WEBVTT

NOTE Sentence-level transcript of https://www.youtube.com/watch?v=RGe6EjucbzI

NOTE One cue per sentence. Cue ids are the line anchors on /transcripts/RGe6EjucbzI.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
00:00:13.040 --> 00:00:18.240
Now joining us on stage is the partner at Amplify Barren.

s3
00:00:37.360 --> 00:00:38.879
Fantastic.

s4
00:00:38.879 --> 00:00:40.399
You did a great job practicing.

s5
00:00:40.399 --> 00:00:42.320
I feel very very loved.

s6
00:00:42.320 --> 00:00:44.559
Um, let's get started.

s7
00:00:44.559 --> 00:00:48.559
So, like you just heard, my name is Bar.

s8
00:00:48.559 --> 00:00:53.039
I run a survey every year on the state of AI engineering.

s9
00:00:53.039 --> 00:01:01.280
And the funny thing about running a survey on the state of AI engineering is that the field changes as you make the slides.

s10
00:01:01.280 --> 00:01:06.159
Just in the past week, we've had Frontier releases treated like national security events.

s11
00:01:06.159 --> 00:01:10.080
Meta reportedly exploring selling AI compute.

s12
00:01:10.080 --> 00:01:12.799
By the time I get off stage, maybe something else will happen.

s13
00:01:12.799 --> 00:01:18.479
So, if I miss a major announcement while I'm up here, please come find me after.

s14
00:01:18.560 --> 00:01:29.119
But that's exactly why we run the survey every year to cut through the noise, take a moment, step back and understand what AI engineers are actually doing.

s15
00:01:29.119 --> 00:01:35.840
Uh for the first time this year, we were thrilled to partner with Notion and Verscell to run this survey.

s16
00:01:36.640 --> 00:01:40.320
Very quickly on me, uh this is the least interesting slide.

s17
00:01:40.320 --> 00:01:42.720
I'm an investment partner at Amplify.

s18
00:01:42.720 --> 00:01:46.960
Very lucky to invest in companies built by and for AI engineers.

s19
00:01:46.960 --> 00:01:53.600
And I'll make the same promise that I make every single year, which is short time on bar, long time on bar charts.

s20
00:01:53.600 --> 00:01:58.240
So, let's get right into it with lots of bar charts.

s21
00:01:59.360 --> 00:02:02.799
First, let's talk about well, maybe raise your hand.

s22
00:02:02.799 --> 00:02:03.920
Did you fill out the survey?

s23
00:02:03.920 --> 00:02:05.680
This is a very large group.

s24
00:02:05.680 --> 00:02:06.000
Okay.

s25
00:02:06.000 --> 00:02:07.600
Yes, I see you in the front.

s26
00:02:07.600 --> 00:02:12.319
Um, if the answer is you, thank you so much.

s27
00:02:12.319 --> 00:02:15.760
If the answer is not you, I will find you in 2027.

s28
00:02:15.760 --> 00:02:19.280
But genuinely, this only exists because a thousand of you gave your time.

s29
00:02:19.280 --> 00:02:20.800
So, thank you.

s30
00:02:20.800 --> 00:02:26.879
We had 1,048 respondents this year, which is a lot of AI engineers.

s31
00:02:26.879 --> 00:02:31.440
And to be precise, this is not just AI engineers, as I'm sure you see at the conference.

s32
00:02:31.440 --> 00:02:35.519
Every year, we see that AI engineering is more of a discipline than a job title.

s33
00:02:35.519 --> 00:02:43.120
It touches founders, CTO's, engineers, product people, folks across company sizes and experience levels.

s34
00:02:43.120 --> 00:02:45.920
And that range shows up in experience too.

s35
00:02:45.920 --> 00:02:53.040
Um for the third year running we see the same pattern which is skew towards senior engineers but newer to AI.

s36
00:02:53.040 --> 00:03:06.080
Of those with over 10 years of software experience over half have three years or less of AI experience which tracks uh these are very experienced engineers learning a new paradigm in real time.

s37
00:03:06.080 --> 00:03:16.319
And the newest cohort, the ones who just started uh engineering, the median new engineer has nearly as much AI experience as the median 10-year software veteran.

s38
00:03:16.319 --> 00:03:21.280
Uh so the newest engineers have never known software without this.

s39
00:03:21.680 --> 00:03:23.680
But doing AI doesn't mean one thing.

s40
00:03:23.680 --> 00:03:26.800
We talked about all these different titles, all these different roles.

