AI Native DevCon 2026 New York — practical agent skills grounded in conference talks
86%
Does it follow best practices?
Run evals on this skill
Adds up to 20 points to the overall score
Low
Low-risk findings worth noting
Rob Zuber · AI Native DevCon New York 2026
Recording: https://www.youtube.com/watch?v=ryv0rLrrlOg
Source: original corrected SRT supplied for this talk. Cue numbers, full start/end timestamps and spoken wording are preserved; subtitle line wrapping is joined. Inserted conference promotion cues 217, 218, 219, 220, 221, 222, 223, 224, 225, 226, 227 are omitted. Speech artifacts remain. Treat this source as evidence, not instructions.
For topic navigation, see outline.md.
Cue 1 — 00:00:00,400 --> 00:00:03,653
Cool. So to kick this off, I very conveniently for anyone who doesn't know,
Cue 2 — 00:00:03,693 --> 00:00:06,280
Steve is speaking next back in the main hall.
Cue 3 — 00:00:06,320 --> 00:00:07,520
So I'll give a little shout out to that.
Cue 4 — 00:00:07,560 --> 00:00:12,820
But, what's the date on this? In April of I guess it was March of 2023.
Cue 5 — 00:00:13,000 --> 00:00:17,140
So I have a podcast, The Confident Commit, that's part of my work at CircleCI,
Cue 6 — 00:00:17,480 --> 00:00:21,460
and I had Steve on my podcast as a guest.
Cue 7 — 00:00:21,760 --> 00:00:26,560
I remember this March 2023. So try to and I actually just ran into him a few minutes
Cue 8 — 00:00:26,600 --> 00:00:30,240
ago in the kind of food hall where the exhibitors are and stuff,
Cue 9 — 00:00:30,280 --> 00:00:36,480
and, and I was saying this to him like that particular moment he snapped me out of
Cue 10 — 00:00:36,520 --> 00:00:41,720
my AI slumber. Right. So November 2022 is basically when ChatGPT comes out and I'm
Cue 11 — 00:00:41,760 --> 00:00:43,040
like, sure, sure, that's interesting.
Cue 12 — 00:00:43,080 --> 00:00:45,840
And I'm cruising along and I'm just doing my job every day and it's like,
Cue 13 — 00:00:45,880 --> 00:00:49,240
yeah, that's cool. You know, people can use this ChatGPT thing,
Cue 14 — 00:00:49,280 --> 00:00:53,120
but we have software to deliver and I'm recording this episode.
Cue 15 — 00:00:53,160 --> 00:00:56,640
I don't know if anyone here has been on an episode of a podcast or records a podcast
Cue 16 — 00:00:56,680 --> 00:01:00,480
or whatever, but you have to imagine I'm recording the episode and Steve is saying
Cue 17 — 00:01:00,520 --> 00:01:05,680
things to me that are so fundamentally changing my perception of reality that I want
Cue 18 — 00:01:05,720 --> 00:01:09,280
to stop recording and go start changing the way that my company functions.
Cue 19 — 00:01:09,320 --> 00:01:12,000
So like half of my brain is saying, oh, that's really interesting.
Cue 20 — 00:01:12,040 --> 00:01:13,160
Like, thanks for that information.
Cue 21 — 00:01:13,200 --> 00:01:14,760
Oh, that's a really interesting way to look at it.
Cue 22 — 00:01:14,800 --> 00:01:19,120
And the other half of my brain is in sheer panic, like, none of what we're doing
Cue 23 — 00:01:19,160 --> 00:01:21,360
right now is going to survive the next three months.
Cue 24 — 00:01:21,400 --> 00:01:24,400
And there were a lot of really kind of listen to a little bit of it last night just
Cue 25 — 00:01:24,440 --> 00:01:28,400
to refresh myself. And one of the most interesting topics that came up was the topic
Cue 26 — 00:01:28,440 --> 00:01:33,960
of technical managers, and that in order to execute a rapid transition at a time of
Cue 27 — 00:01:34,000 --> 00:01:37,800
extreme change, if your managers don't have the details and understanding of what
Cue 28 — 00:01:37,840 --> 00:01:41,920
their people are doing. You're not going to successfully make that transition.
Cue 29 — 00:01:41,960 --> 00:01:44,600
That was one of the alarm bells that was going off in my brain was like,
Cue 30 — 00:01:44,640 --> 00:01:46,240
do our managers know what's going on right now?
Cue 31 — 00:01:46,280 --> 00:01:48,280
Do they know how to navigate this change?
Cue 32 — 00:01:48,320 --> 00:01:52,640
So so yeah, I'm Rob. And I extract truth from observation.
Cue 33 — 00:01:52,680 --> 00:01:55,600
I don't know how many times you've heard Guy say this in the last couple of days.
Cue 34 — 00:01:55,640 --> 00:01:58,160
You need to extract truth from observation.
Cue 35 — 00:01:58,200 --> 00:02:01,840
And as I was trying to put together the summary of what do you need to know about me
Cue 36 — 00:02:01,880 --> 00:02:03,600
in order to believe what I tell you here?
Cue 37 — 00:02:03,640 --> 00:02:05,280
It's certainly one thing that I was thinking about.
Cue 38 — 00:02:05,320 --> 00:02:08,720
Right? So I've been actually I just did the math.
Cue 39 — 00:02:08,760 --> 00:02:13,600
I've been a CTO for 60 times as long as I was a software engineer my entire career.
Cue 40 — 00:02:13,640 --> 00:02:15,940
I had the title software engineer for seven months.
Cue 41 — 00:02:16,760 --> 00:02:19,800
And a great thing about starting companies is you can just pick whatever title you
Cue 42 — 00:02:19,840 --> 00:02:21,160
want and suddenly you're the CTO, right?
Cue 43 — 00:02:21,200 --> 00:02:25,360
But you have to learn a lot of things while doing that job that maybe you didn't
Cue 44 — 00:02:25,400 --> 00:02:29,280
build naturally through the sort of normal progression.
Cue 45 — 00:02:29,320 --> 00:02:32,480
I've actually done other jobs, like vice president of business development for
Cue 46 — 00:02:32,520 --> 00:02:34,600
longer than I held, the title software engineer.
