WEBVTT

NOTE Sentence-level transcript of https://www.youtube.com/watch?v=n97BCfyFIvw

NOTE One cue per sentence. Cue ids are the line anchors on /transcripts/n97BCfyFIvw.html. A cue ends where the next begins, or 2 s after its last word.

s1
00:00:01.309 --> 00:00:03.309
[music]

s2
00:00:20.880 --> 00:00:23.360
Howdy, folks.

s3
00:00:23.600 --> 00:00:29.240
So, good afternoon or good whatever time it is when you're watching this on YouTube.

s4
00:00:29.240 --> 00:00:31.600
I'm really excited to be here.

s5
00:00:31.600 --> 00:00:40.560
And um today I want to talk to you about really uh what it takes to keep the human in the loop where engineering is concerned.

s6
00:00:40.560 --> 00:00:44.960
I really want to start with a human side before we talk about the architecture here.

s7
00:00:44.960 --> 00:00:54.680
I think that the engineer of the future is going to be really defined by the person who is able to choose what is worth doing.

s8
00:00:54.680 --> 00:00:55.960
They're going to own the evidence.

s9
00:00:55.960 --> 00:00:59.880
They're going to own the understanding as well as the verdict.

s10
00:00:59.880 --> 00:01:04.160
around increasingly automated work that's being done by agents.

s11
00:01:04.160 --> 00:01:10.240
Now, when I use the term verdict, I don't mean that we're suddenly all going to be Judge Judy.

s12
00:01:10.240 --> 00:01:11.600
We're not.

s13
00:01:11.600 --> 00:01:15.360
But what I mean really is something just a little bit different.

s14
00:01:15.360 --> 00:01:19.840
I mean, we're going to be accountable for the production decisions.

s15
00:01:19.840 --> 00:01:21.040
Does something ship?

s16
00:01:21.040 --> 00:01:22.240
Do we block it?

s17
00:01:22.240 --> 00:01:25.640
Do we redirect it or accept the risk?

s18
00:01:25.640 --> 00:01:30.520
Quality is something that we all talk about a lot, but quality produces evidence.

s19
00:01:30.520 --> 00:01:33.880
A verdict assigns responsibility.

s20
00:01:33.880 --> 00:01:38.920
And answerability is really what lets us stand behind a verdict.

s21
00:01:38.920 --> 00:01:46.560
And this of course is not the only way that our industry is starting to think about our roles evolving.

s22
00:01:46.640 --> 00:01:51.800
Boris Cherney recently put some useful language around what many teams are starting to feel.

s23
00:01:51.800 --> 00:01:58.000
The old craft boundaries are getting blurry, and roles are re-bumbling around the work itself.

s24
00:01:58.000 --> 00:02:06.640
And the important question here becomes a lot less about what is your title and more what part of the system can you own?

s25
00:02:06.640 --> 00:02:10.560
Now, I like this taxonomy quite a lot.

s26
00:02:10.560 --> 00:02:13.520
Um it's optimistic without being overly vague.

s27
00:02:13.520 --> 00:02:17.200
So, things like prototype, build, sweep, grow, and maintain.

s28
00:02:17.200 --> 00:02:19.920
And these are real engineering modes.

s29
00:02:19.920 --> 00:02:25.040
Agents are going to help with all of them, but the scarce thing is not merely doing the task.

s30
00:02:25.040 --> 00:02:33.360
It's going to be knowing which mode your product needs and what quality bar applies and who owns the result at the end of the day.

s31
00:02:33.360 --> 00:02:38.959
Now, we've been talking about harnesses and loop engineering and software factories over the last couple of days.

s32
00:02:38.959 --> 00:02:41.360
We can talk why this shift is happening.

s33
00:02:41.360 --> 00:02:44.640
We moved past the model as the whole story, right?

s34
00:02:44.640 --> 00:02:49.520
With harness engineering, the coding agent is the model plus the harness around it, right?

s35
00:02:49.520 --> 00:02:52.400
Your context, your tools, your file system, Git.

s36
00:02:52.400 --> 00:02:56.480
And the harness is what turns intelligence into something that you can delegate to.

s37
00:02:56.480 --> 00:03:01.000
The next move was loop engineering, where we weren't just prompting one run anymore.

s38
00:03:01.000 --> 00:03:07.000
We were designing systems that kept prompting, checking, and remembering, and deciding what happened next.

