1
00:00:00,200 --> 00:00:00,840
Okay.

2
00:00:00,880 --> 00:00:02,280
It did indeed finish.

3
00:00:02,320 --> 00:00:03,800
It completed its three searches.

4
00:00:03,800 --> 00:00:06,600
It ended with hooray, which is success.

5
00:00:06,600 --> 00:00:08,200
And I received an email.

6
00:00:08,200 --> 00:00:09,160
And here it is.

7
00:00:09,160 --> 00:00:16,560
A nice HTML email, well formatted with a good introduction and overview, detailed analysis crew.

8
00:00:16,600 --> 00:00:19,680
It's now putting at the top, which is in fact what we'll be doing next week.

9
00:00:19,680 --> 00:00:20,720
Look forward to that.

10
00:00:20,920 --> 00:00:24,160
And then there's generally some great information here.

11
00:00:24,160 --> 00:00:28,040
I would say that there's a couple of misses, but it's generally done pretty well.

12
00:00:28,040 --> 00:00:32,360
And then at the bottom are some references with links since it's an HTML email.

13
00:00:32,360 --> 00:00:35,960
And if I click on that it does appear to work, which is nice.

14
00:00:36,240 --> 00:00:43,240
And I will say, I do find it very satisfying that with this minimal scaffolding with something that

15
00:00:43,240 --> 00:00:46,800
was really very simple indeed, there's not much code to this look.

16
00:00:46,840 --> 00:00:49,240
That's all there is with just this.

17
00:00:49,280 --> 00:00:53,280
We were able to build this whole deep research framework.

18
00:00:53,520 --> 00:01:00,600
And whilst, you know, maybe those those results are reasonably simplistic, I hope you appreciate

19
00:01:00,600 --> 00:01:05,120
how much potential there is with what we've just built, and can think of all the different ways that

20
00:01:05,120 --> 00:01:06,760
we could expand on this.

21
00:01:06,800 --> 00:01:12,520
And in fact, one very simple way we can expand on this is just to turn up that number, which you may

22
00:01:12,520 --> 00:01:17,560
not want to do, because as I say, it will cost that few cents a piece, but there's no reason why

23
00:01:17,560 --> 00:01:24,200
I can't do it for you so that we can get some to show you what it looks like to have that deeper kind

24
00:01:24,200 --> 00:01:25,040
of research.

25
00:01:25,040 --> 00:01:26,800
So I will do that right now.

26
00:01:26,800 --> 00:01:33,720
We will change that to 20 and, uh, go through and make sure that we, um, hang on if I just set that

27
00:01:33,720 --> 00:01:34,560
properly.

28
00:01:35,240 --> 00:01:38,680
Yes, I have that has now recreated that planner agent.

29
00:01:39,000 --> 00:01:48,480
So now we will come back down here and just make sure that we rerun everything and rerun this.

30
00:01:48,480 --> 00:01:55,200
And now we will expect to do rather a lot more searches, 20 more searches.

31
00:01:55,400 --> 00:02:01,360
And this may take a moment longer and then I will come back and show you what results we get.

32
00:02:01,640 --> 00:02:04,920
And that is completed after 20 searches.

33
00:02:05,120 --> 00:02:10,120
And I can show you the results, which actually it looks generally a bit similar.

34
00:02:10,120 --> 00:02:14,760
It's got the same key frameworks, but it's got more information laid around it.

35
00:02:14,800 --> 00:02:19,320
It has applications and benefits associated with each of the frameworks that I identified.

36
00:02:19,320 --> 00:02:23,400
And then it has some commercial implications at the end and a conclusion.

37
00:02:23,400 --> 00:02:26,120
And so it's certainly more substantive than the previous one.

38
00:02:26,120 --> 00:02:27,040
It's a step up.

39
00:02:27,360 --> 00:02:33,360
And obviously, should you wish, I would very much recommend that you experiment with this.

40
00:02:33,520 --> 00:02:36,560
And we should also, of course, go in and look at our trace.

41
00:02:36,760 --> 00:02:43,360
And if we come in to the research trace, we will see that all of these agents ran and they all ran

42
00:02:43,360 --> 00:02:44,360
in parallel.

43
00:02:44,480 --> 00:02:47,560
The planner agent took took took that time to start with.

44
00:02:47,920 --> 00:02:49,120
Then there's load more.

45
00:02:49,840 --> 00:02:52,240
There's all of this and more.

46
00:02:53,280 --> 00:02:57,210
And then at the end was the writer agent and the email agent.

47
00:02:57,410 --> 00:02:58,890
So you can see everything.

48
00:02:58,930 --> 00:03:05,170
And it definitely gives you that sense of how Asyncio ran all of these different searches in parallel

49
00:03:05,170 --> 00:03:06,250
at the same time.

50
00:03:06,410 --> 00:03:10,290
And then, of course, the writer and email were sequential.

51
00:03:10,690 --> 00:03:16,530
And that then is the conclusion of this part of the deep research agent.

52
00:03:17,130 --> 00:03:22,850
The thing I'd like you to do is bear in mind, have a think about how you could make this more substantive.

53
00:03:22,850 --> 00:03:24,010
What more could be done?

54
00:03:24,010 --> 00:03:27,130
What would you like to do to beef this up?

55
00:03:27,370 --> 00:03:31,250
And then we'll do a quick wrap up before we launch into tomorrow.

56
00:03:31,930 --> 00:03:37,690
So I really hope you enjoyed that as much as I did, and that you appreciate how simple it was to build

57
00:03:37,690 --> 00:03:43,610
that framework around a deep research agent in your deeply puzzling over how you can make it better.

58
00:03:43,930 --> 00:03:48,290
But actually, what we're going to be spending tomorrow doing is putting this into an application and

59
00:03:48,290 --> 00:03:52,050
having it be something that is like a takeaway deep research agent.

60
00:03:52,210 --> 00:03:53,810
And I'm very excited for it.

61
00:03:53,810 --> 00:03:54,850
And I will see you then.