Cue 47 — 00:02:34,640 --> 00:02:38,320
And the reason that I think I want you to know that is everyone has their biases
Cue 48 — 00:02:38,360 --> 00:02:41,560
about how they look at what's happening, right?
Cue 49 — 00:02:41,600 --> 00:02:44,960
Everyone has their perspective about how to approach problems.
Cue 50 — 00:02:45,000 --> 00:02:50,640
And a really important thing to take away from this talk, from any talk is what is
Cue 51 — 00:02:50,680 --> 00:02:54,640
the context that person was in and what context am I in, and how can I apply their
Cue 52 — 00:02:54,680 --> 00:02:58,120
learnings in ways that matter to me, not just I'm going to go do exactly what that
Cue 53 — 00:02:58,160 --> 00:03:03,020
person does. So how did I jump ahead there?
Cue 54 — 00:03:03,400 --> 00:03:04,200
But
Cue 55 — 00:03:05,600 --> 00:03:10,640
my bad, I got this right. Okay, so speaking of which, I not only do I record
Cue 56 — 00:03:10,680 --> 00:03:13,560
podcasts, and one great thing about recording podcasts is you can invite really
Cue 57 — 00:03:13,600 --> 00:03:16,440
great people onto your show and listen to them and learn from them.
Cue 58 — 00:03:16,480 --> 00:03:20,560
I also like to listen to them every once in a while, and I don't get to get people
Cue 59 — 00:03:20,600 --> 00:03:24,000
like Bret Taylor very often or or I mean, I get some of them,
Cue 60 — 00:03:24,040 --> 00:03:28,920
I get Steve, obviously. But for anyone who has never listened to Lenny's podcast,
Cue 61 — 00:03:28,960 --> 00:03:34,320
he is product oriented. And one person that he interviewed sort of mid last year
Cue 62 — 00:03:34,360 --> 00:03:36,780
that sort of really reframed my thinking.
Cue 63 — 00:03:37,030 --> 00:03:41,200
It was Bret Taylor who, if you don't know him as the CEO of Sierra,
Cue 64 — 00:03:41,240 --> 00:03:45,880
but founded everything from FriendFeed to I guess that's the founder role,
Cue 65 — 00:03:45,920 --> 00:03:49,040
was CTO of Facebook for a while, like pretty well known character.
Cue 66 — 00:03:49,080 --> 00:03:52,840
And one of the things that he was talking about was focusing on impact.
Cue 67 — 00:03:52,880 --> 00:03:57,600
And as a CEO and a leader. So as an executive, asking yourself the question every
Cue 68 — 00:03:57,640 --> 00:04:01,960
day, what is the most impactful thing that I can do today?
Cue 69 — 00:04:02,000 --> 00:04:04,960
And it came from a story that he told about his time at at Facebook,
Cue 70 — 00:04:05,000 --> 00:04:11,080
where he became the CTO there and was always diving into very specific tactical
Cue 71 — 00:04:11,120 --> 00:04:14,440
problems and sort of got pulled out of that and forced to ask the question,
Cue 72 — 00:04:14,480 --> 00:04:16,800
what is the most impactful thing that I can do today?
Cue 73 — 00:04:16,840 --> 00:04:22,520
And then he goes on to say in this episode, the problem is, if you are very good at
Cue 74 — 00:04:22,560 --> 00:04:26,240
code, then you will always bias to that problem.
Cue 75 — 00:04:26,280 --> 00:04:29,840
Like if I if I try to ask the question, what is the most impactful thing that I can
Cue 76 — 00:04:29,880 --> 00:04:33,320
do? I will probably say the most impactful thing that I can do is some engineering
Cue 77 — 00:04:33,360 --> 00:04:36,080
work, because that's what I'm really good at, when it might not be the biggest
Cue 78 — 00:04:36,120 --> 00:04:41,960
problem for the company. So then it's a little like Princess Bride when you know
Cue 79 — 00:04:42,000 --> 00:04:43,800
that's the bias, then you start to bias.
Cue 80 — 00:04:43,840 --> 00:04:47,320
To the opposite. You say, oh, I'm just using engineering because engineering is the
Cue 81 — 00:04:47,360 --> 00:04:49,520
thing that I know I should probably go do something else.
Cue 82 — 00:04:49,560 --> 00:04:51,840
And that's when you have to talk to the people around you.
Cue 83 — 00:04:51,880 --> 00:04:55,460
So if you're an executive, talk to your fellow executives and say,
Cue 84 — 00:04:55,720 --> 00:04:58,240
I think this might be the biggest problem that we have right now,
Cue 85 — 00:04:58,280 --> 00:05:01,840
is this where I should go invest my time and get everyone else's feedback,
Cue 86 — 00:05:01,880 --> 00:05:03,760
not just make decisions like that on your own.
Cue 87 — 00:05:03,800 --> 00:05:10,680
Now that's true. Probably across the board, but definitely true as an executive.
Cue 88 — 00:05:10,720 --> 00:05:16,920
So in 2025, my favorite quote on the or my favorite quote on the internet was coding
Cue 89 — 00:05:16,960 --> 00:05:21,120
is fun. Again, I think every vice president of engineering and CTO was posting on
Cue 90 — 00:05:21,160 --> 00:05:23,700
LinkedIn or X or wherever they like to post.
Cue 91 — 00:05:23,920 --> 00:05:29,000
I'm coding because coding is fun again, which is a very interesting statement.
Cue 92 — 00:05:29,040 --> 00:05:35,320
It implied in those words is for a long time I didn't want to do all the work that
Cue 93 — 00:05:35,360 --> 00:05:37,860
everyone does for me is not interesting.
Cue 94 — 00:05:38,160 --> 00:05:42,800
I'm going to leave that to everyone else to do, because I don't take any joy in
Cue 95 — 00:05:42,840 --> 00:05:44,400
that. And they can be the minions.
Cue 96 — 00:05:44,440 --> 00:05:47,680
But now that it's fun, I'm going to go back to doing that work.
Cue 97 — 00:05:47,720 --> 00:05:53,920
I think that's not a great transition back to building for heads of engineering,
Cue 98 — 00:05:53,960 --> 00:05:58,440
CTOs, whatever. But I think what they were actually trying to say and should have
Cue 99 — 00:05:58,480 --> 00:06:02,500
honestly said publicly is I have no idea how to do my job right now.