s39
00:03:07.000 --> 00:03:10.840
And that's really when agents [clears throat] started to feel like infrastructure.

s40
00:03:10.840 --> 00:03:14.959
And once you start putting all of those things together, you get that software factory.

s41
00:03:14.959 --> 00:03:22.560
Dex covered this well in his talk, where you have agents that are running inside that inner loop and evidence that comes out.

s42
00:03:22.560 --> 00:03:26.760
Humans still end up making the production decisions in this loop.

s43
00:03:26.760 --> 00:03:29.680
And the wind really isn't moving us from it.

s44
00:03:29.680 --> 00:03:34.800
The wind is moving human judgments the highest leverage checkpoint, I think.

s45
00:03:34.800 --> 00:03:37.120
And this is why it starts to matter now.

s46
00:03:37.120 --> 00:03:42.360
AI-generated and AI-assisted code is becoming normal code for a lot of us.

s47
00:03:42.360 --> 00:03:47.519
One of Sonar's 2026 survey said that AI-assisted code is no longer marginal.

s48
00:03:47.519 --> 00:03:51.320
It's increasingly having a large role in our code bases.

s49
00:03:51.320 --> 00:03:55.360
And once that happens, answerability stops being this philosophical world.

s50
00:03:55.360 --> 00:03:57.760
It becomes an engineering requirement.

s51
00:03:57.760 --> 00:03:59.840
And there's a quality point here as well, right?

s52
00:03:59.840 --> 00:04:04.200
Like we used to care about clean code, code that people could read.

s53
00:04:04.200 --> 00:04:09.160
But cleaner code is actually not just going to help the next human and the next person on your teams.

s54
00:04:09.160 --> 00:04:10.880
It actually helps the next agent.

s55
00:04:10.880 --> 00:04:21.000
Another one of Sonar's research uh studies found that clean and messy repos had roughly the same pass rates, but clean code actually used fewer tokens and caused fewer revisits.

s56
00:04:21.000 --> 00:04:26.240
So there's a lot of benefit to maintainability that can fuel efficiency for your factories.

s57
00:04:26.240 --> 00:04:31.760
Now, making generation cheaper does not automatically make review cheaper, right?

s58
00:04:31.760 --> 00:04:36.600
I think a lot of us are facing this moment and we know that engineers are not naive.

s59
00:04:36.600 --> 00:04:41.200
The Sonar numbers say that almost everybody is skeptical of AI code.

s60
00:04:41.200 --> 00:04:43.320
Now, I love working in my software factory.

s61
00:04:43.320 --> 00:04:45.840
I love building my engineering loops.

s62
00:04:45.840 --> 00:04:48.240
But the problem is still capacity.

s63
00:04:48.240 --> 00:04:58.600
If 96% of people don't fully trust that code, but only about half always verify before committing, we have this danger that we've got distrust without bandwidth.

s64
00:04:58.600 --> 00:05:04.000
And so safety comes from making verification cheaper, clearer, and harder for people to skip.

s65
00:05:04.000 --> 00:05:18.320
And if you zoom out from the individual reviewer to the organization, review and validation start becoming a bottleneck when governance isn't able to catch up and adoption is already moving way faster than any company can go and set their policies.

s66
00:05:18.320 --> 00:05:23.720
And this means that we have some hard questions we have to deal with, like did a model actually touch this file?

s67
00:05:23.720 --> 00:05:27.760
And the hard questions are also like what constraints guided that work?

s68
00:05:27.760 --> 00:05:29.800
What evidence was produced?

s69
00:05:29.800 --> 00:05:33.000
What risk was accepted and who owned the result?

s70
00:05:33.000 --> 00:05:38.040
Now, the agent can ship more than any of us can review, right?

s71
00:05:38.040 --> 00:05:40.280
So what are we still good for?

s72
00:05:40.280 --> 00:05:43.640
I think it's a question that's on a lot of our minds, right?

s73
00:05:43.640 --> 00:05:51.080
And you know, if Homer Simpson's experience automating computers can teach us anything, maybe this is our future.

s74
00:05:51.080 --> 00:05:52.640
I don't think it is.

s75
00:05:52.640 --> 00:05:55.320
But it's one direction things could take.

s76
00:05:55.320 --> 00:05:56.960
Now, let's try that again.

s77
00:05:56.960 --> 00:06:06.200
If change is where humans enter the loop, if generation scales faster than comprehension, the scarce resource becomes judgment that's backed by evidence.

