1
00:00:00,240 --> 00:00:03,280
Well, hello and welcome to week three.

2
00:00:03,320 --> 00:00:04,240
We are here.

3
00:00:04,240 --> 00:00:13,160
We are at Crew Week, and this is the time that I get to unveil the world of crew A, and I want to

4
00:00:13,160 --> 00:00:18,680
say that you might find it a little bit painful to make this mental change now, because we've just

5
00:00:18,680 --> 00:00:21,080
fallen in love with OpenAI agents SDK.

6
00:00:21,120 --> 00:00:23,400
We've got used to it, we know all about it.

7
00:00:23,400 --> 00:00:27,440
And suddenly there's going to be a change new terminology, new constructs.

8
00:00:27,440 --> 00:00:30,440
And the thing is, I would say there's a lot in common.

9
00:00:30,440 --> 00:00:34,320
There's going to be some differences too, but you're going to find you're quickly going to fall in

10
00:00:34,320 --> 00:00:35,880
love with AI as well.

11
00:00:35,920 --> 00:00:40,040
And this is going to be a repeating pattern in the next few weeks that we're going to get comfortable

12
00:00:40,040 --> 00:00:44,040
with something and love it, and then we're going to have to put it to one side and move on to the next

13
00:00:44,040 --> 00:00:45,280
and just keep it in mind.

14
00:00:45,320 --> 00:00:47,160
Keep keep in mind the differences.

15
00:00:47,160 --> 00:00:50,480
What's similar, what's unique, what's better, what's worse.

16
00:00:50,480 --> 00:00:54,320
Because different projects will lend themselves more to different frameworks.

17
00:00:54,320 --> 00:00:56,160
And you're going to find that you'll fall in love with one.

18
00:00:56,160 --> 00:00:59,800
And it may not be the same as me, and you should be able to make their own determination, and you'll

19
00:00:59,800 --> 00:01:01,950
learn something from each of these experiments.

20
00:01:01,990 --> 00:01:04,910
Plus all of our commercial projects will be a bit different.

21
00:01:05,110 --> 00:01:06,630
Okay, that's enough preamble.

22
00:01:06,630 --> 00:01:07,430
Let's get to it.

23
00:01:08,190 --> 00:01:11,510
So crew is actually several different things.

24
00:01:11,510 --> 00:01:13,470
We're gonna we're gonna have this a couple of times.

25
00:01:13,510 --> 00:01:14,950
Autogen is much the same.

26
00:01:14,990 --> 00:01:19,550
But crew in particular, when you when you hear people talk about crew, they might be talking about

27
00:01:19,590 --> 00:01:27,350
something called crew I enterprise, which is crews platform for deploying agents, for running them,

28
00:01:27,350 --> 00:01:30,910
for monitoring and managing them through a number of different screens.

29
00:01:31,350 --> 00:01:33,670
They sometimes just call it the career AI platform.

30
00:01:33,710 --> 00:01:37,430
Sometimes I think officially it's branded crew AI enterprise.

31
00:01:37,470 --> 00:01:46,230
And if you go to their landing page at currycomb, not com, you'll see that this is what sort of presented

32
00:01:46,230 --> 00:01:47,670
first and foremost.

33
00:01:48,390 --> 00:01:56,670
Secondly, there is a product called crew crew I, UI studio, which is one of these low code, no code

34
00:01:56,670 --> 00:02:00,070
platforms for piecing together agent interactions.

35
00:02:00,190 --> 00:02:02,950
A bit like an addendum that we looked at at the very, very beginning.

36
00:02:03,390 --> 00:02:06,910
But it's a it's a nice, elegant end user tool.

37
00:02:07,150 --> 00:02:13,910
And then thirdly, it is something called the framework, the crew AI framework, which is an open source

38
00:02:13,910 --> 00:02:15,190
framework for.

39
00:02:15,230 --> 00:02:21,310
And I quote from, from the site orchestrating high performing AI agents with ease and scale.

40
00:02:21,710 --> 00:02:25,150
So these are the three different offerings that they have.

41
00:02:25,470 --> 00:02:30,710
And hopefully no surprise to you, we are going to be focusing on the open source framework because

42
00:02:30,710 --> 00:02:32,750
we're here to build agents ourselves.

43
00:02:32,750 --> 00:02:34,110
We're writing the code.

44
00:02:34,110 --> 00:02:39,470
We're not going to be using the low code tooling, which is there, and we're also not going to be necessarily

45
00:02:39,470 --> 00:02:45,350
needing to do something where we're deploying them and paying for a hosting platform, which is really

46
00:02:45,350 --> 00:02:47,150
crew AI enterprise.