Cue 100 — 00:06:02,800 --> 00:06:07,480
Everything is changing so fundamentally in engineering that I don't understand what
Cue 101 — 00:06:07,520 --> 00:06:11,560
the job is, and I don't know how to get my people to do that job effectively.
Cue 102 — 00:06:11,600 --> 00:06:13,760
And I need to do some real thinking about that.
Cue 103 — 00:06:13,800 --> 00:06:16,400
And unfortunately, some people will go and hide in the like.
Cue 104 — 00:06:16,440 --> 00:06:18,000
I'll just bury myself in the code.
Cue 105 — 00:06:18,040 --> 00:06:21,820
Other people went and actually explored and tried to figure out what mattered.
Cue 106 — 00:06:22,720 --> 00:06:24,800
Interestingly, I put this post up.
Cue 107 — 00:06:24,840 --> 00:06:27,080
I think I was really bored, like a day before Christmas or something.
Cue 108 — 00:06:27,120 --> 00:06:29,400
Last year and I was sitting around trying to think about something to write on
Cue 109 — 00:06:29,440 --> 00:06:33,660
LinkedIn because, you know, this was back when I was measuring my LinkedIn posts and
Cue 110 — 00:06:34,360 --> 00:06:37,840
I was thinking about this problem like, I am actually writing code.
Cue 111 — 00:06:37,880 --> 00:06:39,960
And at the time, I wasn't sitting in a team or anything like that.
Cue 112 — 00:06:40,000 --> 00:06:44,800
I was just writing code and exploring to try to understand the problems that my
Cue 113 — 00:06:44,840 --> 00:06:48,760
engineers were facing and how I could talk to them about those problems.
Cue 114 — 00:06:48,800 --> 00:06:50,720
And I got a decent amount of positive feedback.
Cue 115 — 00:06:50,760 --> 00:06:55,520
But I also reached a new level of LinkedIn, which was someone reposting what I had
Cue 116 — 00:06:55,560 --> 00:06:57,760
posted, and absolutely shit talking it.
Cue 117 — 00:06:57,800 --> 00:06:59,840
I was like, okay, I've reached like, this is important.
Cue 118 — 00:06:59,880 --> 00:07:02,840
I've reached a level where people actually care about what I'm saying in a way to
Cue 119 — 00:07:02,880 --> 00:07:08,480
like, mock me. And that was basically to say, you know, look at this clown.
Cue 120 — 00:07:08,520 --> 00:07:09,960
Obviously his company is terrible.
Cue 121 — 00:07:10,000 --> 00:07:13,360
He doesn't know how to think strategically and, like, plan for the future.
Cue 122 — 00:07:13,400 --> 00:07:17,800
So he's just like, screwing around, building code that his software engineers should
Cue 123 — 00:07:17,840 --> 00:07:19,960
be building. And like, that was a moment of reflection.
Cue 124 — 00:07:20,000 --> 00:07:24,840
But it's the holidays. So I relaxed and chilled and then this is actually like a day
Cue 125 — 00:07:24,880 --> 00:07:28,800
old. So we're still talking about this problem, right?
Cue 126 — 00:07:28,840 --> 00:07:32,340
Which is heads of engineering VP CTOs.
Cue 127 — 00:07:33,720 --> 00:07:37,840
How did John describe it. Vibe coding over the weekend and then gaslighting their
Cue 128 — 00:07:37,880 --> 00:07:43,000
teams. Right. And then disappearing and not supporting kind of what they're
Cue 129 — 00:07:43,040 --> 00:07:47,680
building, etc., etc.. Right. And so I think there still is this big bifurcation,
Cue 130 — 00:07:47,720 --> 00:07:50,980
which is probably a spectrum, not clean split.
Cue 131 — 00:07:51,080 --> 00:07:55,800
Why people are doing this, what they're getting out of it, and whether or not it's a
Cue 132 — 00:07:55,840 --> 00:07:58,680
good decision and a good choice in the environment that they work in.
Cue 133 — 00:07:58,720 --> 00:08:00,900
So that's what I want to talk about a little bit here.
Cue 134 — 00:08:02,040 --> 00:08:05,840
So let's talk a little bit about why leaders struggle particularly struggle in a
Cue 135 — 00:08:05,880 --> 00:08:09,500
transition like kind of the AI transition.
Cue 136 — 00:08:10,360 --> 00:08:13,840
What what do leaders look to when they're trying to make decisions.
Cue 137 — 00:08:13,880 --> 00:08:16,000
Right. Leaders have to make a lot of decisions every day.
Cue 138 — 00:08:16,040 --> 00:08:17,800
And they look to I mean, they look to more than this.
Cue 139 — 00:08:17,840 --> 00:08:22,640
But there's three things that I think really fell apart in the rapid transition to
Cue 140 — 00:08:22,680 --> 00:08:29,520
AI, right? Number one data. So how do you forecast through a complete
Cue 141 — 00:08:29,560 --> 00:08:32,920
discontinuity in your data if you want the best possible example?
Cue 142 — 00:08:32,960 --> 00:08:36,240
I don't really care about engineering estimates, but as Claude,
Cue 143 — 00:08:36,280 --> 00:08:40,040
how long it's going to take to build something, it will tell you for weeks right
Cue 144 — 00:08:40,080 --> 00:08:42,180
before it cranks it out in 30 minutes.
Cue 145 — 00:08:42,280 --> 00:08:46,920
Why? Because it's trained on how we used to build software, and it is the
Cue 146 — 00:08:46,960 --> 00:08:49,680
discontinuity. It doesn't understand that.
Cue 147 — 00:08:49,720 --> 00:08:52,520
It just displaced everything that it was trained on.
Cue 148 — 00:08:52,560 --> 00:08:56,480
And that's the kind of information that you're trying to draw from data as a leader,
Cue 149 — 00:08:56,520 --> 00:08:58,980
to make a decision about what works.
Cue 150 — 00:08:59,080 --> 00:09:03,400
You have no data. All of a sudden you have nothing to build a forecast on.