s78
00:06:06.200 --> 00:06:13.600
So, the question is no longer how much can the agent do, but where does human judgment still create leverage?

s79
00:06:13.600 --> 00:06:22.320
Now, I want to talk to you about two terms that I'm going to use for the career part of this talk, alpha and decay.

s80
00:06:22.320 --> 00:06:28.440
Alpha is the gap between what you can do today and what current models can do.

s81
00:06:28.440 --> 00:06:34.400
That gap is a very real thing, and decay is the clock on that gap.

s82
00:06:34.400 --> 00:06:41.440
If the thing that makes you special is a capability, the frontier is eventually going to come for it.

s83
00:06:41.440 --> 00:06:42.320
Right?

s84
00:06:42.320 --> 00:06:44.400
And there's a whole conversation around this.

s85
00:06:44.400 --> 00:06:47.280
This is one of the reasons why taste keeps coming up.

s86
00:06:47.280 --> 00:06:49.720
Paul Graham had a point here that I think is very right.

s87
00:06:49.720 --> 00:06:55.800
When anyone can make anything, choosing what to make becomes very important, and I buy that.

s88
00:06:55.800 --> 00:07:05.840
But I also think that we have to be very careful because taste can become a magic word for whatever part of the work we don't want to explain just yet.

s89
00:07:05.840 --> 00:07:09.800
Mitchell Hashimoto gave us a more useful version of this definition.

s90
00:07:09.800 --> 00:07:16.520
Taste is the ability to make high-quality qualitative judgments where no objective metric exists yet.

s91
00:07:16.520 --> 00:07:22.320
That matters because it puts taste before the benchmark and before the market has fully voted.

s92
00:07:22.320 --> 00:07:34.720
When you try out a model and you see the kind of UX and the kind of experiences that it builds, you can often tell when you think it has taste or lacks taste or where there's a gap there that humans can fill.

s93
00:07:34.720 --> 00:07:43.160
Now, this is also only useful if we can turn some of this concept around taste into critique, examples, and better judgment over time.

s94
00:07:43.160 --> 00:07:47.800
So, yes, taste matters when production gets cheaper.

s95
00:07:47.800 --> 00:07:54.920
And if anyone can generate 10 options, the scarce skill is really knowing which option deserves to exist.

s96
00:07:54.920 --> 00:07:58.560
But taste is not some eternal moat.

s97
00:07:58.560 --> 00:08:00.920
It's alpha as well.

s98
00:08:00.920 --> 00:08:03.440
Now, the people with taste are still going to matter.

s99
00:08:03.440 --> 00:08:06.360
I personally think they're still going to matter for a long time.

s100
00:08:06.360 --> 00:08:11.160
But the best version of that skill is not mystique.

s101
00:08:11.160 --> 00:08:17.440
It's making better calls and leaving behind examples that your team and the system can learn from.

s102
00:08:17.440 --> 00:08:19.680
Now, let's apply the decay test.

s103
00:08:19.680 --> 00:08:21.600
Well, we used to have speed.

s104
00:08:21.600 --> 00:08:22.680
That decayed.

s105
00:08:22.680 --> 00:08:24.160
We used to have recall.

s106
00:08:24.160 --> 00:08:26.080
You know, harnesses have memory.

s107
00:08:26.080 --> 00:08:31.160
Verification is moving into harnesses, evals, static checks, and model critique.

s108
00:08:31.160 --> 00:08:38.560
Taste, I continue to think this is going to decay much more slowly, but it still resets as models learn from examples and preferences.

s109
00:08:38.560 --> 00:08:41.680
Even judgment in some ways is a slope rather than a wall.

s110
00:08:41.680 --> 00:08:44.960
So, the strategy is not to cling to any one capability.

s111
00:08:44.960 --> 00:08:48.320
It's for us to keep moving our edges up a level.

s112
00:08:48.320 --> 00:08:54.200
So, this is one of the reasons why what can the agent do is not the best strategic question anymore.

s113
00:08:54.200 --> 00:08:57.960
The list of things that agents can't do just keeps shrinking.

s114
00:08:57.960 --> 00:09:03.120
The better question for us is really what can only a human be answerable for.

s115
00:09:03.120 --> 00:09:10.760
Not because, you know, any of us are are magical in any way, but because some decisions actually require ownership.

