1
00:00:00,05 --> 00:00:02,03
- [Laurence] It's time to write some code.

2
00:00:02,03 --> 00:00:04,00
So, in this video, we're going to build

3
00:00:04,00 --> 00:00:06,00
a very simple Hello World agent,

4
00:00:06,00 --> 00:00:08,01
and this will help us see the concepts

5
00:00:08,01 --> 00:00:12,01
of intent, planning, tools, and reflection in action.

6
00:00:12,01 --> 00:00:13,09
Our agent's goal is going to be very simple,

7
00:00:13,09 --> 00:00:15,05
it's to greet the user,

8
00:00:15,05 --> 00:00:17,05
but instead of just printing a static message,

9
00:00:17,05 --> 00:00:20,07
it's going to use a tool to generate the greeting.

10
00:00:20,07 --> 00:00:22,04
So, let's get started with this one,

11
00:00:22,04 --> 00:00:23,09
and I've started by just creating

12
00:00:23,09 --> 00:00:26,04
this file called hello_agent.py,

13
00:00:26,04 --> 00:00:28,08
and I'm going to start coding in it right away.

14
00:00:28,08 --> 00:00:31,00
Okay, first of all is to define the tool,

15
00:00:31,00 --> 00:00:34,02
so I'm going to call the tool create_greeting.

16
00:00:34,02 --> 00:00:36,07
It's going to take the name as a string.

17
00:00:36,07 --> 00:00:38,09
I'll just have that return a string.

18
00:00:38,09 --> 00:00:43,04
And then here I'm going to return, "Hello,"

19
00:00:43,04 --> 00:00:46,07
to the name that was passed in.

20
00:00:46,07 --> 00:00:49,01
And maybe just something simple like,

21
00:00:49,01 --> 00:00:52,04
"Welcome to the world

22
00:00:52,04 --> 00:00:55,05
of AI agents."

23
00:00:55,05 --> 00:00:58,05
Okay, so that's my very simple tool.

24
00:00:58,05 --> 00:01:00,06
So, now let's take a look at the main logic

25
00:01:00,06 --> 00:01:02,01
of the actual agent itself,

26
00:01:02,01 --> 00:01:06,08
so I'm just going to call that run_agent.

27
00:01:06,08 --> 00:01:10,05
So, step one is going to be to understand the intent.

28
00:01:10,05 --> 00:01:13,02
Now in this case, understanding the intent is very simple.

29
00:01:13,02 --> 00:01:15,04
We're just getting the user's name.

30
00:01:15,04 --> 00:01:16,08
Typically with an agent,

31
00:01:16,08 --> 00:01:19,01
you're going to have like a vague thing

32
00:01:19,01 --> 00:01:22,04
that you want the agent to do that's an unstructured text,

33
00:01:22,04 --> 00:01:25,01
like, "Find me a place where I could hike today,"

34
00:01:25,01 --> 00:01:27,06
or, "Find me a restaurant that serves my favorite dish

35
00:01:27,06 --> 00:01:30,07
that's open after nine o'clock on Mondays

36
00:01:30,07 --> 00:01:33,05
and is within 25 miles of me."

37
00:01:33,05 --> 00:01:35,03
And the whole idea of intent is being able

38
00:01:35,03 --> 00:01:37,02
to take that big block of text

39
00:01:37,02 --> 00:01:39,04
and really understand what the user wants.

40
00:01:39,04 --> 00:01:41,08
It's one of the magical things of large language models

41
00:01:41,08 --> 00:01:45,01
compared to old-style natural language processing,

42
00:01:45,01 --> 00:01:47,01
where you can throw that block of text and say,

43
00:01:47,01 --> 00:01:48,03
"Help me figure out the intent.

44
00:01:48,03 --> 00:01:49,08
What does the person want to do?

45
00:01:49,08 --> 00:01:52,00
Do they want to hike, do they want to go to a restaurant?

46
00:01:52,00 --> 00:01:53,00
Tell me those details."

47
00:01:53,00 --> 00:01:54,05
And then to turn that unstructured data

48
00:01:54,05 --> 00:01:55,09
into structured data.

49
00:01:55,09 --> 00:01:58,08
But in this case, the agent is much, much simpler than that.

50
00:01:58,08 --> 00:02:01,09
The intent is to just figure out the person's name

51
00:02:01,09 --> 00:02:03,03
so that you can say hello to them.