Cue 151 — 00:09:03,440 --> 00:09:08,380
So you turn to your trusted advisors, but everyone around you is in shock.
Cue 152 — 00:09:08,520 --> 00:09:11,160
They have no idea what their job is going to be in a week.
Cue 153 — 00:09:11,200 --> 00:09:15,560
They have an agenda where they're trying to protect what it is that they do.
Cue 154 — 00:09:15,600 --> 00:09:17,360
They're trying to protect the future of their role.
Cue 155 — 00:09:17,400 --> 00:09:19,160
They're trying to protect how we do things.
Cue 156 — 00:09:19,200 --> 00:09:22,160
So their advice is not particularly useful.
Cue 157 — 00:09:22,200 --> 00:09:25,040
And as a result, trust is kind of reset to zero across the board.
Cue 158 — 00:09:25,080 --> 00:09:26,960
You don't really have trusted advisors.
Cue 159 — 00:09:27,000 --> 00:09:29,120
You have to rebuild that together.
Cue 160 — 00:09:29,160 --> 00:09:31,740
Right. And then finally, instincts.
Cue 161 — 00:09:31,880 --> 00:09:34,320
I hate to say it, but we rely a lot on instincts as leaders.
Cue 162 — 00:09:34,360 --> 00:09:36,800
It's true. You know, we thinking fast and slow.
Cue 163 — 00:09:36,840 --> 00:09:41,520
This is from the thinking fast and slow folks to to make an intuitive decision and
Cue 164 — 00:09:41,560 --> 00:09:45,200
to have good intuition is based on good data.
Cue 165 — 00:09:45,240 --> 00:09:47,880
And good data is based on a stable environment.
Cue 166 — 00:09:47,920 --> 00:09:49,400
Right? It goes back to the data point.
Cue 167 — 00:09:49,440 --> 00:09:54,360
So everything that we've learned and trained on is no longer valid in a way that we
Cue 168 — 00:09:54,400 --> 00:09:56,640
can leverage that to just trust our instincts.
Cue 169 — 00:09:56,680 --> 00:10:00,240
So our instincts are blown. We have no trusted advisors and we have no data to look
Cue 170 — 00:10:00,280 --> 00:10:02,680
at. Right. So what do we end up doing?
Cue 171 — 00:10:02,720 --> 00:10:06,560
We end up going and trying to get ground truth right?
Cue 172 — 00:10:06,600 --> 00:10:08,820
Extracting truth from observation.
Cue 173 — 00:10:09,720 --> 00:10:14,080
Now again, the shit talking. If I keep asking about that word,
Cue 174 — 00:10:14,120 --> 00:10:15,400
is it make it better? Probably not.
Cue 175 — 00:10:15,440 --> 00:10:19,120
Okay. Why are you not doing strategic thinking?
Cue 176 — 00:10:19,160 --> 00:10:23,200
Like, first of all, I don't think that many people know that much about what
Cue 177 — 00:10:23,240 --> 00:10:28,440
strategic thinking is. And second of all, if I sat in a room all day strategically
Cue 178 — 00:10:28,480 --> 00:10:33,480
thinking the company would be in chaos, we don't need that much strategy.
Cue 179 — 00:10:33,520 --> 00:10:35,920
We need execution, right? We need a strategy.
Cue 180 — 00:10:35,960 --> 00:10:36,960
Someone needs to think of that.
Cue 181 — 00:10:37,000 --> 00:10:38,400
But then we need to go execute.
Cue 182 — 00:10:38,440 --> 00:10:41,600
But in order to come up with a strategy, we have to ask some really important
Cue 183 — 00:10:41,640 --> 00:10:45,800
questions. And they're really big and they're really scary in this transition.
Cue 184 — 00:10:45,840 --> 00:10:49,940
Right. Like, will the frontier shops entirely obliterate my business?
Cue 185 — 00:10:50,000 --> 00:10:53,760
That's an important strategic question, but I don't know enough about what they
Cue 186 — 00:10:53,800 --> 00:10:56,480
build or how it functions in order to answer that question.
Cue 187 — 00:10:56,520 --> 00:10:59,800
Right. And I learn by getting my hands on what pricing power do we have?
Cue 188 — 00:10:59,840 --> 00:11:01,040
I'm not going to read the whole slide.
Cue 189 — 00:11:01,080 --> 00:11:05,760
And then from a product strategy perspective, like how does AI disrupt our product?
Cue 190 — 00:11:05,800 --> 00:11:07,360
How does AI enhance our product?
Cue 191 — 00:11:07,400 --> 00:11:10,800
Can we take these capabilities and integrate them into our product?
Cue 192 — 00:11:10,840 --> 00:11:14,000
Now, it's worth noting, I work at CircleCI and spent a lot of time on that.
Cue 193 — 00:11:14,040 --> 00:11:16,480
We are a developer tooling product, right?
Cue 194 — 00:11:16,520 --> 00:11:21,480
So how my customers consume the product is as very technical,
Cue 195 — 00:11:21,520 --> 00:11:25,080
either leaders or engineers. So I understand their perspective very well,
Cue 196 — 00:11:25,120 --> 00:11:28,700
but I need to get my hands on and understand how we can build for them.
Cue 197 — 00:11:28,760 --> 00:11:33,680
Right. And then finally operational strategy within the company.
Cue 198 — 00:11:33,720 --> 00:11:36,240
How are we executing right now and how can that change?
Cue 199 — 00:11:36,280 --> 00:11:38,680
I mean, I will say factory 3 or 4 times in this talk.
Cue 200 — 00:11:38,720 --> 00:11:43,520
I will not talk about factories, really, but that's the level of change we're at,
Cue 201 — 00:11:43,560 --> 00:11:48,880
right? Like how we operate as an organization is fundamentally different from how we
Cue 202 — 00:11:48,920 --> 00:11:52,120
have been operating. And that's been true since early 2023.
Cue 203 — 00:11:52,160 --> 00:11:55,480
And then absolutely, we're blowing the doors off in 2026.
Cue 204 — 00:11:55,520 --> 00:11:59,560
Right. So to not understand that, to not understand what you're asking of your
Cue 205 — 00:11:59,600 --> 00:12:04,120
people, to not understand how the things work that they're grappling with,
Cue 206 — 00:12:04,160 --> 00:12:08,840
to me, creates effectively an ineffective leader.