s116
00:09:10.760 --> 00:09:16.920
They require context, risk acceptance, and responsibility after that work ships.

s117
00:09:16.920 --> 00:09:21.440
This is why the word engineer has to get just a little bit stricter.

s118
00:09:21.440 --> 00:09:26.120
More people than ever can now make computers do things, and I think that's truly awesome.

s119
00:09:26.120 --> 00:09:31.360
The total addressable market for builders has never been larger, and that's so cool.

s120
00:09:31.360 --> 00:09:33.520
But it's a huge expansion of leverage.

s121
00:09:33.520 --> 00:09:39.360
An engineer is not merely somebody who can code, you know, and and get things to exist.

s122
00:09:39.360 --> 00:09:41.360
An engineer can reason about systems.

s123
00:09:41.360 --> 00:09:48.600
They think about constraints, you defend trade-offs, you can manage risk, and you're the person that can be reached out to when things start to break.

s124
00:09:48.600 --> 00:09:54.840
So, what are things that engineers should avoid if we want to stay effective and accountable in this moment?

s125
00:09:54.840 --> 00:09:59.080
Well, the first thing to avoid really is cognitive debt.

s126
00:09:59.080 --> 00:10:05.480
Now, cognitive debt is the erosion of your understanding and memory around how to solve problems.

s127
00:10:05.480 --> 00:10:09.320
I think a lot of us start to feel this the more that we're using agents every single day.

s128
00:10:09.320 --> 00:10:11.040
I know that I feel this a lot.

s129
00:10:11.040 --> 00:10:15.400
And it's because we're deferring more and more to AI to solve our problems.

s130
00:10:15.400 --> 00:10:23.480
For code, it's the gap between how much code exists in your repo and how much any human on your team genuinely understands.

s131
00:10:23.480 --> 00:10:26.440
And this is why things like delegation debt end up mattering.

s132
00:10:26.440 --> 00:10:38.280
You can have a build that passes, you know, your tests, a PR that you can merge, but your team can still end up losing its ability to actually explain the system that they are shipping to production.

s133
00:10:38.280 --> 00:10:41.920
Now, a very real pressure is all much is also how much we delegate.

s134
00:10:41.920 --> 00:10:47.480
So, agents can now stay inside the system long enough for the human to lose the thread.

s135
00:10:47.480 --> 00:10:56.400
So, a 30-second run, right, can feel like an interaction, but an hour or a day-scale task, so something long horizon, that's a work stream.

s136
00:10:56.400 --> 00:11:04.600
And when tasks can end up, you know, lasting that long, especially when you begin running many of them in parallel, review can't just be a glance at the end.

s137
00:11:04.600 --> 00:11:07.600
It has to become a whole control system.

s138
00:11:07.600 --> 00:11:11.000
The second thing to avoid is cognitive surrender.

s139
00:11:11.000 --> 00:11:15.720
Now, this is when you blindly accept AI's um responses.

s140
00:11:15.720 --> 00:11:22.160
Like, delegation is important cuz delegation says, "Do the work, then show me enough evidence that I can judge it."

s141
00:11:22.160 --> 00:11:25.120
I still make a judgment in that situation.

s142
00:11:25.120 --> 00:11:31.560
Surrender is really saying, "Hey, your answer is now my answer before I have formed any opinions myself.

s143
00:11:31.560 --> 00:11:35.880
Now, um Wharton did a study that kind of offers us a warning light here.

s144
00:11:35.880 --> 00:11:44.680
When AI was wrong, 73% of people still thought that they they you know they picked the wrong answer and they felt more sure.

s145
00:11:44.680 --> 00:11:50.200
So, the failure mode is not using AI, but it's borrowed confidence.

s146
00:11:50.200 --> 00:11:53.520
The third thing to avoid is orchestration tax.

s147
00:11:53.520 --> 00:12:01.760
Now, if you've been in the Bay Area, you will see people who for better or worse are still walking around with their laptops open or talking to you about cloud agents.

s148
00:12:01.760 --> 00:12:09.920
And we're increasingly trying to run more and more and more in parallel or telling each other that we're shipping with hundreds of agents or thousands of agents.

s149
00:12:09.920 --> 00:12:14.960
More AI agents running does not mean that there is more of you available.

s150
00:12:14.960 --> 00:12:18.360
Your cognitive bandwidth does not parallelize.

