Agentic AI · Ipsos Client Workshop01

The agent has
entered the building.

What autonomous AI agents are, why they spread faster than any open-source software before them, and what a research and marketing leader should do about it. A hands-on briefing by BASIC for the Ipsos client workshop series.
The next thirty minutes
Part 01
The shift
Chatbots became agents
Part 02
The evidence
Adoption without permission
Part 03
Hands on
Two live agents, on your phone
Part 04
The playbook
Govern it in ninety days

Agentic AI Workshopaiagent.research.my
Agentic AI · Ipsos Client Workshop02
•The Shift

A chatbot answers. An agent acts.

The word "agent" is doing real work here: this is software that pursues a goal, not software that completes a sentence.

What you have used
The chatbot

You ask, it answers, it forgets. It waits for you, gives one reply at a time, and touches nothing: no files, no tools, no follow-through. Every task starts again from zero.
What arrived next
The agent

You set a goal, it works. It plans the steps, operates real tools (browser, files, email, databases), checks its own output, remembers what it learned, and keeps going while you sleep.
A working definition: an AI agent is a model given tools, memory and permission to act toward a goal without step-by-step supervision.
·Agentic AI Workshopaiagent.research.my
Agentic AI · Ipsos Client Workshop03
•The Shift

Six things changed in barely twelve months.

None of these is science fiction. Each shipped, publicly, between late 2025 and now.

01

Tool use

Models learned to operate software: run commands, browse, query databases, send messages.
02

Memory

Agents keep context across days and tasks. They know your projects, your style, your data.
03

Long-running loops

Work continues unattended. The agent plans, acts, checks its result, and tries again.
04

Messaging as interface

The agent lives in WhatsApp and Telegram, not in yet another app your team must learn.
05

Prices collapsed

Budget models now run junior-analyst work at 1 to 3 percent of frontier token prices.
06

Open source

Complete agent stacks are free on GitHub and run on a laptop. No vendor required.
·Agentic AI Workshopaiagent.research.my
Agentic AI · Ipsos Client Workshop04
•The Evidence

This is not a vendor slide. It is a measurement.

Two open-source personal agents, built outside every major AI lab, produced these numbers in under nine months. Every figure below is dated and sourced.

386K

GitHub stars, one project

OpenClaw, shipped 24 November 2025 as one developer's weekend project, counted 386,400 stars by 16 August 2026: one of the fastest-growing open-source projects in history.
36.4T

Tokens in thirty days

Hermes Agent's volume on OpenRouter in the 30 days to 16 August 2026, holding the number one global rank among all AI applications.
Feb 26

Machines overtook humans

The month agent-driven traffic passed human traffic on OpenRouter, at roughly 15 times the tokens per request.
<9

Months old, the whole category

From first release to trillions of tokens a month, with no marketing budget and no sales team.
Tap any figure for the full source list.
·Agentic AI Workshopaiagent.research.my
Agentic AI · Ipsos Client Workshop05
•The Evidence

It did not arrive through IT. It arrived through WhatsApp.

Every previous enterprise software wave came top-down: procurement, licences, rollout. This one is bottom-up, and that changes what "adoption" means.

01

A developer ships it free

Open source, November 2025. No licence, no sales team.
→
02

Your analyst installs it

One evening, on a laptop. No procurement, no ticket, no approval.
→
03

It plugs into their life

Email, files, chats, calendar, a personal API key.
→
04

It touches company data

Ungoverned, unlogged, invisible to the organisation.
The uncomfortable question for this room: if your team includes anyone technical, the realistic assumption is that an agent is already inside the building. And in Malaysia, where WhatsApp reaches 90.7 percent of internet users, the interface was already installed.
·Agentic AI Workshopaiagent.research.my
Agentic AI · Ipsos Client Workshop06
•How It Works

The loop is the product. Six steps, repeated.

Strip away the hype and every agent is the same simple machine. Understanding this loop is all the technical depth a leader needs.