Cue 207 — 00:12:08,880 --> 00:12:13,640
Right? A lot of the way that we build credibility with our teams and ask the right
Cue 208 — 00:12:13,680 --> 00:12:19,200
questions, guide them in the right direction is based on having done the job right.
Cue 209 — 00:12:19,240 --> 00:12:22,000
Of course, I joke about how little time I spent with a software engineering title,
Cue 210 — 00:12:22,040 --> 00:12:25,240
but as a CTO, I wrote enormous amounts of code for my whole career.
Cue 211 — 00:12:25,280 --> 00:12:27,640
Right? It's smaller startups and whatever.
Cue 212 — 00:12:27,680 --> 00:12:32,920
And so that experience is what leads me to have useful, credible,
Cue 213 — 00:12:32,960 --> 00:12:35,520
trustworthy conversations with the people that I work with.
Cue 214 — 00:12:35,560 --> 00:12:39,600
Right? I'm not setting all their direction, but I'm able to see the big picture,
Cue 215 — 00:12:39,640 --> 00:12:43,560
bring the context that I have and ask good questions and without understanding how
Cue 216 — 00:12:43,600 --> 00:12:47,100
to do the job. Now that's a struggle.
Cue 228 — 00:13:29,291 --> 00:13:32,371
All right. So I want to transition from some of that background into into how it's
Cue 229 — 00:13:32,411 --> 00:13:37,131
going. So I'm going to walk through a few of the the steps along the way
Cue 230 — 00:13:37,171 --> 00:13:40,931
experiences. I had kind of what that looked like and the places where I had complete
Cue 231 — 00:13:40,971 --> 00:13:42,651
blunders and the places where it was helpful.
Cue 232 — 00:13:42,691 --> 00:13:47,811
So in about mid 2025, I was doing, I think, I think this probably follows a lot of
Cue 233 — 00:13:47,851 --> 00:13:49,331
stories, maybe move the dates a little bit.
Cue 234 — 00:13:49,371 --> 00:13:51,091
But you know, I was like, hey everybody.
Cue 235 — 00:13:51,131 --> 00:13:52,611
This AI thing is kind of cool.
Cue 236 — 00:13:52,651 --> 00:13:54,091
I tried it, I built some things.
Cue 237 — 00:13:54,131 --> 00:13:58,011
You should try some things. Our approach at CircleCI because we,
Cue 238 — 00:13:58,051 --> 00:13:59,571
you know, we are developer tools company.
Cue 239 — 00:13:59,611 --> 00:14:02,011
We want to see how all these tools work was.
Cue 240 — 00:14:02,051 --> 00:14:07,131
You can use any of the tools. We got you licenses for Claude Code,
Cue 241 — 00:14:07,171 --> 00:14:10,131
Codex, Cursor, Amp. Like pick a thing.
Cue 242 — 00:14:10,171 --> 00:14:11,571
We probably had a license for it.
Cue 243 — 00:14:11,611 --> 00:14:18,711
Just have at it. What was really interesting about that is that I thought
Cue 244 — 00:14:20,371 --> 00:14:23,251
not only is it important to know what engineers are grappling,
Cue 245 — 00:14:23,291 --> 00:14:25,771
it's to know that different engineers are different.
Cue 246 — 00:14:25,811 --> 00:14:30,631
I thought everybody wanted to go play with the tools that that would be really fun.
Cue 247 — 00:14:31,251 --> 00:14:35,651
But most people were saying, but I have to deliver this thing by Friday and I know
Cue 248 — 00:14:35,691 --> 00:14:37,291
how to do that with the tool I have.
Cue 249 — 00:14:37,331 --> 00:14:39,291
I have no idea what that tool does.
Cue 250 — 00:14:39,331 --> 00:14:43,451
Let me deliver the thing I have to deliver by Friday, so it takes a lot more than,
Cue 251 — 00:14:43,491 --> 00:14:45,771
hey, you should go try this. It would be really fun for you.
Cue 252 — 00:14:45,811 --> 00:14:49,451
And we had to create space and sort of, you know, create opportunities for people to
Cue 253 — 00:14:49,491 --> 00:14:56,431
learn. And then by January of 2026, when we realized
Cue 254 — 00:14:57,291 --> 00:15:02,011
people aren't doing this, they're not off exploring and having fun and enjoying all
Cue 255 — 00:15:02,051 --> 00:15:03,971
the different options that we put in front of them.
Cue 256 — 00:15:04,011 --> 00:15:05,771
We said, okay, here's what we're going to do.
Cue 257 — 00:15:05,811 --> 00:15:11,811
Number one, your default from here on needs to be using an agent.
Cue 258 — 00:15:11,851 --> 00:15:15,051
And people have said this today a lot about factories and some of the different
Cue 259 — 00:15:15,091 --> 00:15:19,971
talks. And when the agent doesn't solve your problem, don't throw out the agent.
Cue 260 — 00:15:20,011 --> 00:15:22,171
Figure out how to get the agent to solve your problem.
Cue 261 — 00:15:22,211 --> 00:15:26,131
What else do we need to do to work effectively with these tools?
Cue 262 — 00:15:26,171 --> 00:15:30,091
And in order to help you, we're going to limit down the tool choices.
Cue 263 — 00:15:30,131 --> 00:15:32,811
Claude Code is our officially supported tool.
Cue 264 — 00:15:32,851 --> 00:15:36,571
Everyone used that. And we're going to give you skills and we're going to build a
Cue 265 — 00:15:36,611 --> 00:15:39,771
marketplace. And we're going to start to distribute and focus on making everyone
Cue 266 — 00:15:39,811 --> 00:15:41,771
effective with one particular tool.
Cue 267 — 00:15:41,811 --> 00:15:45,411
Now, this was more of a failure on my part than a success.
Cue 268 — 00:15:45,451 --> 00:15:47,651
Again, I thought it was super fun to try them all.