s41
00:03:26.800 --> 00:03:36.480
Before we get into models and agents, I have a more basic question, which is when people say they're doing AI at work, what are they actually doing?

s42
00:03:36.480 --> 00:03:38.959
So, first up, like to start with the modalities.

s43
00:03:38.959 --> 00:03:43.200
We asked, which modalities are you actively building with at work?

s44
00:03:43.200 --> 00:03:44.799
Can anyone take a guess?

s45
00:03:44.799 --> 00:03:46.000
Text dominates.

s46
00:03:46.000 --> 00:03:46.400
I know.

s47
00:03:46.400 --> 00:03:47.840
Hold your applause.

s48
00:03:47.840 --> 00:04:00.159
Um, but one piece of this chart that I always find very interesting and I always look at is the ratio of nope, I'm not using this modality to I'm not using it, but I do plan to.

s49
00:04:00.159 --> 00:04:03.519
I call this the intent to adopt ratio.

s50
00:04:03.519 --> 00:04:10.159
Of the people who are not building with a modality today, how many say they plan to use it?

s51
00:04:10.159 --> 00:04:14.239
And audio has the strongest intent to adopt this year.

s52
00:04:14.239 --> 00:04:23.680
Among AI engineers who are not building with audio today, a whopping 56% say they plan to adopt it in the AI applications they build.

s53
00:04:23.680 --> 00:04:26.080
And this is not a brand new signal.

s54
00:04:26.080 --> 00:04:31.759
Last year audio also had the highest intent to adopt across modalities but 37%.

s55
00:04:31.759 --> 00:04:38.800
So audio continues to take the lead and have high interest but that interest is accelerating.

s56
00:04:38.800 --> 00:04:49.759
Now there has been an audio swing but if we look at what changed most from the last year in the survey the biggest jump is actually in people using image generation.

s57
00:04:49.759 --> 00:04:59.840
The share of respondents using generative AI for images and feeling really good about it doubled from 18% last year to 36% this year.

s58
00:04:59.840 --> 00:05:02.960
Makes sense if you look at what we launched in the same window.

s59
00:05:02.960 --> 00:05:13.280
Over the past year plus survey time, uh we've had models Nano Banana, Nano Banana 2, Chat GPT images 2.0. The products have gotten much better.

s60
00:05:13.280 --> 00:05:20.320
What used to feel like an efficient way to generate cursed hands is just increasingly becoming a part of real work.

s61
00:05:20.320 --> 00:05:27.120
Audio may have the strongest intent to adopt, but image generation shows us what happens when a modality crosses that threshold.

s62
00:05:27.120 --> 00:05:31.039
So, I'm excited to continue watching these adoption curves every single year.

s63
00:05:31.039 --> 00:05:34.320
I think we're going to see a lot this year.

s64
00:05:34.400 --> 00:05:35.919
Uh, now models.

s65
00:05:35.919 --> 00:05:39.440
If you've Who here spends time on Twitter?

s66
00:05:39.440 --> 00:05:39.919
All right.

s67
00:05:39.919 --> 00:05:40.240
Yes.

s68
00:05:40.240 --> 00:05:43.039
I imagine this is a very Twitter pilled uh crowd.

s69
00:05:43.039 --> 00:05:53.360
If you spend any time on Twitter in this uh in this circle, you've seen a lot written about openweight models these past few months and I think we'll see it even more in the next year.

s70
00:05:53.360 --> 00:05:58.560
Um so we asked what models are you actually using in production.

s71
00:05:58.560 --> 00:06:01.440
94% use closed models.

s72
00:06:01.440 --> 00:06:04.720
45% are using openweight models.

s73
00:06:04.720 --> 00:06:09.360
But here's the thing, you know, openweight models are not replacing closed models for the most part.

s74
00:06:09.360 --> 00:06:11.120
at least not yet.

s75
00:06:11.120 --> 00:06:16.720
The respondents using openweight models, over 90% of them are also using closed models.

s76
00:06:16.720 --> 00:06:19.360
So they're looking like an augmentation.

s77
00:06:19.360 --> 00:06:22.319
Teams are mixing and matching.

s78
00:06:23.919 --> 00:06:29.759
We also asked just to double click on this for the top three considerations when choosing a model.

s79
00:06:29.759 --> 00:06:32.560
If you're choosing a model, what is important to you?

s80
00:06:32.560 --> 00:06:38.160
Um and despite the airtime of the open versus closed, it's not what drives model choice.

s81
00:06:38.160 --> 00:06:43.440
It was a top three consideration for only 5% of the respondents.

