1
00:00:03,661 --> 00:00:07,661
Hi everyone, I'm Matt Ragone and I'm a financial planner at Range.

2
00:00:07,661 --> 00:00:14,421
I want to show you how our financial planners ended up writing the agents that we use every day.

3
00:00:14,641 --> 00:00:18,581
And none of us, the financial planners for that matter, write code.

4
00:00:18,801 --> 00:00:20,461
This is the build guide.

5
00:00:20,641 --> 00:00:25,681
How we did it step by step so you can take it back to your own teams, companies, etc.

6
00:00:26,381 --> 00:00:31,001
So a quick bit about me because it matters for how you'll hear this talk.

7
00:00:31,201 --> 00:00:32,341
I'm not an engineer.

8
00:00:32,341 --> 00:00:34,241
I'm a certified financial planner.

9
00:00:34,241 --> 00:00:40,841
and I spend my days with clients talking about cash flow, taxes, insurance, retirement, estate, you name it.

10
00:00:40,981 --> 00:00:48,341
I also lead the AI adoption for the planning team at Range, which is an all-in-one wealth management company.

11
00:00:48,721 --> 00:00:53,401
Most agent talks you'll hear come from the people who built the platform.

12
00:00:53,741 --> 00:00:58,621
This talk comes from someone on the other side, the people you hand the platform to.

13
00:00:58,621 --> 00:01:13,141
We work in a regulated business. The output is financial advice to real clients. And the review step is not optional. So this isn't a vendor pitch. It's a method and it's what actually worked for us.

14
00:01:14,441 --> 00:01:18,581
We started where everyone starts, AI as a chatbot.

15
00:01:18,781 --> 00:01:23,521
I'd write a client message, paste it in, and ask it to make it a little clearer or friendlier.

16
00:01:23,681 --> 00:01:24,741
And that was useful.

17
00:01:24,981 --> 00:01:30,001
It saved me a few minutes here and there, but I was still doing all of the work manually.

18
00:01:30,221 --> 00:01:32,741
I was just doing it a little bit faster.

19
00:01:33,001 --> 00:01:36,421
The change that actually mattered came later.

20
00:01:37,341 --> 00:01:38,581
Here's what we figured out.

21
00:01:38,801 --> 00:01:42,701
A planner's constraint was never typing speed, for example.

22
00:01:43,141 --> 00:01:46,561
It's that we can only be in one place at a time.

23
00:01:46,761 --> 00:01:50,901
If I'm on a call with a client, I'm not building someone else's plan.

24
00:01:51,501 --> 00:01:53,781
Writing faster doesn't fix that problem.

25
00:01:54,061 --> 00:01:58,041
What fixes it is work that happens while I'm somewhere else.

26
00:01:58,241 --> 00:02:01,221
So here's what a day looks like now.

27
00:02:01,441 --> 00:02:11,401
I'm on a call with one client at the same time an agent is drafting a full cash flow analysis or an insurance analysis for a different client.

28
00:02:12,001 --> 00:02:20,121
When the call ends, I open that draft, I fully review it, adjust what needs to be adjusted, and send the client their recommendations.

29
00:02:20,521 --> 00:02:23,181
I want to be really clear about that last step.

30
00:02:23,381 --> 00:02:25,721
The agent doesn't send anything.

31
00:02:26,041 --> 00:02:31,461
A planner reviews every single recommendation before a client sees it.

32
00:02:31,701 --> 00:02:37,121
What changed is that the first draft no longer needs me sitting in front of it.

33
00:02:37,641 --> 00:02:41,121
Getting there took a lot more than a good prompt, however.

34
00:02:41,401 --> 00:02:42,901
Two things had to be true.

35
00:02:43,261 --> 00:02:47,321
First, the agent has to work where the work lives,

36
00:02:47,541 --> 00:02:51,121
not in a separate chat window I'm copying and pasting it into.