s151
00:12:18.360 --> 00:12:25.040
So, every loop that you create ends up causing more decisions to route, merge, verify, and integrate.

s152
00:12:25.040 --> 00:12:31.240
And the fix is not necessarily fewer agents, but it's about designing your attention like a system.

s153
00:12:31.240 --> 00:12:34.600
Like where you enter, what you require, what you reuse.

s154
00:12:34.600 --> 00:12:37.960
You just want to be very intentional about it.

s155
00:12:37.960 --> 00:12:42.280
Now, accountability can be a scary word for a lot of people.

s156
00:12:42.280 --> 00:12:49.480
And I wouldn't be surprised if it made you want to go hide in the bushes and just tell your agent to deal with it.

s157
00:12:49.480 --> 00:12:52.760
But, accountability is not what remains after agents get good.

s158
00:12:52.760 --> 00:12:56.280
It's what lets the rest of the whole system scale.

s159
00:12:56.280 --> 00:13:06.000
If agents can do more work, if they can do it faster and parallel, better than what many of us could do, the scarce thing becomes the ability to explain

s160
00:13:06.000 --> 00:13:13.720
intent, to inspect evidence, to accept risk, and improve the system when the decision was wrong.

s161
00:13:14.960 --> 00:13:16.680
Now, here is the career math.

s162
00:13:16.680 --> 00:13:20.600
The half-life of an edge might be one model release.

s163
00:13:20.600 --> 00:13:25.839
Speed, recall, verification, even taste all move as the frontier moves.

s164
00:13:25.839 --> 00:13:32.040
But, the half-life of a signature, your credibility, your expertise is much longer.

s165
00:13:32.040 --> 00:13:41.040
And by signature, I really mean the name on the work, the person, the team, the institution, whoever stands behind what's actually shipped.

s166
00:13:41.040 --> 00:13:45.840
So, skills can earn leverage, accountability can turn leverage into trust.

s167
00:13:45.840 --> 00:13:48.960
And this is one of the lines that I want to draw pretty clearly.

s168
00:13:48.960 --> 00:13:54.640
Agents can choose, they can route, they can merge, they can escalate, they can operate inside policy.

s169
00:13:54.640 --> 00:13:57.360
And in many systems, you know, they can, they should.

s170
00:13:57.360 --> 00:14:00.400
But execution and responsibility are very different things.

s171
00:14:00.400 --> 00:14:04.680
The agent can follow your runbook, but it can't inherit the consequences.

s172
00:14:04.680 --> 00:14:08.000
When something fails, the question is, who understood the policy?

s173
00:14:08.000 --> 00:14:11.640
Who accepted the risk and who owns the blast radius?

s174
00:14:11.640 --> 00:14:18.160
High agency is something that a lot of us talk about these days as being like this thing that we're looking for when we're hiring.

s175
00:14:18.160 --> 00:14:21.680
High agency is actively taking ownership of your outcomes.

s176
00:14:21.680 --> 00:14:27.400
So, knowing when to delegate, when to inspect, when to stop, and when to put your name on the result.

s177
00:14:27.400 --> 00:14:31.839
High agency in this world is not I personally do everything.

s178
00:14:31.839 --> 00:14:33.440
You know, that version doesn't really scale.

s179
00:14:33.440 --> 00:14:38.480
It's not just hustle theater, but it's ownership with judgment attached.

s180
00:14:38.480 --> 00:14:42.920
This agency ladder tries to make that a little bit more concrete.

s181
00:14:42.920 --> 00:14:47.640
At the bottom, you've got someone that flags a problem and leaves it for the system.

s182
00:14:47.640 --> 00:14:52.480
Higher up, they execute, diagnose, propose, recommend, and resolve.

s183
00:14:52.480 --> 00:14:55.080
And the rare top movement is discernment.

s184
00:14:55.080 --> 00:14:59.200
You know, maybe you find a problem and you decide whether or not it's worth investing in.

s185
00:14:59.200 --> 00:15:01.760
Maybe it's not, and maybe you move on.

s186
00:15:01.760 --> 00:15:07.440
But when agents make more paths possible, agency is not chasing every single path.

s187
00:15:07.440 --> 00:15:12.839
It's really just deciding which paths deserve your ownership and attention.

s188
00:15:12.839 --> 00:15:15.120
So, translate that into an operating model.

s189
00:15:15.120 --> 00:15:18.440
Agents can run much more of the inner execution loop.