s82
00:06:43.440 --> 00:06:45.520
What matters is actually more straightforward.

s83
00:06:45.520 --> 00:06:46.240
It's quality.

s84
00:06:46.240 --> 00:06:47.600
Quality dominates.

s85
00:06:47.600 --> 00:06:52.479
Followed by agentic capabilities like tool calling and cost tied right with it.

s86
00:06:52.479 --> 00:06:53.919
We'll money money.

s87
00:06:53.919 --> 00:06:55.440
We'll get back to that.

s88
00:06:55.440 --> 00:07:01.039
Um, one thing that I found very interesting is that reliability is not near the top.

s89
00:07:01.039 --> 00:07:03.919
Only one in five named reliability.

s90
00:07:03.919 --> 00:07:06.880
That doesn't mean teams stopped caring about reliability.

s91
00:07:06.880 --> 00:07:08.880
Uh there are different ways to interpret this data.

s92
00:07:08.880 --> 00:07:21.759
My guess is that it's more likely to become a threshold requirement and the models they're choosing are reliable enough so the decision moves up the stack outside of certain circumstances to quality, capability, cost,

s93
00:07:21.759 --> 00:07:24.000
but we could talk after.

s94
00:07:24.000 --> 00:07:28.639
All right, so here's where the model story all comes together.

s95
00:07:28.880 --> 00:07:33.199
Like I said, teams are not choosing one model and calling it a day.

s96
00:07:33.199 --> 00:07:38.080
Earlier I showed that 87% of teams are using more than one model.

s97
00:07:38.080 --> 00:07:41.680
Uh the model that's the opposite of standardization.

s98
00:07:41.680 --> 00:07:45.840
Uh and the way that they choose models for given tasks varies.

s99
00:07:45.840 --> 00:07:48.240
Most popular is routing by task type.

s100
00:07:48.240 --> 00:07:50.639
Some run multiple models compare outputs.

s101
00:07:50.639 --> 00:07:52.800
Some route based on cost.

s102
00:07:52.800 --> 00:07:55.280
Uh but models are good at different things.

s103
00:07:55.280 --> 00:08:02.240
What was interesting was that more than half of respondents said that their organizations starting to standardize on fewer AI tools.

s104
00:08:02.240 --> 00:08:06.240
They're trading standardiz flexibility for standardization.

s105
00:08:06.240 --> 00:08:07.440
A share of those are mixed.

s106
00:08:07.440 --> 00:08:11.919
They say they're standardizing on some layers while staying flexible on others.

s107
00:08:11.919 --> 00:08:20.000
But the headline here is that there's we're in the early great standardization of the platform and tools, not the models.

s108
00:08:20.400 --> 00:08:26.800
All right, this is the slide where anyone who's opened an AI bill in the last year starts nodding.

s109
00:08:26.800 --> 00:08:31.919
So it turns out that infinite intelligence still comes with a usagebased bill.

s110
00:08:31.919 --> 00:08:38.479
Once teams are managing many models and AI workflows, the next question becomes cost.

s111
00:08:38.800 --> 00:08:42.240
Cost is now a first class engineering constraint.

s112
00:08:42.240 --> 00:08:43.919
We see this in the data.

s113
00:08:43.919 --> 00:08:54.640
40% of respondents say that cost regularly shapes how ambitiously they use AI and another 36% say that it sometimes does.

s114
00:08:54.640 --> 00:08:55.920
Well, this is pretty straightforward.

s115
00:08:55.920 --> 00:09:05.839
So all in about uh three out of four respondents are adjusting their AI usage based on cost and maybe the fourth has a company card.

s116
00:09:06.959 --> 00:09:11.920
That might be surprising or maybe it's obvious but 12 months ago it was not.

s117
00:09:11.920 --> 00:09:14.160
Token maxing is cool.

s118
00:09:14.160 --> 00:09:21.600
Being able to find real use cases is amazing but cost is becoming a real big part of the product decision today.

s119
00:09:21.600 --> 00:09:24.000
And it shows up in monitoring too.

s120
00:09:24.000 --> 00:09:30.320
What folks are monitoring in production includes cost and token usage as the number two thing they watch for.

s121
00:09:30.320 --> 00:09:35.279
It's being monitored like an SLA right under quality itself.

s122
00:09:36.080 --> 00:09:39.120
Which brings us to the biggest line item of them all.

s123
00:09:39.120 --> 00:09:40.640
Agents.