37
00:02:51,461 --> 00:02:54,301
Second, it has to know how we do it,

38
00:02:54,421 --> 00:02:57,001
not how financial planning works in general.

39
00:02:57,221 --> 00:02:59,021
The model already knows that,

40
00:02:59,141 --> 00:03:02,781
but the agent needs to know how range does it,

41
00:03:02,901 --> 00:03:05,141
what we look at first, what we always flag,

42
00:03:05,361 --> 00:03:07,581
how we phrase a specific recommendation.

43
00:03:07,921 --> 00:03:10,361
That combination is what a skill is.

44
00:03:10,361 --> 00:03:15,001
our firm's way of doing one specific piece of work written

45
00:03:15,021 --> 00:03:17,481
down so an agent can follow it.

46
00:03:18,821 --> 00:03:31,181
The bulk of this talk is how you build one. Five steps. Pick, teach, pack, test, and then ship. None of them require writing code.

47
00:03:31,181 --> 00:03:35,421
Every one of them requires knowing the work really well,

48
00:03:35,601 --> 00:03:40,461
which is exactly why the people doing the work should also be the ones building it.

49
00:03:42,361 --> 00:03:45,321
Step one, pick a workflow you repeat.

50
00:03:45,581 --> 00:03:49,321
Your initial instinct is to go for the flashiest work,

51
00:03:49,421 --> 00:03:51,841
whatever's taking the most amount of time and the most complex.

52
00:03:52,081 --> 00:03:57,501
Don't. Pick something you do every week where you already know what good looks like,

53
00:03:57,541 --> 00:04:00,121
where you can spot a bad draft at a glance.

54
00:04:00,721 --> 00:04:09,201
Additionally, we recommend choosing some of the least complex workflows to start, but some that require the most manual work.

55
00:04:09,361 --> 00:04:14,401
For us, that meant things like cash flow analysis and insurance reviews.

56
00:04:14,661 --> 00:04:18,561
Just for us here at Range, that required the most manual work.

57
00:04:18,721 --> 00:04:23,841
And if you can't judge the output quickly, you can't test it and you can't trust it.

58
00:04:24,021 --> 00:04:28,841
So look for two things. You do it often and you know how to judge it.

59
00:04:30,381 --> 00:04:35,521
Step two, show it several real cases for previous plans you've delivered.

60
00:04:35,761 --> 00:04:41,421
Take several actual client cases and walk through the work start to finish.

61
00:04:41,621 --> 00:04:44,641
And as you do it, narrate your judgment out loud.

62
00:04:44,801 --> 00:04:48,641
Not just, next I open the pay stub, why you're opening it.

63
00:04:49,081 --> 00:04:50,461
What you're looking for.

64
00:04:50,701 --> 00:04:53,461
What would make you stop and ask a question.

65
00:04:53,641 --> 00:04:56,281
That narration is the whole game.

66
00:04:56,461 --> 00:04:58,041
The steps are easy to write down.

67
00:04:58,041 --> 00:05:04,001
the judgment is what makes it your firm's work instead of generic advice.

68
00:05:05,721 --> 00:05:13,021
Step three, pack it into a skill. A skill isn't just a list of steps. Two things matter most.

69
00:05:13,141 --> 00:05:19,381
The questions it must ask. Before it starts, what does it need to know? What should it confirm with

70
00:05:19,381 --> 00:05:25,981
me instead of simply guessing? And the limits it must respect. What it never does on its own.

71
00:05:25,981 --> 00:05:30,581
In our world, the big one is that it never sends anything to a client.

72
00:05:30,761 --> 00:05:35,801
It drafts, and then a planner decides if it gets sent or needs to be adjusted.

73
00:05:36,061 --> 00:05:40,761
If you only write the steps, you get something that simply just runs on its own.

74
00:05:40,921 --> 00:05:45,401
If you write the questions and set the limits, you can get something you can trust.