s190
00:15:18.440 --> 00:15:20.920
They can investigate, implement, test, and report.

s191
00:15:20.920 --> 00:15:25.600
I think that there's leverage in that, but that outer loop is still engineering.

s192
00:15:25.600 --> 00:15:30.120
So, deciding, verifying, approving, owning, that inner loop is capability.

s193
00:15:30.120 --> 00:15:33.000
The outer loop is agency.

s194
00:15:33.000 --> 00:15:35.320
And this is a boundary that I really care about.

s195
00:15:35.320 --> 00:15:37.720
Your agent returns evidence.

s196
00:15:37.720 --> 00:15:46.040
It returns diffs, tests, logs, rationale, traces, trajectories, screenshots, whatever the work itself requires.

s197
00:15:46.040 --> 00:15:48.560
But then the engineering really begins.

s198
00:15:48.560 --> 00:15:51.160
We decide whether the work was worth doing.

s199
00:15:51.160 --> 00:15:57.920
We verify whether the evidence is enough, and we approve or redirect or own what reaches production.

s200
00:15:57.920 --> 00:16:03.240
It doesn't matter if you're someone that's just working with a small number of agents, or whether you're working with thousands of agents.

s201
00:16:03.240 --> 00:16:06.080
I still very much think that these ideas apply.

s202
00:16:06.080 --> 00:16:09.560
So, the boundary is not human looks at AI output.

s203
00:16:09.560 --> 00:16:12.920
The boundary is evidence and responsibility.

s204
00:16:13.600 --> 00:16:15.360
So, here's an operational rule.

s205
00:16:15.360 --> 00:16:17.240
Explain it or don't ship it.

s206
00:16:17.240 --> 00:16:25.280
And it's not because humans have to type every line or read every line, but because someone has to understand the work well enough to defend it.

s207
00:16:25.280 --> 00:16:36.200
If you've ever worked in a large code base or an enterprise code base, some code bases have this concept of an owners file or certain subdirectories where there are people who are on the hook for that part of the system.

s208
00:16:36.200 --> 00:16:38.320
You can think about this in a very similar way.

s209
00:16:38.320 --> 00:16:42.960
Who is accountable for that part of your architecture and your code base?

s210
00:16:42.960 --> 00:16:52.160
Your model might write the code, and the question is really still whether you can explain those changes that the agent is shipping, whether you've got the evidence where you understand the risks.

s211
00:16:52.160 --> 00:16:54.880
Now, this is one of the things I want you to remember near the end.

s212
00:16:54.880 --> 00:16:58.760
Automation moves the floor for all of us.

s213
00:16:58.760 --> 00:17:12.560
Engineering continues to move up a level, and our new work might be loop design, evidence design, and brownfield stewardship, but fewer keystrokes doesn't mean less engineering over the next few years.

s214
00:17:12.560 --> 00:17:20.319
It means that there is more surface area that needs taste, verification, ownership, and ultimately care.

s215
00:17:20.319 --> 00:17:25.280
I don't think I've ever been more excited about the future of this field.

s216
00:17:25.280 --> 00:17:31.760
Every time that we have made it easier to write software, we've predicted that the world would need less of it.

s217
00:17:31.760 --> 00:17:33.880
And in fact, the opposite happened.

s218
00:17:33.880 --> 00:17:37.960
Higher-level languages happened, frameworks, cloud, low code.

s219
00:17:37.960 --> 00:17:40.720
The pattern always went the other way.

s220
00:17:40.720 --> 00:17:43.960
And when you lower the cost, latent demand ends up appearing.

s221
00:17:43.960 --> 00:17:48.560
Those ideas that people didn't think were feasible to build and get out there are suddenly unlocked.

s222
00:17:48.560 --> 00:17:51.480
And agents are going to do the same thing for a lot of people.

s223
00:17:51.480 --> 00:17:53.680
It's not going to remove engineering work.

s224
00:17:53.680 --> 00:17:59.440
It's going to move the bottleneck from can we build this to should this exist and can we answer for it?

s225
00:17:59.440 --> 00:18:03.680
So, build the factories, keep the lights on, own the verdict.

s226
00:18:03.680 --> 00:18:05.040
I hope this was useful.

s227
00:18:05.040 --> 00:18:07.240
Thank you.

s228
00:18:23.032 --> 00:18:25.032
[music]