52
00:02:03,03 --> 00:02:05,07
And in this case, we're not going to bother with an LLM,

53
00:02:05,07 --> 00:02:07,05
we're just going to ask them their name.

54
00:02:07,05 --> 00:02:10,07
So, then the second part is when your agent understands

55
00:02:10,07 --> 00:02:12,06
what the tools are available to it,

56
00:02:12,06 --> 00:02:15,00
it's got to plan to be able to use those.

57
00:02:15,00 --> 00:02:16,06
So, I'm going to put the plan together,

58
00:02:16,06 --> 00:02:18,03
but of course, here my plan

59
00:02:18,03 --> 00:02:19,09
is just to use the greeting tool.

60
00:02:19,09 --> 00:02:22,02
We're just going to say hello to that person.

61
00:02:22,02 --> 00:02:25,02
So, my plan is I'll just put out something,

62
00:02:25,02 --> 00:02:28,06
I'll print that the agent is going to say,

63
00:02:28,06 --> 00:02:32,09
"I should use the create greeting tool."

64
00:02:32,09 --> 00:02:34,02
I know I'm cheating a little bit,

65
00:02:34,02 --> 00:02:37,08
but the whole idea is just to show that level of

66
00:02:37,08 --> 00:02:40,09
breaking it down into the steps that an agent would do.

67
00:02:40,09 --> 00:02:43,08
And again, typically, if you're thinking about

68
00:02:43,08 --> 00:02:45,08
the vague intent that I mentioned earlier on,

69
00:02:45,08 --> 00:02:48,01
I would like to have dinner of a particular cuisine,

70
00:02:48,01 --> 00:02:50,04
or I'd like to hike in a particular place.

71
00:02:50,04 --> 00:02:53,00
Then once you've established the intent from that,

72
00:02:53,00 --> 00:02:56,00
then you could also call a large language model to say,

73
00:02:56,00 --> 00:02:57,07
okay, my user wants to do this.

74
00:02:57,07 --> 00:02:59,06
These are the tools that are available to me.

75
00:02:59,06 --> 00:03:02,02
Help me break it down into a plan to use those tools,

76
00:03:02,02 --> 00:03:03,03
that type of thing.

77
00:03:03,03 --> 00:03:05,09
But in this case, I just have a single tool

78
00:03:05,09 --> 00:03:09,08
and I'm going to bypass all of that and just create a greeting.

79
00:03:09,08 --> 00:03:13,05
So, create_greeting, and the user has entered their name.

80
00:03:13,05 --> 00:03:16,05
So, now I'll call that create greeting tool to do that.

81
00:03:16,05 --> 00:03:17,05
And then finally, of course,

82
00:03:17,05 --> 00:03:20,05
the final step was to use the tools.

83
00:03:20,05 --> 00:03:21,09
I should put that here.

84
00:03:21,09 --> 00:03:25,02
This we're actually using the tools was step three.

85
00:03:25,02 --> 00:03:28,05
And then finally step four is to reflect on what we've done.

86
00:03:28,05 --> 00:03:31,09
And in this case, reflecting would be, did it work?

87
00:03:31,09 --> 00:03:33,00
Did we create a greeting?

88
00:03:33,00 --> 00:03:34,06
Did something go wrong?

89
00:03:34,06 --> 00:03:37,04
And because I'm returning my greeting, I could just say,

90
00:03:37,04 --> 00:03:39,08
if greeting,

91
00:03:39,08 --> 00:03:42,09
print,

92
00:03:42,09 --> 00:03:44,04
agent

93
00:03:44,04 --> 00:03:46,01
says

94
00:03:46,01 --> 00:03:48,07
the greeting that was returned.

95
00:03:48,07 --> 00:03:51,09
Otherwise it failed on the greeting,

96
00:03:51,09 --> 00:03:54,01
so I'm just going to say, print,

97
00:03:54,01 --> 00:03:55,08
agent says

98
00:03:55,08 --> 00:03:57,05
I failed

99
00:03:57,05 --> 00:04:01,00
to create a greeting.

100
00:04:01,00 --> 00:04:01,09
If you wanted to make it

101
00:04:01,09 --> 00:04:03,05
a little bit more sophisticated there,

102
00:04:03,05 --> 00:04:06,00
you could have it do something like,

103
00:04:06,00 --> 00:04:07,03
if it failed to create the greeting,

104
00:04:07,03 --> 00:04:10,04
maybe it could go back and ask or tell the user to be

105
00:04:10,04 --> 00:04:12,08
more specific about the data that they put in,

106
00:04:12,08 --> 00:04:14,04
or something along those lines.