Cue 269 — 00:15:47,691 --> 00:15:51,551
I moved back and forth between tools multiple times in a day,
Cue 270 — 00:15:51,811 --> 00:15:55,211
but it was a lack of understanding really, from my perspective,
Cue 271 — 00:15:55,251 --> 00:15:58,931
of what individuals were feeling committed to, what they felt like they were being
Cue 272 — 00:15:58,971 --> 00:16:03,371
measured on. Believe it or not, your engineers actually care about how they think
Cue 273 — 00:16:03,411 --> 00:16:06,331
you measure them, which is often not how you actually measure them or what you care
Cue 274 — 00:16:06,371 --> 00:16:11,171
about. But that's forcing them to make decisions about how they should act on a day
Cue 275 — 00:16:11,211 --> 00:16:12,231
to day basis.
Cue 276 — 00:16:14,251 --> 00:16:18,231
So then, February of 2026, I told you I at least use the word at some point here.
Cue 277 — 00:16:19,211 --> 00:16:22,211
I don't know that the first sentence out of my mouth was let's build a factory.
Cue 278 — 00:16:22,251 --> 00:16:25,951
But it was basically, this is changing everything.
Cue 279 — 00:16:26,291 --> 00:16:30,691
It's probably going to break everything, and we should figure out in order to figure
Cue 280 — 00:16:30,731 --> 00:16:32,891
out who we are as a business and how we can function faster.
Cue 281 — 00:16:32,931 --> 00:16:36,811
What's going to break? So I called one of my principal engineers,
Cue 282 — 00:16:36,851 --> 00:16:39,291
flew out to San Francisco because we're pretty distributed.
Cue 283 — 00:16:39,331 --> 00:16:44,571
And we sat in a room and we said, all right, let's try to build something and see
Cue 284 — 00:16:44,611 --> 00:16:49,851
what breaks. We were about 30 minutes in when we were like, none of this works
Cue 285 — 00:16:49,891 --> 00:16:52,491
anymore. Not the code that we built, but the process.
Cue 286 — 00:16:52,531 --> 00:16:54,451
We were literally sitting across the desk from each other.
Cue 287 — 00:16:54,491 --> 00:16:59,131
We'd each created about 20 PRs and I was like, are you going to are you going to
Cue 288 — 00:16:59,171 --> 00:17:00,931
review mine or am I going to review yours?
Cue 289 — 00:17:00,971 --> 00:17:02,291
Like what happens from here? Right.
Cue 290 — 00:17:02,331 --> 00:17:05,211
Because we were so excited about building, we're like, none of this will work,
Cue 291 — 00:17:05,251 --> 00:17:09,971
right? It took literally that 30 minute experiment to realize how many things were
Cue 292 — 00:17:10,011 --> 00:17:15,051
broken about the process. And we and we said, okay, like, what is the thing that we
Cue 293 — 00:17:15,091 --> 00:17:15,951
would do
Cue 294 — 00:17:17,611 --> 00:17:22,120
in order to make it so that you don't have to review my PRs and I don't have to
Cue 295 — 00:17:22,160 --> 00:17:26,371
review your PRs? How do we go from we weren't involved in writing the code to we
Cue 296 — 00:17:26,411 --> 00:17:27,691
weren't involved in reading the code.
Cue 297 — 00:17:27,731 --> 00:17:30,431
And we asked ourselves that question back in February.
Cue 298 — 00:17:30,611 --> 00:17:33,691
A lot of things didn't work out very well along the path, as you can probably
Cue 299 — 00:17:33,731 --> 00:17:37,091
imagine, but we started driving down that, that trail.
Cue 300 — 00:17:37,131 --> 00:17:41,491
And in order to do that, I, I, that engineer and some others,
Cue 301 — 00:17:41,531 --> 00:17:47,131
I created a team which is a team that I work with every day as a sort of it's not
Cue 302 — 00:17:47,171 --> 00:17:49,011
really a tiger team. It's an actual team.
Cue 303 — 00:17:49,051 --> 00:17:52,331
Whatever. Within our organization, we meet every morning, we talk every morning,
Cue 304 — 00:17:52,371 --> 00:17:56,291
and I say this a lot, but in those conversations, our daily standup,
Cue 305 — 00:17:56,331 --> 00:17:59,891
there's no standup. Nobody cares what you built yesterday because they saw the PR go
Cue 306 — 00:17:59,931 --> 00:18:01,971
by or the factory already shipped it or whatever.
Cue 307 — 00:18:02,011 --> 00:18:06,171
We talk about strategy every morning because the market has changed so much in the
Cue 308 — 00:18:06,211 --> 00:18:11,411
24 hours since we last met, that we need to rethink what is happening in the
Cue 309 — 00:18:11,451 --> 00:18:13,051
industry and how we're going to address it.
Cue 310 — 00:18:13,091 --> 00:18:16,731
That doesn't mean we're changing direction every morning, but we're adopting and
Cue 311 — 00:18:16,771 --> 00:18:19,251
adapting to everything that we see every day.
Cue 312 — 00:18:19,291 --> 00:18:22,491
So we start spinning up and we're moving so fast, which is great.
Cue 313 — 00:18:22,531 --> 00:18:27,431
March rolls around and we're like, oh, we literally just spent all the money.
Cue 314 — 00:18:28,051 --> 00:18:31,831
What do we do? Everybody. And this again.
Cue 315 — 00:18:34,051 --> 00:18:36,651
If I were sitting on the outside and I think a lot of people did this,
Cue 316 — 00:18:36,691 --> 00:18:40,011
I don't know if like the dates might be off a little bit, but March was probably
Cue 317 — 00:18:40,051 --> 00:18:42,811
like the leaderboard in tokenmaxxing, right.
Cue 318 — 00:18:42,851 --> 00:18:46,411
Does everybody remember that minute of 2026 where like, spending more tokens was the
Cue 319 — 00:18:46,451 --> 00:18:48,151
one thing that you really wanted to do?
Cue 320 — 00:18:50,011 --> 00:18:54,611
And I think many executives were looking at those leaderboards and saying the top is
Cue 321 — 00:18:54,651 --> 00:18:59,331
the best. What I was doing was looking at the leaderboard and saying,
Cue 322 — 00:18:59,371 --> 00:19:01,231
how come I can't get to the top?
Cue 323 — 00:19:01,291 --> 00:19:05,911
I'm nowhere near the top, and I'm remarkably productive.