01

Goal

"Code these 2,000 open ends by tonight."
→
02

Plan

Break the goal into steps it can execute.
→
03

Tools

Files, browser, database, code, email.
→
04

Act

Do the step. Really do it, not describe it.
→
05

Check

Inspect its own result. Retry if wrong.
→
06

Remember

Write what it learned to memory for next time.
The double edge: the same autonomy that finishes the job unattended is the autonomy that makes mistakes at speed. That is why the second half of this session is about governance.
·Agentic AI Workshopaiagent.research.my
Agentic AI · Ipsos Client Workshop07
•Scenario 01

Always-on brand and competitor intelligence.

The Monday-morning competitive scramble, retired.

Today
The Monday scramble

A junior analyst spends Monday morning pulling competitor prices, campaign launches, review scores and social chatter into a deck nobody reads until Tuesday. Coverage is whatever one person could check before the meeting.
With an agent
The Sunday-night sweep

An agent sweeps the same sources every night: pricing pages, campaigns, ratings, store listings. It drafts the digest, flags what moved and why it matters, and delivers it to the team's WhatsApp before the 9am meeting, with every source logged.
What changes: the analyst stops being the collector and becomes the editor. Coverage goes from "what we had time for" to "everything, nightly".
·Agentic AI Workshopaiagent.research.my
Agentic AI · Ipsos Client Workshop08
•Scenario 02

The research production layer, compressed.

The honest version for a room of research buyers: agents do not replace insight. They compress the hours between fieldwork and insight.

Overnight

Open-end coding

Thousands of verbatims themed, counted and quoted by morning, with a human reviewing the codeframe instead of building it.
Same day

Transcript synthesis

Eight focus groups in, one structured findings document out, every claim traceable to the transcript line it came from.
Before launch

Questionnaire QA

Routing logic, translations and quota maths checked by an agent that never gets bored on question 47.
First draft

Report drafting

The agent drafts the data story overnight. Your senior team spends its time on the "so what", not the chart formatting.
Where the value moves: as production compresses, judgment, design and the client conversation rise in value. The winners will be the teams whose analysts direct fleets of agents.
·Agentic AI Workshopaiagent.research.my
Agentic AI · Ipsos Client Workshop09
•Scenario 03

A chief of staff in your pocket.

The scenario everyone in this room can start personally, this quarter, without a committee.

Morning

Inbox triage

Overnight email sorted into "needs you", "needs a reply I drafted", and "noise", waiting on WhatsApp when you wake.
Before meetings

Meeting prep

One page per meeting: who, the history, the open items, the numbers you will be asked about.
On demand

Brief drafting

Message it a voice note in the car; a structured brief is in your inbox before you park.
Always

Follow-through

It remembers what you promised, to whom, by when, and nags you the way a good chief of staff does.
Why start here: it teaches you, personally, what agents can and cannot do, before you make decisions about them for your whole organisation.
·Agentic AI Workshopaiagent.research.my
Agentic AI · Ipsos Client Workshop10
•Hands On

Stop listening to me. Message one yourself.

Two live agents, on WhatsApp, right now. Same tools, same data, very different brains. Scan, say hello, and ask about Malaysia.

Frontier model · the heavyweight
Agent GAJAH
QR code for Agent Gajah on WhatsApp

aiagent.research.my/wa1

Deep reasoning, careful analysis, frontier-class pricing. Ask it the hard, multi-step questions.
Budget model · the lightweight
Agent KANCIL
QR code for Agent Kancil on WhatsApp

aiagent.research.my/wa2

Fast, cheap, surprisingly capable. Ask it the same questions and feel the difference for yourself.
Try: "Compare median household income in Johor Bahru and Shah Alam" · "How many cafes are in Penang?" · "Which Selangor district has the highest spending power?"  Replies take 30 to 60 seconds. Both agents are sandboxed and query public Malaysian data only.
·Agentic AI Workshopaiagent.research.my
Agentic AI · Ipsos Client Workshop11
•Hands On

What you are actually querying: an agent is only as good as its data.

The demo agents are not browsing the open web. They are querying structured data assets, the same way an agent inside your business would query yours.