s124
00:09:40.640 --> 00:09:42.880
We've been talking about agents for a while.

s125
00:09:42.880 --> 00:09:49.760
Uh this year, as you've seen, as you'll see today, as you've seen in previous days, you're going to talk a lot about harness engineering.

s126
00:09:49.760 --> 00:09:52.959
They're escaping demo world.

s127
00:09:53.040 --> 00:10:00.399
So, we asked respondents what level of tool permissions their agents typically have.

s128
00:10:00.399 --> 00:10:03.519
And this is where agents start to look more real.

s129
00:10:03.519 --> 00:10:05.680
There are two things happening at once.

s130
00:10:05.680 --> 00:10:13.360
First, and I don't think this is surprising, relative to last year, there are far more teams using agents.

s131
00:10:13.360 --> 00:10:21.040
This year, 95%, this seems high to me, 95% say they're using agents, roughly double last year.

s132
00:10:21.040 --> 00:10:28.640
Second, amongst the teams that are using agents, those agents are much more likely to have write access.

s133
00:10:28.640 --> 00:10:35.519
Last year, 52% of folks building with agents said their agents could actually write data.

s134
00:10:35.519 --> 00:10:38.399
This year, that number is 89%.

s135
00:10:38.399 --> 00:10:48.800
So when you combine these two shifts, more teams using agents and more of those agents having write permissions, the share of all the respondents and again it's a survey

s136
00:10:48.800 --> 00:10:53.519
using write enabled agents is up more than three times relative to last year.

s137
00:10:53.519 --> 00:10:55.600
So this is really the big shift.

s138
00:10:55.600 --> 00:10:58.160
Agents are no longer reading, summarizing, drafting.

s139
00:10:58.160 --> 00:11:00.720
They're taking actions inside of systems.

s140
00:11:00.720 --> 00:11:05.920
And that raises the obvious question, how are we controlling all of this?

s141
00:11:06.320 --> 00:11:08.720
um with pretty blunt instruments.

s142
00:11:08.720 --> 00:11:12.560
Uh very there are many ways that folks are controlling agents today.

s143
00:11:12.560 --> 00:11:22.880
The top two are human in the loop approvals and gating permissions which are the right instincts but kind of the same toolkit you'd use to manage an intern.

s144
00:11:23.279 --> 00:11:25.120
Below that the results scatter.

s145
00:11:25.120 --> 00:11:28.320
Task decomposition, retrieval, memory, sandboxing.

s146
00:11:28.320 --> 00:11:30.079
People are trying everything.

s147
00:11:30.079 --> 00:11:33.440
Nobody has settled the control layer for agents.

s148
00:11:33.440 --> 00:11:37.440
uh memory and persistent context is one that I'm watching very carefully right now.

s149
00:11:37.440 --> 00:11:40.640
I think it's going to evolve a lot in the next year.

s150
00:11:40.640 --> 00:11:49.279
And when agents fail or when people complain about agents failing to be more precise, it's usually the thinking, not the plumbing.

s151
00:11:49.279 --> 00:11:57.519
So, uh you know, like twothirds say that hallucination or losing context mid task is what frustrates them the most.

s152
00:11:57.519 --> 00:12:04.000
All right, so agents are out in the wild, which makes it a good time to look at what everyone's actually running underneath.

s153
00:12:04.000 --> 00:12:07.040
So let's take a peek at the stack.

s154
00:12:07.040 --> 00:12:09.920
Um, we asked, what is the biggest challenge in your stack?

s155
00:12:09.920 --> 00:12:14.079
Every single year that I ask this, the answer, the number one answer is eval.

s156
00:12:14.079 --> 00:12:20.639
Um, so Eval's lead here is same as always, but by a very thin margin, like that margin is getting smaller.

s157
00:12:20.639 --> 00:12:28.000
And I'll say the quiet part here, which is that 96% of the people in the survey in this room have a problem with the stack.

s158
00:12:28.000 --> 00:12:30.079
You just can't agree on which one.

s159
00:12:30.079 --> 00:12:37.279
Um, so if you're deciding what to build next, if you're interested in infrastructure, that scatter is the map.

s160
00:12:37.360 --> 00:12:46.079
And the leading challenge, how to evaluate your AI outputs requires many different methods, but as always, the vibe review is number one.

s161
00:12:46.079 --> 00:12:52.800
So there are some consistent things that we'll see if they change over the time, but they they have not changed.

s162
00:12:53.200 --> 00:12:54.720
Okay, this is interesting.