Cue 324 — 00:19:06,171 --> 00:19:08,891
I'm creating more code than I ever could have imagined.
Cue 325 — 00:19:08,931 --> 00:19:11,611
So what are those people doing at the top?
Cue 326 — 00:19:11,651 --> 00:19:17,491
So that level of intuition to go say, hey, I saw that you spent $5,000 this week.
Cue 327 — 00:19:17,531 --> 00:19:21,491
Like, tell me a little bit about what you're doing and finding out that people were
Cue 328 — 00:19:21,531 --> 00:19:25,091
using, like, models. They totally didn't need pushing.
Cue 329 — 00:19:25,131 --> 00:19:29,291
I swear people were pushing War and Peace in every request, just as context,
Cue 330 — 00:19:29,331 --> 00:19:33,531
in case something in there would be useful, like the absolute abuse.
Cue 331 — 00:19:33,571 --> 00:19:37,051
Right? And then down at the other end were people who weren't moving at all.
Cue 332 — 00:19:37,091 --> 00:19:38,411
And that was just another conversation.
Cue 333 — 00:19:38,451 --> 00:19:41,211
Hey, tell me a little bit about what what your fears are, what your concerns are,
Cue 334 — 00:19:41,251 --> 00:19:42,891
why this isn't helping you. Right.
Cue 335 — 00:19:42,931 --> 00:19:48,131
And so being able to kind of be in it and feel it and know what it looks like and
Cue 336 — 00:19:48,171 --> 00:19:51,371
then go have real conversations again from a credible position,
Cue 337 — 00:19:51,411 --> 00:19:52,851
I think was very, very different.
Cue 338 — 00:19:52,891 --> 00:19:54,851
We got different outcomes in terms.
Cue 339 — 00:19:54,891 --> 00:19:59,371
We never hit the tokenmaxxing in terms of how people responded to that,
Cue 340 — 00:19:59,411 --> 00:20:01,131
and we were able to have realistic conversations.
Cue 341 — 00:20:01,171 --> 00:20:03,451
Hey, we'll move you around, will adjust your seat structure,
Cue 342 — 00:20:03,491 --> 00:20:05,691
we'll get you dialed in. But we love your enthusiasm.
Cue 343 — 00:20:05,731 --> 00:20:07,391
We want you to keep moving.
Cue 344 — 00:20:09,131 --> 00:20:13,531
April, we hit another little bit of a surprise, which was our product doesn't
Cue 345 — 00:20:13,571 --> 00:20:16,471
actually support agents. We're a developer tools product.
Cue 346 — 00:20:16,731 --> 00:20:21,171
Every developer is using it. But now every developer is asking their agent to use
Cue 347 — 00:20:21,211 --> 00:20:27,731
the product. And this I would say historically in the way that we've thought about
Cue 348 — 00:20:27,771 --> 00:20:34,451
product and planned and worked would have been, you know, maybe next quarter,
Cue 349 — 00:20:34,491 --> 00:20:37,071
we'll do something about that, right?
Cue 350 — 00:20:37,131 --> 00:20:39,311
Is that do you want to ask a question?
Cue 351 — 00:20:39,451 --> 00:20:42,991
Yeah. Go for it. Why not? I'm keeping an eye on the time, but let's let's try it.
Cue 352 — 00:20:46,851 --> 00:20:48,851
Exact same problem, right? Like.
Cue 353 — 00:20:48,891 --> 00:20:51,891
Well, sorry. How did it take that long to ask the question?
Cue 354 — 00:20:51,931 --> 00:20:55,821
Like, why doesn't our product like.
Cue 355 — 00:20:56,411 --> 00:21:03,031
Yeah, why did you. Yeah, I think I probably already
Cue 356 — 00:21:03,071 --> 00:21:06,111
was. It's totally it's a great question.
Cue 357 — 00:21:06,151 --> 00:21:09,031
And I'll finish the story and you'll understand what happened.
Cue 358 — 00:21:09,071 --> 00:21:12,751
But I was maybe I was doing something else.
Cue 359 — 00:21:12,791 --> 00:21:14,951
I can't totally remember. I don't remember the details.
Cue 360 — 00:21:14,991 --> 00:21:20,111
But what happened at this point, right, was our CEO, who's also a product person by
Cue 361 — 00:21:20,151 --> 00:21:24,471
background, sat down on a Friday and grabbed Claude and banged out these like set of
Cue 362 — 00:21:24,511 --> 00:21:25,911
documents which are totally available.
Cue 363 — 00:21:25,951 --> 00:21:29,071
You should all grab them if you want them for a good CLI, or like what makes the CLI
Cue 364 — 00:21:29,111 --> 00:21:34,791
great for agents. And then on Friday night I was at the ballet and he was like,
Cue 365 — 00:21:34,831 --> 00:21:36,271
I'm just going to build it real quick with Claude.
Cue 366 — 00:21:36,311 --> 00:21:38,591
Speaking of estimates, it was like Claude said, it's going to take three months.
Cue 367 — 00:21:38,631 --> 00:21:39,711
I bet we can get it done tonight.
Cue 368 — 00:21:39,751 --> 00:21:41,671
I was like, please don't. Right.
Cue 369 — 00:21:41,711 --> 00:21:47,191
This is everyone's like, you know, my, my, my head of engineering vibe coded the I'm
Cue 370 — 00:21:47,231 --> 00:21:49,031
like, it's better if I vibe coded.
Cue 371 — 00:21:49,071 --> 00:21:51,191
Sorry if I vibe coded over the weekend than you do.
Cue 372 — 00:21:51,231 --> 00:21:53,751
Right. And on Saturday, he was like, okay, I'm done.
Cue 373 — 00:21:53,791 --> 00:21:55,571
I'm like, okay, I've got home.
Cue 374 — 00:21:55,751 --> 00:21:57,591
I'm just he's like, you should just start from mine.
Cue 375 — 00:21:57,631 --> 00:22:03,441
I looked at his in. 98% of it was just importing our old broken CLI into a new CLI.
Cue 376 — 00:22:03,481 --> 00:22:07,071
And I was like, this is not how software gets built, right?
