WEBVTT

00:00:00.000 --> 00:00:01.840
Your AI agent can write the message.

00:00:02.480 --> 00:00:03.740
Who gave it permission to hit Send?

00:00:04.400 --> 00:00:07.040
I’m Jimmy Standaert. Let me show you how
we approach that

00:00:07.040 --> 00:00:11.220
question with whitewater.ai. Imagine a
customer asks for pricing.

00:00:11.880 --> 00:00:15.320
Your AI finds an old proposal, writes a
convincing answer, and

00:00:15.320 --> 00:00:16.260
gets ready to send it.

00:00:16.700 --> 00:00:17.540
The writing looks good.

00:00:18.060 --> 00:00:19.300
The price could still be wrong.

00:00:19.900 --> 00:00:22.980
Before your AI agent hits Send, ask six
questions.

00:00:23.600 --> 00:00:24.240
What is the task?

00:00:24.900 --> 00:00:25.640
What can it access?

00:00:26.340 --> 00:00:27.300
Who approved this action?

00:00:27.920 --> 00:00:29.060
What evidence supports it?

00:00:29.060 --> 00:00:30.860
When must a human step in?

00:00:31.380 --> 00:00:34.280
And can we reconstruct what happened if
something goes wrong?

00:00:34.760 --> 00:00:36.480
Here’s how those questions become a
workflow.

00:00:37.600 --> 00:00:41.040
First, define the job: prepare a pricing
response for this customer.

00:00:41.640 --> 00:00:43.580
That gives us a specific action to
evaluate.

00:00:44.660 --> 00:00:45.920
Second, identify the evidence.

00:00:46.620 --> 00:00:49.000
The current approved price list should
govern the quote.

00:00:49.600 --> 00:00:52.440
An old proposal can provide background,
but its price needs to

00:00:52.440 --> 00:00:54.500
be checked. Third, set the permissions.

00:00:55.340 --> 00:00:58.400
In this example, the agent can read
approved documents and prepare

00:00:58.400 --> 00:01:02.380
a draft. Changing prices or offering a
discount requires a person’s

00:01:02.380 --> 00:01:05.000
approval. Fourth, define the stopping
point.

00:01:05.560 --> 00:01:08.560
If the agent finds conflicting prices or
can’t verify the source,

00:01:09.020 --> 00:01:11.400
it should pause and bring the issue to
the person responsible.

00:01:12.120 --> 00:01:15.960
Finally, keep a record: the source used,
the draft, the approval,

00:01:16.420 --> 00:01:18.080
and whether the message was actually
sent.

00:01:18.620 --> 00:01:20.560
That gives your team something concrete
to review.

00:01:21.520 --> 00:01:22.280
Then test the workflow.

00:01:23.080 --> 00:01:24.120
Give it an expired quote.

00:01:24.640 --> 00:01:25.920
Ask for an unauthorized discount.

00:01:26.620 --> 00:01:30.200
Check whether the controls hold before
giving the agent more responsibility.

00:01:30.200 --> 00:01:34.320
That’s the practical work behind
whitewater.ai: connecting the task,

00:01:34.660 --> 00:01:38.840
evidence, permissions, human decisions,
and action history around the AI

00:01:38.840 --> 00:01:42.180
tools you use. You can apply these same
six questions to

00:01:42.180 --> 00:01:45.100
customer service, research, or updating
business records.

00:01:45.780 --> 00:01:47.500
Bring me one workflow you want to
improve.

00:01:48.160 --> 00:01:50.680
Message me “WORKFLOW” for a free
brainstorming session.

00:01:50.680 --> 00:01:52.340
whitewater.ai.