsea.synapxe-ai.sg/agents
AgentSea
My Agents
Discover
Recent chats
Agent Builder
Preview — Executive Deck Builder
Chat with AWS Healthcare Keynote…
Settings & Help
DiscoveryAgentSea (Beta)

Create powerful
AI agents

DESIGN • BUILD • SHARE
+ Create AgentDiscover CommunityFAQ & User Guide
MY AGENTS
📊Executive Deck Builder

Creates polished PowerPoint presentations tailored…

DRAFT🔒 PRIVATE
🏥AWS Healthcare Keynote

Creates professional keynote presentation decks…

DRAFT🔒 PRIVATE
STEP 1 OF 7
Create your agent

Start here to build your first agent — no code needed.

SkipNext

Interactive recreation of the AgentSea landing page and Discover space · demo 1 (video 2489)

🤝 Build privately. Reuse collectively.

What you make stays yours until you choose to share it — and there's already a growing library to borrow from, not a blank slate.

🛡️

My Agents · private

Your working space. Create, test and refine before deciding whether anything should be shared.

🧩

Discover · shared

Browse agents colleagues built and Save a Copy to adapt for your own context.

🌊

AgentSea Lifeguard

Ask for help, give feedback or raise a ticket in plain language. Shortcut: Ctrl/Cmd+M.

📈 Early traction observed
201 public agents already shared across public healthcare — with named authors from Synapxe, NUHS and KTPH. Real reuse, not a blank slate. (Observed 14 Jul 2026.)

🛠️ Describe the outcome. AgentSea builds the configuration.

There's no code and no blank form to fill in. You type a goal in plain language, and AgentSea drafts the whole setup — description, system prompt, model and skills — which you can then edit. Press play to watch one sentence become a working agent.

agentsea.synapxe-ai.sg/agents/create
CREATE AN AGENTSave Agent
💬 Create
CreateConfigure
DESCRIPTION
Creates professional PowerPoint presentations tailored for executive audiences.
SYSTEM PROMPT
You are a professional presentation assistant specializing in executive-level PowerPoint decks. Before creating a presentation, confirm the topic, audience and key points, then use the Document Builder skill to produce a polished PPTX…
987 / 30000
MODEL
Claude Sonnet 4.5
SKILLS ✓ auto-selected
Document Builder
Generate PDF, DOCX, PPTX, XLSX from scratch
Document Insights
Microsoft Exchange
TEST AGENT RESPONSE
📊
PowerPoint Creator
Creates professional PowerPoint presentations tailored for executive audiences.
Create a quarterly business review presentation
Generate slides for an executive strategy meeting
Message PowerPoint Creator…Max 30 MB

Interactive recreation of the Configure tab, auto-generated from one sentence · demo 2 (video 2490)

👥 One platform, two depths

✍️

Create — for everyone

Describe the goal in plain language, let AgentSea propose the setup, and test it on the same screen. This is the F3 starting point.

⚙️

Configure — for power users

Hand-edit the full system prompt, choose skills and privacy, and tune the starter prompts. Same agent, more control.

Skills — what an agent can do (available today)

SkillWhat it does
📄 Document BuilderGenerate professional documents from scratch — PDF, DOCX, PPTX, XLSX
🔍 Document InsightsRead and summarise long documents — contracts, reports, policies, meeting notes
📧 Microsoft ExchangeRead and search email, calendar events and meeting schedules
💡
The model here is Claude Sonnet 4.5 — a signal of the Amazon Bedrock engine underneath AgentSea. Business users never need to think about it; the technical tracks go deeper into Bedrock, AgentCore and Strands.

📊 Use case A · From rough content to an on-brand slide

A real example: the presenter had an executive keynote's base deck, plus a few slides of rough plain-text content. The ask was simple — "match the style, tone and design of the rest of the deck, and keep the flow coherent." Press play to watch Document Builder turn the raw notes into a finished, on-brand slide.