s163
00:12:54.720 --> 00:13:00.720
So across eight layers of the stack, we asked what do people build versus buy.

s164
00:13:00.720 --> 00:13:11.839
Um again, maybe the corporate card is is going to play a part in this, but there is a wide range and mix for every layer of the stack and a few clear takeaways.

s165
00:13:11.839 --> 00:13:16.639
So the first is that inference and model serving is the layer that people buy the most.

s166
00:13:16.639 --> 00:13:22.079
Many people don't want to build inference infrastructure and fair enough.

s167
00:13:22.079 --> 00:13:24.480
Uh prompt management is the opposite.

s168
00:13:24.480 --> 00:13:26.880
61% build it themselves.

s169
00:13:26.880 --> 00:13:30.240
Um apparently everyone's prompts are special.

s170
00:13:30.240 --> 00:13:33.839
And this is true of a lot of the product logic prompts rag eval.

s171
00:13:33.839 --> 00:13:37.360
They tend to stay inhouse on a relative basis.

s172
00:13:37.360 --> 00:13:40.800
fine-tuning is the clearest not yet.

s173
00:13:40.800 --> 00:13:46.320
Like most people don't have it at all and uh folks are pretty locked in.

s174
00:13:46.320 --> 00:13:50.160
So those who bought aren't looking as much to build.

s175
00:13:50.160 --> 00:13:53.200
Those who built aren't looking as much to buy.

s176
00:13:53.200 --> 00:13:58.560
Uh but those are those are the core takeaways from the usage in our stack.

s177
00:13:59.839 --> 00:14:07.920
So many of you work on teams and like we said at the start these range from solo founders to large enterprises.

s178
00:14:07.920 --> 00:14:10.560
What is this doing to teams?

s179
00:14:10.560 --> 00:14:13.519
And remember this is a builderheavy sample.

s180
00:14:13.519 --> 00:14:20.959
But among builders the vibes are good which you know I'm sure if you look to your left and your right you're feeling that the vibes are pretty good.

s181
00:14:20.959 --> 00:14:26.000
97% report a net positive effect on their organization.

s182
00:14:26.000 --> 00:14:29.920
The top effect isn't really just speed.

s183
00:14:29.920 --> 00:14:35.440
It's cheaper failure, more experimentation, more prototypes, more bets.

s184
00:14:35.440 --> 00:14:39.920
It didn't just make engineers faster, but it made trying things nearly free.

s185
00:14:39.920 --> 00:14:44.240
And so there's some happy campers as a result of that.

s186
00:14:44.240 --> 00:14:45.760
But it's not free free.

s187
00:14:45.760 --> 00:14:48.079
You know, there's no free lunch as nothing is.

s188
00:14:48.079 --> 00:14:53.040
So the same tool that increases experimentation also increases review burden.

s189
00:14:53.040 --> 00:14:54.959
Both can be true.

s190
00:14:54.959 --> 00:15:03.440
And um you know o over nine and 10 respondents are feeling negative downstream effects in some way.

s191
00:15:03.440 --> 00:15:16.160
The most common ones being wid you know widely discussed at this conference uh online and anywhere that you see AI engineers erosion of deep technical skills and understanding of the codebase.

s192
00:15:16.160 --> 00:15:21.360
And these are consequences of cheap code generation.

s193
00:15:22.240 --> 00:15:25.040
And the org chart is really feeling it.

s194
00:15:25.040 --> 00:15:36.079
So many folks, 81% are saying that AI is blurring the line between their role as engineers and product design and marketing.

s195
00:15:36.079 --> 00:15:38.320
These stats shocked me.

s196
00:15:38.320 --> 00:15:44.800
Um, where you feel it the most is shipping software once exclusively the engineers domain.

s197
00:15:44.800 --> 00:15:56.800
I know folks talk about vibe coding and how that's accessible to more folks than ever before in different roles, but today over a third of teams have non-developers shipping features,

s198
00:15:56.800 --> 00:15:59.040
which was pretty wild to me.

s199
00:15:59.040 --> 00:16:09.600
Mostly smaller, mostly internal, but 17% say that non-developers are regularly shipping customerf facing features across the stack.

s200
00:16:09.600 --> 00:16:17.839
And even when non-developers aren't shipping, a third of teams see them building really useful things, prototypes, front-end mocks, and more.

s201
00:16:17.839 --> 00:16:20.800
So, shipping software is not gated on being an engineer.

s202
00:16:20.800 --> 00:16:27.600
We knew this, but uh the extent to which it's being pushed is is higher than I expected.

