WEBVTT

1
00:00:08.570 --> 00:00:12.315
Agentic AI has become
the buzzword of the moment.

2
00:00:12.672 --> 00:00:17.115
Every major technology vendor is racing
to position their products as the gateway

3
00:00:17.115 --> 00:00:23.455
to an autonomous, decision-making future:
AI that doesn't just assist, it acts.

4
00:00:24.292 --> 00:00:26.556
The vision is compelling.

5
00:00:26.913 --> 00:00:31.841
But the most mature organisations
aren't focused on agentic AI.

6
00:00:32.197 --> 00:00:35.861
They're focused on automation
and integration.

7
00:00:38.081 --> 00:00:41.858
The difference is clearest
in a painfully simple framework,

8
00:00:41.858 --> 00:00:44.258
and the best frameworks always are.

9
00:00:44.615 --> 00:00:48.646
It maps AI solutions
across two dimensions.

10
00:00:49.003 --> 00:00:50.320
Integration:

11
00:00:50.320 --> 00:00:54.786
how deeply a solution connects
to your existing systems and data.

12
00:00:55.143 --> 00:00:59.676
And automation: how much it
can do without human intervention.

13
00:01:00.513 --> 00:01:03.257
That creates four distinct categories:

14
00:01:03.257 --> 00:01:07.695
assistants, copilots,
autopilots and agents.

15
00:01:08.052 --> 00:01:12.640
Without automation and integration,
there are no agents.

16
00:01:13.860 --> 00:01:19.494
Assistants sit in the bottom-left corner:
low integration, low automation.

17
00:01:19.851 --> 00:01:22.133
They're the simplest form of AI,

18
00:01:22.133 --> 00:01:25.715
like a chatbot that answers
questions from a knowledge base.

19
00:01:26.552 --> 00:01:30.692
Many organisations
dismiss this as too basic.

20
00:01:31.048 --> 00:01:34.146
But this is where users
learn to interact with AI,

21
00:01:34.146 --> 00:01:36.701
where you learn what people
actually ask about,

22
00:01:36.701 --> 00:01:40.745
and where you discover gaps in your
knowledge content you never knew existed.

23
00:01:41.582 --> 00:01:43.968
This step can't be skipped.

24
00:01:44.325 --> 00:01:46.947
It builds the conversational
design muscles

25
00:01:46.947 --> 00:01:50.570
that every subsequent AI
initiative will depend on.

26
00:01:51.790 --> 00:01:57.198
Move up the integration axis, and you
arrive at copilots: high integration,

27
00:01:57.198 --> 00:01:59.177
but still low automation.

28
00:01:59.533 --> 00:02:03.392
They work alongside people,
inside their existing tools,

29
00:02:03.392 --> 00:02:06.906
like Copilot suggesting
email replies in Outlook.

30
00:02:07.263 --> 00:02:12.150
The AI suggests, and the human
decides what happens next.

31
00:02:12.987 --> 00:02:15.323
Autopilots are high automation,

32
00:02:15.323 --> 00:02:20.228
low integration, like a chatbot that
guides users through resetting a password,

33
00:02:20.228 --> 00:02:23.566
without querying multiple
systems behind the scenes.

34
00:02:24.403 --> 00:02:29.004
And then there's the top-right
quadrant: agents, the Holy Grail.

35
00:02:29.361 --> 00:02:33.242
They combine high integration
and high automation.

36
00:02:33.599 --> 00:02:38.884
They take context from multiple systems,
reason about it, and act on their own,

37
00:02:38.884 --> 00:02:41.610
like handling an order refund end to end.

38
00:02:42.830 --> 00:02:44.945
It sounds incredible.

39
00:02:45.302 --> 00:02:48.001
And it is, when it works.

40
00:02:48.358 --> 00:02:51.586
But agents are the hardest place to start.

41
00:02:52.423 --> 00:02:55.929
They need deep integration
across systems and data,

42
00:02:55.929 --> 00:02:59.212
and automation that handles
edge cases gracefully.

43
00:02:59.569 --> 00:03:02.159
Each is challenging on its own.

44
00:03:02.516 --> 00:03:06.777
Together, they multiply
the difficulty, and the risk.

45
00:03:07.614 --> 00:03:10.667
The integration work can take months.

46
00:03:11.023 --> 00:03:14.610
Edge cases that seemed trivial
in planning become black holes

47
00:03:14.610 --> 00:03:17.038
that consume development resources.

48
00:03:17.395 --> 00:03:20.085
And the project never
sees the light of day,

49
00:03:20.085 --> 00:03:25.325
or arrives so late and so descoped that
stakeholder confidence has evaporated.

50
00:03:28.045 --> 00:03:33.013
So start in the bottom-left quadrant,
and work your way up deliberately.

51
00:03:33.370 --> 00:03:37.591
Deploy an assistant,
and learn what users ask about.

52
00:03:37.948 --> 00:03:41.883
Add integration,
and move towards copilots.

53
00:03:42.240 --> 00:03:47.792
Add fulfilment to the well-understood
requests, and move towards autopilots.

54
00:03:48.631 --> 00:03:51.357
All of this capability stacks.

55
00:03:51.712 --> 00:03:53.383
The conversational flows you

56
00:03:53.383 --> 00:03:57.753
design for your assistant become
the foundation for your copilots.

57
00:03:58.110 --> 00:04:02.146
The automations you build for routine
requests become the building blocks

58
00:04:02.146 --> 00:04:04.355
for more complex orchestration.

59
00:04:05.192 --> 00:04:09.739
You're not delaying your journey
to agentic AI by starting simple.

60
00:04:10.096 --> 00:04:12.931
You're building the road you'll travel on.