1 Before · raw plain text
The Next Frontier of Human + AI
Human judgement. Agentic execution. System learning.
Human owns
Clinical judgement · Context · Ethical decisions · Accountability
AI provides
Research · Analysis · Documentation · Coordination · Execution support
Healthcare system benefits
Shared learning · Faster innovation · Consistent standards · Population-scale impact
Key statement: the future of public healthcare is not autonomous AI. It is trusted AI working alongside healthcare professionals, within shared governance, platforms and data foundations.
2 After · on-brand slide
Human-plus-AI in practice
From AI architecture to everyday healthcare work
We've built the architecture. The next step is helping every healthcare professional safely use AI — without needing to understand models, infrastructure or prompt engineering.
Access
  • Multiple frontier models
  • Best model for each task
  • Unified experience
Govern
  • Enterprise guardrails
  • Security & compliance
  • Auditability & oversight
Scale
  • Shared prompts & workflows
  • Reusable patterns
  • Organisation-wide learning
One common Human+AI workspace where innovation can scale safely.

Interactive recreation — AgentSea matched the base deck's style, tone and layout (~80%); the presenter did the final polish · demo 3 (video 2492)

🎯
Template-matching, not generic generation. Document Builder followed the recipe of the existing deck rather than inventing its own look — which is exactly why the output was usable. It did roughly 80%; the human made minor font and line edits. That's the right trust posture: a powerful first draft plus human judgement.

📈 Use case B · Rows of supplier responses → a decision-ready view

Everyone works with spreadsheets. A procurement analyst normally scans a huge RFP workbook row by row, isolating the questions that matter and comparing every supplier by hand. Press play: the agent keeps only the sustainability questions, then scores each supplier.

X Event Content.xlsx — RFP responses
RFP questionSP Sus OfficeSP Transf.Si2
Company registration number
Scope 1 & 2 emissions reported?YesYesNo
Years in operation1285
Net-zero commitment + target year?20402045No
Standard payment terms (days)304560
Renewable energy in operations?YesPartNo
Support SLA tierGoldSilverBronze
Overall Sustainability Performance
Category score · higher is better
0.0%
0.0%
0.0%
SP Sus Office
SP Transformation
Si2 Pte Ltd
🧾 Gap analysis
Low scores across the board — the report flagged high-priority gaps for all suppliers on net-zero commitments, renewable energy and circular-economy practices; carbon-measurement coverage was mixed. It ended with a clear recommendation, not just raw data.

Interactive recreation of the filtered workbook and generated chart — real scores 23.8 / 20.8 / 6.2 · demo 4 (video 2495)

⚙️ What the agent does

01

Filter

Keep only the relevant questions via an identifier.

02

Evaluate

Score each supplier's response with evidence.

03

Tabulate

Multi-sheet XLSX: scores, details, gap analysis.

04

Visualise

Editable PowerPoint charts, ready to present.

⏱️
Built by a non-technical user in 5–10 minutes. The analyst created this agent herself and now uses it routinely. Treat that as her reported experience, not a universal performance promise — but it shows how low the barrier has become.

🛡️ Fast is useful. Verified is safe.

Every one of these outcomes was a strong first draft — not a signed-off deliverable. AgentSea makes that boundary explicit. This line sits in the footer of the live chat:

"AgentSea handles up to Restricted, Sensitive (Normal) information. Always verify responses."

It has two parts. The classification boundary tells you what the platform is approved to handle. "Always verify responses" defines your job: a human stays in the loop.

Three checks before you rely on any output

🔎

Check the source

Can the answer point to the guideline section or the uploaded evidence it used?

Check the output

Review the facts, calculations, tone and formatting before the deliverable goes anywhere.

⚠️

Escalate uncertainty

If the source doesn't cover it, the honest answer is "not covered" — never an invented one.

🏥
Why this matters in healthcare admin: a confidently wrong answer about leave, claims or procurement can cause real harm. Grounding plus human verification is what makes an agent safe to rely on — and it's exactly what you'll build in the next step.

🚀 Now build one yourself

You've seen the pattern four times over: orient in the platform, describe what you want, let AgentSea build the configuration, then apply it to real work and verify the result. Next you'll run that same loop on a document everyone deals with — an internal guideline.

The Guideline Assistant — in four steps

STEP 01

Configure

Give the agent a system prompt: role, grounding rules and "not covered" behaviour.

STEP 02

Add the skill

Switch on Document Analysis so it can read the guideline you upload.

STEP 03

Ground & test

Ask one covered question and one deliberately uncovered one.

STEP 04

Briefing

Turn the answer into a plain-language, brand-compliant document.

🎯
Success looks like: a cited answer, an honest "not covered" when the guideline is silent, and a polished briefing — the same goal + context + skills + verification loop you just watched.
🌊 Start the Guideline Assistant build →

Prefer the projected version for a live session? The instructor slide deck covers this same story.