Official statistics

OpenDOSM, normalised

Malaysia's official statistics, made query-ready
612K

population fact rows, plus income, poverty and Gini for 999 geographies down to DUN level

 Business census

Southeast Asia, mapped

Geocoded business listings across eight markets
17.2M

businesses, 1.9M of them Malaysian, with 27.9M contact records

 Property

Live listings pipeline

Continuous capture from Malaysian property portals
Live

listings, prices and media refreshed continuously
🛡Hosted in Malaysia on infrastructure BASIC controls  ·  the demo never sends your questions to a third-party data broker.
·Agentic AI Workshopaiagent.research.my
Agentic AI · Ipsos Client Workshop12
•The Honest Part

Now the slide the vendors skip.

Everything that makes an agent useful, the tools, the memory, the autonomy, is also the attack surface. Four risks, all real, all documented, all with dates.

Risk 01

Prompt injection

One crafted email was enough: the zero-click EchoLeak flaw made a production enterprise copilot leak private data with no user interaction. Patched, never exploited, proven possible.
Risk 02

Credential exposure

Scans found 40,214 self-hosted agent instances exposed on the open internet in February 2026, 12,812 of them open to remote code execution.
Risk 03

Shadow agents

One in five breached organisations traced the breach to unsanctioned AI, at an average of USD 670,000 in extra cost (IBM, 2025).
Risk 04

No audit trail

63 percent of breached companies had no AI governance policy at all. By May 2026, six national cyber agencies had issued joint guidance demanding full traceability of agent actions.
⚠None of this argues for a ban. It argues for governance. A ban does not remove the agents, it just keeps them invisible.
·Agentic AI Workshopaiagent.research.my
Agentic AI · Ipsos Client Workshop13
•The Playbook

What a governed agent looks like.

Six rules. The two agents you just messaged run under all of them, which is why we were comfortable handing them to a room full of strangers.

01

Sandboxed

Agents run on isolated infrastructure, never on someone's personal laptop.
02

Least privilege

Read-only data, a whitelist of tools, nothing it does not need.
03

Approval gates

A human approves anything irreversible: send, publish, pay, delete.
04

Logged

Every message, tool call and data access recorded and reviewable.
05

Right-sized models

Frontier brains for hard judgment, budget brains for volume work. You just felt the difference.
06

Data stays home

Company and client data stays in your jurisdiction, on infrastructure you control.
·Agentic AI Workshopaiagent.research.my
Agentic AI · Ipsos Client Workshop14
•The Playbook

The first ninety days: see, pilot, decide.

The goal is not transformation. It is an informed decision, made with your own evidence instead of a vendor's.

01
Weeks 01 to 02

See

Inventory what is already running in your teams, amnesty included. Write the two-page policy: what is allowed, what is banned, what needs approval.
02
Weeks 03 to 08

Pilot

One governed pilot, one workflow, real data. Pick a production task you measured in hours, run it under the six rules, and keep the humans in the loop.
03
Weeks 09 to 12

Decide

Measure hours saved, quality, and incidents. Scale what worked, kill what did not, and publish the decision internally so the shadow agents come into the light.
·Agentic AI Workshopaiagent.research.my
Agentic AI · Ipsos Client Workshop15

Do not ask whether your company
will use agents. Ask who is governing
the ones it already has.

·Agentic AI Workshopaiagent.research.my
Agentic AI · Ipsos Client Workshop16
Thank you

Keep the agents.
Keep them governed.

The demo agents stay live for a week. The brief below carries everything from today, sources included.

This workshop
aiagent.research.my
Enquiries
info@basicinsight.my
Web
basicinsight.my
Download the brief A4 · print-ready

Agentic AI Workshopaiagent.research.my
01 / 16

Keyboard shortcuts

Arrows / Space
Previous / next slide
Home / End
First / last slide
1 to 16, Enter
Jump to slide
F
Fullscreen
I
Toggle Layer 2 notes
O
Overview grid
?
This help
Esc
Close any overlay