1
00:00:00,05 --> 00:00:01,09
- [Instructor] Our agent is fully functional,

2
00:00:01,09 --> 00:00:03,06
and that's fantastic.

3
00:00:03,06 --> 00:00:06,04
But what happens if an external API is down,

4
00:00:06,04 --> 00:00:09,08
or our network connection fails, or other issues like that?

5
00:00:09,08 --> 00:00:12,09
Right now, there's a risk that our agent would crash.

6
00:00:12,09 --> 00:00:14,04
So one of the things I'd like to do

7
00:00:14,04 --> 00:00:17,00
is to look at how to make the agent more robust

8
00:00:17,00 --> 00:00:19,01
by adding error handling.

9
00:00:19,01 --> 00:00:22,00
And we can do this manually or we can use an AI assistant

10
00:00:22,00 --> 00:00:26,00
like the Gemini CLI to do most of that heavy lifting for us.

11
00:00:26,00 --> 00:00:28,01
The goal is to take all of the code

12
00:00:28,01 --> 00:00:30,03
in our hiking_agent directory

13
00:00:30,03 --> 00:00:33,02
and make it more resilient to errors.

14
00:00:33,02 --> 00:00:35,02
If we can wrap the external calls

15
00:00:35,02 --> 00:00:37,08
to the location, weather, and parks APIs,

16
00:00:37,08 --> 00:00:39,06
and try...except blocks,

17
00:00:39,06 --> 00:00:42,02
then this would allow our agent to fail gracefully

18
00:00:42,02 --> 00:00:44,06
with a clear message instead of crashing.

19
00:00:44,06 --> 00:00:46,04
Indeed, we saw some of the code for this

20
00:00:46,04 --> 00:00:47,08
a little bit earlier on,

21
00:00:47,08 --> 00:00:50,08
but I want to show what it would be like with the initial code

22
00:00:50,08 --> 00:00:52,00
that many of us might build

23
00:00:52,00 --> 00:00:54,03
that might be a little bit more naive.

24
00:00:54,03 --> 00:00:57,04
For example, let's start with the parks.py.

25
00:00:57,04 --> 00:00:59,06
And it's making these URL requests

26
00:00:59,06 --> 00:01:02,01
that are common areas of failure.

27
00:01:02,01 --> 00:01:03,09
So for example, something like this one,

28
00:01:03,09 --> 00:01:06,06
the common pattern is to define the URL,

29
00:01:06,06 --> 00:01:08,05
to call requests to get the response,

30
00:01:08,05 --> 00:01:11,01
and then to return the response.json.

31
00:01:11,01 --> 00:01:13,08
Now, there are a number of areas where this could go wrong,

32
00:01:13,08 --> 00:01:15,05
and naively, we could say, to fix it,

33
00:01:15,05 --> 00:01:18,07
we could manually add a try...except block,

34
00:01:18,07 --> 00:01:22,00
but there may be corner cases that we don't see.

35
00:01:22,00 --> 00:01:29,00
So what if we ask an AI assistant to do this work for us?

36
00:01:29,00 --> 00:01:31,07
So I'm going to do that by starting Gemini.

37
00:01:31,07 --> 00:01:34,00
You can use any AI assistant you like of course,

38
00:01:34,00 --> 00:01:36,03
but I like using the Gemini CLI

39
00:01:36,03 --> 00:01:39,00
because I'm in the same directory as the code

40
00:01:39,00 --> 00:01:42,00
and I can work directly within the code itself.

41
00:01:42,00 --> 00:01:44,04
So I can specifically say something like,

42
00:01:44,04 --> 00:01:48,08
in my parks.py file,

43
00:01:48,08 --> 00:01:54,00
please update the get_parks function

44
00:01:54,00 --> 00:01:59,02
to handle potential network errors.

45
00:01:59,02 --> 00:02:05,06
It should use a try...except block.

46
00:02:05,06 --> 00:02:11,06
Check for bad HTTP responses,

47
00:02:11,06 --> 00:02:14,01
network failures,

48
00:02:14,01 --> 00:02:18,08
or any other issues you can think of

49
00:02:18,08 --> 00:02:22,00
and return None

50
00:02:22,00 --> 00:02:24,07
when an error occurs.

51
00:02:24,07 --> 00:02:27,05
This is one of the ways that your skills as a developer

52
00:02:27,05 --> 00:02:29,05
are actually more useful, I think,

53
00:02:29,05 --> 00:02:32,04
in the age of AI because you can be much more specific

54
00:02:32,04 --> 00:02:34,07
with your prompt here by the kind of things

55
00:02:34,07 --> 00:02:37,03
that you're experienced, like HTTP errors,

56
00:02:37,03 --> 00:02:39,05
bad H responses, network failures,

57
00:02:39,05 --> 00:02:41,03
all of those kind of things.

58
00:02:41,03 --> 00:02:43,00
And I just did get something like this

59
00:02:43,00 --> 00:02:46,01
and have it kind of generate the code for me.

60
00:02:46,01 --> 00:02:47,05
It's the first time I'm running it,

61
00:02:47,05 --> 00:02:49,06
so sometimes it's going to ask me for permission

62
00:02:49,06 --> 00:02:51,06
to be able to write and stuff like that.