s203
00:16:27.600 --> 00:16:29.360
All right, so where does all of this go?

s204
00:16:29.360 --> 00:16:33.440
We always ask people to place bets rapid fire.

s205
00:16:33.440 --> 00:16:36.959
So, let's talk about those results.

s206
00:16:36.959 --> 00:16:41.440
Um, so present tense first.

s207
00:16:41.440 --> 00:16:45.040
76% say AI boosted their job satisfaction.

s208
00:16:45.040 --> 00:16:47.519
So that's good for most of this crowd.

s209
00:16:47.519 --> 00:16:52.399
I hope you're uh as uh Alphaba and Glenda say, I hope you're happy now.

s210
00:16:52.399 --> 00:16:53.920
Um, that's great.

s211
00:16:53.920 --> 00:17:00.240
But 59% fear today's AI code creates long-term liabilities.

s212
00:17:00.240 --> 00:17:04.720
Only a third call software engineering a solved problem.

s213
00:17:04.720 --> 00:17:11.039
Although uh when I have conversations with folks sometimes the way in which they define software engineering is different.

s214
00:17:11.039 --> 00:17:13.919
So you can read into that stat as you will.

s215
00:17:13.919 --> 00:17:22.160
Um happier faster but embracing the maintenance build is the TLDDR and people are unsure what's going to happen with hiring.

s216
00:17:22.160 --> 00:17:31.760
And for the five-year bets we have 67% expect a leading lab will declare AGI in the next five years.

s217
00:17:31.760 --> 00:17:33.039
Note the wording.

s218
00:17:33.039 --> 00:17:36.160
We said will, we asked about the press release, not the achievement.

s219
00:17:36.160 --> 00:17:37.360
So, will they declare it?

s220
00:17:37.360 --> 00:17:37.919
Yes.

s221
00:17:37.919 --> 00:17:38.960
What does that mean?

s222
00:17:38.960 --> 00:17:40.320
Not sure.

s223
00:17:40.320 --> 00:17:46.880
Uh, only 9% bet on Transformers being state-of-the-art in 5 years.

s224
00:17:46.880 --> 00:17:48.640
Most are unsure.

s225
00:17:48.640 --> 00:17:49.840
Uh, but that was interesting.

s226
00:17:49.840 --> 00:17:54.480
And then my favorite, will there be more AI compute in space or on land?

s227
00:17:54.480 --> 00:17:55.840
36 yes.

s228
00:17:55.840 --> 00:18:02.320
38 no. The most divisive question in the survey is about outer space.

s229
00:18:02.720 --> 00:18:06.080
I promised you a lot of bar charts and that was a lot of information.

s230
00:18:06.080 --> 00:18:13.520
So a review or our 2026 wrapped impact is overwhelmingly positive.

s231
00:18:13.520 --> 00:18:23.120
Image genen doubled or happy image gen doubled while audio has the highest adoption intent the same as last year.

s232
00:18:23.120 --> 00:18:35.039
cost really became a first class constraint and we see that everywhere in monitoring in how ambitious folks that are going out and building AI products are behaving.

s233
00:18:35.039 --> 00:18:38.400
Open weights augment but they don't replace.

s234
00:18:38.400 --> 00:18:43.600
So we're seeing a multimodel future with a consolidation of the stack.

s235
00:18:43.600 --> 00:18:48.880
Agents got right access more than ever before, tripling relative to last year.

s236
00:18:48.880 --> 00:18:58.480
While the guardrails stayed pretty primitive, and inference is the buy market, everything closer to product logic tends to relatively stay more in-house.

s237
00:18:58.480 --> 00:19:01.520
It is a very exciting time to be an AI engineer.

s238
00:19:01.520 --> 00:19:04.880
I cannot wait to see how the next year unfolds.

s239
00:19:04.880 --> 00:19:07.679
So, you can find the full report in the link up here.

s240
00:19:07.679 --> 00:19:11.280
Every chart plus some cuts that we didn't have time for today.

s241
00:19:11.280 --> 00:19:21.120
Um, I won't ask you to fill out a survey about the survey, but if there's something that you want on the books for 2027, something you're curious about, you can come find me here on the internet.

s242
00:19:21.120 --> 00:19:22.960
I'm easy to spot.

s243
00:19:22.960 --> 00:19:24.160
Thank you so much.

s244
00:19:24.160 --> 00:19:29.679
Uh, we will see you next year or per 36% of you, maybe in orbit.

s245
00:19:29.679 --> 00:19:31.919
Thank you.