47
00:02:47,550 --> 00:02:53,270
But it's perhaps worth pointing out that when we think of the differences between these different platforms,

48
00:02:53,550 --> 00:02:58,790
OpenAI and anthropic have the great benefit that they already have a reason for being.

49
00:02:58,830 --> 00:03:04,620
They they have their models and that is their source of revenue when it comes to groups like cry.

50
00:03:04,660 --> 00:03:08,580
Of course, they they need to be very mindful of a monetization strategy.

51
00:03:08,740 --> 00:03:12,180
And the open source framework is very popular, very successful.

52
00:03:12,180 --> 00:03:16,340
But of course, they also need ways to monetize and cry enterprise.

53
00:03:16,340 --> 00:03:22,820
And their their tooling is their path to doing that which makes complete sense and which I don't I don't

54
00:03:22,820 --> 00:03:24,220
hold it against them for a moment.

55
00:03:24,220 --> 00:03:28,940
But it does, of course, mean that when you go to things like their website, there is a lot of upselling

56
00:03:28,940 --> 00:03:35,020
going on that they want to try and win people over so that not only will they use the open source toolkit,

57
00:03:35,020 --> 00:03:42,060
but then they'll end up paying for hosting and deployment within the broader cry platform.

58
00:03:42,980 --> 00:03:47,860
And so now for the rest of what we talk about, it will always be the open source framework that we

59
00:03:47,860 --> 00:03:49,300
will be working with.

60
00:03:49,340 --> 00:03:54,660
And when you work with the framework, there are in fact two different flavors, two different approaches

61
00:03:54,660 --> 00:03:57,860
that you can use for all of your work with the framework.

62
00:03:57,900 --> 00:04:05,500
And one of them is called crew I Crews, which is when you have autonomous solutions with teams working

63
00:04:05,540 --> 00:04:07,660
together, agents of different roles.

64
00:04:08,020 --> 00:04:12,380
Crew is crews word for a sort of team of agents, a crew of agents.

65
00:04:12,940 --> 00:04:16,060
And then there's also something called crew I flows.

66
00:04:16,380 --> 00:04:21,700
And this I think this is actually a newer part of crew because at least I wasn't aware of it about six

67
00:04:21,700 --> 00:04:23,180
months ago when I last used crew.

68
00:04:23,220 --> 00:04:24,140
It may have been there, but.

69
00:04:24,140 --> 00:04:28,100
And I just didn't notice it, but but it's certainly now quite prominently in the documentation.

70
00:04:28,340 --> 00:04:36,660
Cry flows are more, um, prescribed workflows, fixed workflows where you divide a problem into multiple

71
00:04:36,660 --> 00:04:42,340
steps and you have workflows that lead through it with decision points and outcomes and so on.

72
00:04:42,860 --> 00:04:48,060
And their their documentation suggests that you should choose crews when you're looking for autonomous

73
00:04:48,060 --> 00:04:56,180
problem solving, creative collaboration, or exploratory tasks versus flows, which is more about deterministic

74
00:04:56,220 --> 00:05:00,180
outcomes, auditability or precise control.

75
00:05:00,300 --> 00:05:06,690
And I probably I imagine and this is just speculation, but I'm imagining that flows has come out as

76
00:05:06,690 --> 00:05:13,650
a result of some of the concerns people have about running crews in production, where there is this

77
00:05:13,650 --> 00:05:21,690
greater level of uncertainty and lack of auditability and sometimes a tighter defined flow, a work

78
00:05:21,730 --> 00:05:26,770
flow is what's required rather than a fully agentic autonomous solution.

79
00:05:27,570 --> 00:05:31,410
So that hopefully gives you a bit of context, and again, no surprise to know that we're going to be

80
00:05:31,410 --> 00:05:34,650
focusing on the crews, because this is all about agents.

81
00:05:34,690 --> 00:05:40,890
This course and building workflows is something that you can also do, but it's a bit more straightforward.

82
00:05:40,890 --> 00:05:46,690
And of course, you could also do that simply by calling LMS directly and by interpreting their responses.

83
00:05:46,810 --> 00:05:51,010
What we're interested in is the more autonomous aspect of this.

84
00:05:51,050 --> 00:05:56,850
It is about when we allow different LMS to choose their own adventure and to go about solving their

85
00:05:56,850 --> 00:05:58,850
problems in an agentic way.

86
00:05:59,210 --> 00:06:01,890
So that will be our focus for this week.