107
00:04:14,04 --> 00:04:16,07
But again, this is just a very simple

108
00:04:16,07 --> 00:04:19,04
Hello World-type thing.

109
00:04:19,04 --> 00:04:21,08
And we just want to show all of the steps

110
00:04:21,08 --> 00:04:24,03
that would be in creating an agent.

111
00:04:24,03 --> 00:04:26,05
So, if you're a Python developer, you'll be like,

112
00:04:26,05 --> 00:04:28,05
why on earth is he doing it this way?

113
00:04:28,05 --> 00:04:30,02
There's a much simpler way of doing that.

114
00:04:30,02 --> 00:04:32,03
All you have to do is input and print.

115
00:04:32,03 --> 00:04:33,04
But the whole idea here

116
00:04:33,04 --> 00:04:35,07
is to start getting into that mindset

117
00:04:35,07 --> 00:04:37,09
when you're building an agentic AI

118
00:04:37,09 --> 00:04:40,06
to think about breaking it into those steps

119
00:04:40,06 --> 00:04:43,05
of understanding intent, making a plan,

120
00:04:43,05 --> 00:04:45,03
using tools to achieve that plan,

121
00:04:45,03 --> 00:04:47,01
and then reflecting on them.

122
00:04:47,01 --> 00:04:49,04
So, this is our app, let's see what happens if we run it.

123
00:04:49,04 --> 00:04:52,08
Hopefully I don't have any bugs in there, but...

124
00:04:52,08 --> 00:04:54,07
So, now to run it, I'm just going to do

125
00:04:54,07 --> 00:04:57,03
python hello_agent.py.

126
00:04:57,03 --> 00:05:00,08
It's going to ask me my name, I'll tell it the truth, Laurence.

127
00:05:00,08 --> 00:05:01,07
And you can see here,

128
00:05:01,07 --> 00:05:03,06
the agent decided to use the greeting tool,

129
00:05:03,06 --> 00:05:05,03
the create greeting tool created the string,

130
00:05:05,03 --> 00:05:07,06
"Hello, Laurence, welcome to the world of AI agents."

131
00:05:07,06 --> 00:05:09,09
And then the agent outputted that to me.

132
00:05:09,09 --> 00:05:14,01
So, let's quickly break that down into our agentic loop.

133
00:05:14,01 --> 00:05:15,05
And you'll hear me saying this one a lot,

134
00:05:15,05 --> 00:05:17,04
but it is important.

135
00:05:17,04 --> 00:05:19,04
Breaking it down your applications

136
00:05:19,04 --> 00:05:21,05
when you want to build AI agents

137
00:05:21,05 --> 00:05:22,08
is to really think about

138
00:05:22,08 --> 00:05:25,07
how you're going to meet those four steps,

139
00:05:25,07 --> 00:05:27,04
understanding the user's intent.

140
00:05:27,04 --> 00:05:29,08
Now, you can do that with AI or not.

141
00:05:29,08 --> 00:05:32,08
Making a plan, you can do that with AI or not.

142
00:05:32,08 --> 00:05:34,09
Using the tools, here's the important part

143
00:05:34,09 --> 00:05:37,01
where the agent executes the plan

144
00:05:37,01 --> 00:05:39,06
by calling whatever tools are available to it.

145
00:05:39,06 --> 00:05:42,02
In this case, it was just that create greeting function,

146
00:05:42,02 --> 00:05:43,05
and then reflect.

147
00:05:43,05 --> 00:05:45,00
And in this case, the agent was able

148
00:05:45,00 --> 00:05:46,04
to check the output of the tool.

149
00:05:46,04 --> 00:05:48,05
It got a valid greeting, so it printed it.

150
00:05:48,05 --> 00:05:50,07
Otherwise it would acknowledge that it failed.

151
00:05:50,07 --> 00:05:51,05
And there you have it.

152
00:05:51,05 --> 00:05:55,02
This is the very first, very, very basic AI agent.

153
00:05:55,02 --> 00:05:57,01
The next module, we're going to start building out

154
00:05:57,01 --> 00:06:00,00
our much more advanced hiking assistant agent.