WEBVTT

00:00:00.200 --> 00:00:07.599
You provide the idea.

00:00:08.250 --> 00:00:17.600
Your E.P. assigns the tasks.

00:00:18.250 --> 00:00:24.600
You make the call.

00:00:25.000 --> 00:00:29.800
Brava by mevanu

00:00:30.050 --> 00:00:31.829
Hi, I’m Anthony from Mevanu.

00:00:32.380 --> 00:00:36.049
In our previous build update, Alice asked what would count as learning.

00:00:36.630 --> 00:00:41.469
This week, we’re asking when learning requires us to stop asking for one more fix.

00:00:42.170 --> 00:00:45.000
There’s a lively argument about keeping a “human in the loop”—

00:00:45.469 --> 00:00:48.329
a person who reviews or directs what an AI does.

00:00:48.850 --> 00:00:53.560
Critics see an exhausting future: machines produce work faster than people can check it,

00:00:53.939 --> 00:00:59.409
and the human ends up clicking Approve, carrying responsibility for decisions they barely understand.

00:01:00.079 --> 00:01:07.129
The counterargument is that consequential work still needs someone who can examine the evidence, challenge the goal and stop an action.

00:01:07.719 --> 00:01:10.520
Removing that person does not make the system trustworthy.

00:01:11.319 --> 00:01:13.879
Both positions ask something important of the design.

00:01:14.479 --> 00:01:16.740
How much can the reviewer reasonably judge?

00:01:17.189 --> 00:01:19.170
And what power do they have when they disagree?

00:01:19.859 --> 00:01:22.250
That became concrete in our Brava engineering this week.

00:01:22.849 --> 00:01:28.870
One ordinary quit-and-reopen test passed independent review: the saved brief and work were preserved.

00:01:29.460 --> 00:01:31.140
That gives us a result we can build on.

00:01:31.870 --> 00:01:35.670
Elsewhere, a navigation control did not visibly open its panel.

00:01:36.170 --> 00:01:40.719
We had not established whether the automated click actually reached the intended control.

00:01:41.239 --> 00:01:46.680
The app, the testing method and the observation were still different possible sources of the problem.

00:01:47.269 --> 00:01:49.689
Another code change could have repaired the wrong thing.

00:01:50.299 --> 00:01:55.329
We paused further device checks to get a direct observation before deciding what to change.

00:01:55.930 --> 00:01:58.129
A repair loop can look remarkably productive.

00:01:58.659 --> 00:02:00.790
Each attempt creates something new to inspect.

00:02:01.409 --> 00:02:03.640
Eventually, even the fixes need fixing.

00:02:04.319 --> 00:02:06.260
Sometimes the answer is a small correction.

00:02:06.799 --> 00:02:08.519
Sometimes it is a different test.

00:02:09.040 --> 00:02:14.340
Sometimes we need to step back and reconsider a design that has become too difficult to reason about.

00:02:14.889 --> 00:02:17.310
More attempts alone cannot tell us which.

00:02:17.980 --> 00:02:19.800
The human can prolong the loop too.

00:02:20.340 --> 00:02:23.209
“Try again” is an instruction, not evidence.

00:02:23.670 --> 00:02:27.300
That includes everyone directing this work, human and agent alike.

00:02:27.969 --> 00:02:30.930
Useful oversight needs to leave room for a different question:

00:02:31.490 --> 00:02:33.709
what would convince us that our approach is wrong?

00:02:34.379 --> 00:02:38.569
An agent can help investigate that question, propose alternatives and test them.

00:02:39.159 --> 00:02:44.819
The person directing the work needs enough evidence to choose among them—and permission to change course.

00:02:45.270 --> 00:02:47.460
That is the partnership we want to build.

00:02:47.939 --> 00:02:49.830
Routine checks should do their share.

00:02:50.219 --> 00:02:57.000
Human attention should go where it can change a consequential decision, including the decision to simplify the work itself.

00:02:57.060 --> 00:03:01.680
Since we recorded this update, a direct human click has opened the navigation panel.

00:03:02.189 --> 00:03:05.129
The automated interaction remains under investigation.

00:03:05.520 --> 00:03:08.219
That gives us a better basis for deciding what to change.

00:03:08.639 --> 00:03:14.849
The larger goal remains a complete everyday task, followed through to its result and independently assessed.

00:03:15.240 --> 00:03:16.889
Savings will need evidence too.

00:03:17.250 --> 00:03:21.120
Have an idea? Drop a prompt, comment, suggestion, or question.

00:03:21.479 --> 00:03:22.139
Mevanu.

00:03:22.409 --> 00:03:23.909
Just. Smarter.