75
00:05:45,741 --> 00:05:47,721
Think of a skill as a recipe.

76
00:05:48,041 --> 00:05:53,181
Once Claude knows how to do the skill, it will repeat your process on command.

77
00:05:54,021 --> 00:06:01,241
Step number four, run it five or 10 times on real cases before anyone else sees it.

78
00:06:01,381 --> 00:06:04,121
This is the step everyone wants to skip.

79
00:06:04,301 --> 00:06:12,681
Every run teaches you something the skill didn't know yet, an edge case, a question it should have asked, a spot where it sounded confident and was wrong.

80
00:06:12,881 --> 00:06:18,381
You need to fix it, save the skill, and run it again and again and again.

81
00:06:18,381 --> 00:06:28,861
5-10 runs sounds like a lot, but it's a lot less work than cleaning up after you hand a half-baked skill to 15, 20, 100 people in your team.

82
00:06:30,141 --> 00:06:34,601
Step 5. Ship it with an owner and a reviewer.

83
00:06:34,761 --> 00:06:43,301
The owner keeps it current. Our processes change, tax rules change, and a skill that was right in March can be wrong by December.

84
00:06:43,301 --> 00:06:50,201
Remember, the reviewer checks it before the team relies on it, and it's someone other than the author.

85
00:06:50,621 --> 00:06:53,581
Same as a plan. You don't grade your own work.

86
00:06:53,761 --> 00:06:58,221
Once other people depend on a skill, somebody has to be accountable for it.

87
00:06:58,341 --> 00:07:04,441
That's the difference between a clever personal shortcut and something a team can build on.

88
00:07:05,261 --> 00:07:07,201
So where are we at currently?

89
00:07:07,201 --> 00:07:13,281
Six months in, we have 24 skills across the company and 17 of them are in planning.

90
00:07:13,461 --> 00:07:15,521
And here's the part I didn't expect.

91
00:07:15,761 --> 00:07:20,041
It spread to sales, product, marketing, all without a mandate.

92
00:07:20,301 --> 00:07:21,541
Nobody sent a memo.

93
00:07:21,921 --> 00:07:25,821
People saw what the planners were doing and started building their own agents.

94
00:07:26,801 --> 00:07:34,681
Every one of those skills was written and is maintained by planners, marketers, and salespeople, not developers or engineers.

95
00:07:36,101 --> 00:07:42,581
A couple of things changed once subject matter experts started building for their own niche.

96
00:07:42,761 --> 00:07:44,981
The skills got a lot more specific.

97
00:07:45,261 --> 00:07:49,581
The person who knows insurance, for example, best builds the insurance skill.

98
00:07:49,821 --> 00:07:55,161
The person who best knows tax, tax code, tax law should be building the tax one.

99
00:07:55,441 --> 00:07:57,521
And a review came naturally.

100
00:07:57,781 --> 00:08:01,481
We already review each other's plans, and that's just how planning works.

101
00:08:01,481 --> 00:08:08,841
So now we review each other's skills in that same way. It's the same habit, just a new thing to review.

102
00:08:10,181 --> 00:08:20,821
So why does this matter right now? Because the bottleneck has moved. Whether a model can draft a cash flow analysis, that question's already been settled. It can.

103
00:08:21,141 --> 00:08:30,501
What's not settled, who writes down how your firm does it? Who tests it before it ships? Who owns it once 15 other people depend on it?

104
00:08:30,501 --> 00:08:37,741
If you're being asked to push agents out past the engineering org, you're about to hit these exact questions.

105
00:08:38,041 --> 00:08:42,921
There isn't much field data on this yet, but this is how we've done it here at Range.

106
00:08:45,361 --> 00:08:48,321
So that's it. Thank you so much for listening to this talk.

107
00:08:48,541 --> 00:08:52,981
I'm Matt Ragone from Range, and you can reach me at mattatrange.com.

108
00:08:53,101 --> 00:08:53,781
Have a great day.