63
00:02:51,06 --> 00:02:53,00
And we can see the code

64
00:02:53,00 --> 00:02:55,07
that Gemini came up with here for me.

65
00:02:55,07 --> 00:02:57,06
It's using this raise_for_status,

66
00:02:57,06 --> 00:03:00,07
which can check for HTTP errors

67
00:03:00,07 --> 00:03:02,06
and do other goodies like this as well.

68
00:03:02,06 --> 00:03:06,07
So it actually also spotted this one where I could have had,

69
00:03:06,07 --> 00:03:09,02
if I look at it here briefly before I continue,

70
00:03:09,02 --> 00:03:11,01
that there are corner cases

71
00:03:11,01 --> 00:03:14,09
where if there's a null in the data, that it would crash.

72
00:03:14,09 --> 00:03:18,03
And the if data in parks_data would actually crash

73
00:03:18,03 --> 00:03:20,01
rather than failing

74
00:03:20,01 --> 00:03:22,09
and gracefully allowing the agent to continue,

75
00:03:22,09 --> 00:03:24,01
there'll be an actual crash here.

76
00:03:24,01 --> 00:03:27,02
So we're saying like, you know, it's going to update that.

77
00:03:27,02 --> 00:03:30,00
And the agent being the agent, the Gemini one,

78
00:03:30,00 --> 00:03:31,04
is it's going to actually run it

79
00:03:31,04 --> 00:03:33,00
and just check if it's working

80
00:03:33,00 --> 00:03:36,00
and do other things and fixes for us like that.

81
00:03:36,00 --> 00:03:37,08
I find that's really, really useful,

82
00:03:37,08 --> 00:03:39,03
and I think it's one of those things

83
00:03:39,03 --> 00:03:41,03
that would be very important for you

84
00:03:41,03 --> 00:03:45,01
to be able to check errors in your files like this.

85
00:03:45,01 --> 00:03:47,06
In this case, actually, I didn't run it

86
00:03:47,06 --> 00:03:48,08
in a virtual environment,

87
00:03:48,08 --> 00:03:50,07
so it found that that package was missing.

88
00:03:50,07 --> 00:03:53,05
And with that package missing, it wants to install that,

89
00:03:53,05 --> 00:03:56,01
and then retest the file with the package there.

90
00:03:56,01 --> 00:03:58,07
And we can see it actually ran parks.py

91
00:03:58,07 --> 00:04:00,06
and was able to find some trails for us,

92
00:04:00,06 --> 00:04:02,06
so we know that it's working now.

93
00:04:02,06 --> 00:04:05,04
The changes that it did to add error handling

94
00:04:05,04 --> 00:04:08,00
have not changed anything.

95
00:04:08,00 --> 00:04:11,09
And I'm just allowing it to remove any files that it added.

96
00:04:11,09 --> 00:04:13,01
While it was working through this,

97
00:04:13,01 --> 00:04:15,02
sometimes it writes temporary files and stuff like that.

98
00:04:15,02 --> 00:04:17,01
I'm allowing it to remove them.

99
00:04:17,01 --> 00:04:21,01
So this is a good example of being able to use

100
00:04:21,01 --> 00:04:24,02
a AI agent or an AI model.

101
00:04:24,02 --> 00:04:27,01
I love to do it as a command-line interface, personally,

102
00:04:27,01 --> 00:04:29,04
so it save a lot of cutting and pasting when I'm working

103
00:04:29,04 --> 00:04:32,02
and stuff like this to update and improve your code.

104
00:04:32,02 --> 00:04:35,01
And we can see the updated and improved code here.

105
00:04:35,01 --> 00:04:36,09
So as you can see,

106
00:04:36,09 --> 00:04:40,06
the idea here is you can write the code yourself,

107
00:04:40,06 --> 00:04:44,00
you can create code for building an agent like this one

108
00:04:44,00 --> 00:04:45,04
using your own understanding,

109
00:04:45,04 --> 00:04:48,07
but there are always corner cases that you may not see,

110
00:04:48,07 --> 00:04:50,07
and I find that using an AI assistant

111
00:04:50,07 --> 00:04:53,03
can dramatically speed up the process

112
00:04:53,03 --> 00:04:56,03
of that polishing your code and hardening your code

113
00:04:56,03 --> 00:04:58,05
and finding those corner cases.

114
00:04:58,05 --> 00:05:01,07
I gave a quick example of a clear semantic prompt,

115
00:05:01,07 --> 00:05:04,02
and I would encourage you to think about that,

116
00:05:04,02 --> 00:05:06,02
think about the code that you're doing,

117
00:05:06,02 --> 00:05:08,06
and provide those prompts that will allow you

118
00:05:08,06 --> 00:05:12,01
to delegate the important and repetitive tasks

119
00:05:12,01 --> 00:05:14,05
of doing things like adding error handling.

120
00:05:14,05 --> 00:05:17,00
I've provided in the download finished versions of the code

121
00:05:17,00 --> 00:05:19,07
that I worked on that I find is more robust

122
00:05:19,07 --> 00:05:21,05
than I built it in that way.

123
00:05:21,05 --> 00:05:24,01
And in the next video, I'm going to use the same technique

124
00:05:24,01 --> 00:05:27,00
to add documentation to the code.