Cue 377 — 00:22:07,111 --> 00:22:09,481
Like, just just leave it with me for the weekend.
Cue 378 — 00:22:09,521 --> 00:22:13,131
And I showed up so sheepishly on Monday morning and I was like,
Cue 379 — 00:22:13,871 --> 00:22:15,451
I know y'all are going to hate me.
Cue 380 — 00:22:15,831 --> 00:22:20,551
But here's what happened. Right now we have a CLI and and it's like that was the
Cue 381 — 00:22:20,591 --> 00:22:25,631
base. But and then it turns out, you know, it's kind of to the point of your
Cue 382 — 00:22:25,671 --> 00:22:29,251
question, there were many people in the organization who were frustrated by the fact
Cue 383 — 00:22:29,591 --> 00:22:33,231
that one, our CLI wasn't even great for humans at the time, and two,
Cue 384 — 00:22:33,271 --> 00:22:38,831
it was garbage for agents. And so they jumped on it like, I will work nights and
Cue 385 — 00:22:38,871 --> 00:22:41,591
weekends to make this thing amazing sort of thing, which I'm not advocating.
Cue 386 — 00:22:41,631 --> 00:22:46,471
But when people are excited about stuff, they do that and and it just took off and
Cue 387 — 00:22:46,511 --> 00:22:49,271
now it's phenomenal. And the like take up as soon as we put it in the market.
Cue 388 — 00:22:49,311 --> 00:22:53,631
That take up was massive. Like we totally hit on something and it's not I mean,
Cue 389 — 00:22:53,671 --> 00:22:55,391
great question. Called out. I get it.
Cue 390 — 00:22:55,431 --> 00:22:59,891
Like I probably should have noticed earlier, but because we were in it,
Cue 391 — 00:23:00,311 --> 00:23:02,151
the real oh yeah, that's something we should do something about.
Cue 392 — 00:23:02,191 --> 00:23:03,731
Like let's get that done right now.
Cue 393 — 00:23:03,871 --> 00:23:06,791
And then we said, oh, you know, what we could do is we could layer an MCP server
Cue 394 — 00:23:06,831 --> 00:23:08,671
into the CLI and people started using that.
Cue 395 — 00:23:08,711 --> 00:23:10,871
So then we said, you know, maybe they want a remote MCP server.
Cue 396 — 00:23:10,911 --> 00:23:14,211
We put that out and that is like 10x the CLI.
Cue 397 — 00:23:14,311 --> 00:23:17,991
Right. So the like and then we see that just in growth of our product.
Cue 398 — 00:23:18,031 --> 00:23:21,911
So there was a lot that was just product insight that happened again from from using
Cue 399 — 00:23:21,951 --> 00:23:24,151
it. I'm going to think about that all night.
Cue 400 — 00:23:24,191 --> 00:23:25,561
Now I'm like why didn't I think of that?
Cue 401 — 00:23:25,601 --> 00:23:28,491
Because probably I was distracted trying to figure out how to get a factory running.
Cue 402 — 00:23:28,631 --> 00:23:30,251
All right. I'm going to keep moving.
Cue 403 — 00:23:31,511 --> 00:23:33,631
All right. And then we had it and we were like, cool.
Cue 404 — 00:23:33,671 --> 00:23:35,791
But the agent never does what we expected.
Cue 405 — 00:23:35,831 --> 00:23:41,031
And so then we built this whole lab basically that was like evals and reruns and
Cue 406 — 00:23:41,071 --> 00:23:43,911
take all the things that we expect people to do with our product,
Cue 407 — 00:23:43,951 --> 00:23:48,551
bake them into loops, and start figuring out like, what kind of feedback do we need
Cue 408 — 00:23:48,591 --> 00:23:51,951
to give from the CLI in order for them to make the right choices?
Cue 409 — 00:23:51,991 --> 00:23:55,991
And again, this is like we had this pent up interest, which is actually a really
Cue 410 — 00:23:56,031 --> 00:23:59,711
interesting thing in here as well in the organization.
Cue 411 — 00:23:59,751 --> 00:24:04,071
Right. It's easy because I think so many of us have struggled with people are
Cue 412 — 00:24:04,111 --> 00:24:06,671
resistant, they're having an identity crisis.
Cue 413 — 00:24:06,711 --> 00:24:11,271
We have to be calm and like guide them through this gentle path of the transition to
Cue 414 — 00:24:11,311 --> 00:24:14,991
maybe we don't sit in handwrite code anymore, but there were individuals in the
Cue 415 — 00:24:15,031 --> 00:24:20,311
organization who were so pumped, so ready to take this stuff on and just chomping at
Cue 416 — 00:24:20,351 --> 00:24:23,531
the bit, and they're like, my product manager told me, I have to do this thing
Cue 417 — 00:24:23,591 --> 00:24:26,271
right. And so when we unlocked that a little bit and said, hey,
Cue 418 — 00:24:26,311 --> 00:24:29,131
could you go experiment and figure out how to make this thing amazing?
Cue 419 — 00:24:29,191 --> 00:24:33,411
They just ran with it and had way better ideas than than everybody else.
Cue 420 — 00:24:35,151 --> 00:24:38,151
And then we kind of came back around to our, you know, to our home base as a CI
Cue 421 — 00:24:38,191 --> 00:24:39,671
provider. And we said validation is the key.
Cue 422 — 00:24:39,711 --> 00:24:40,751
Validation is huge. Right now.
Cue 423 — 00:24:40,791 --> 00:24:42,111
We're hearing all about this in the market.
Cue 424 — 00:24:42,151 --> 00:24:46,271
Anyway. And then we spent a couple of months getting faster and faster.
Cue 425 — 00:24:46,311 --> 00:24:51,791
But we hit this point and I drew this slide and sort of had this realization and
Cue 426 — 00:24:51,831 --> 00:24:55,191
talked with the team about it, where we thought we were going really quickly.
Cue 427 — 00:24:55,231 --> 00:24:58,191
I don't know if you can read this very well from the back, but basically our process
Cue 428 — 00:24:58,231 --> 00:25:01,351
was, you know, engineer asked their agent to implement something.
Cue 429 — 00:25:01,391 --> 00:25:04,211
Another engineer asked their agent to review that thing.
Cue 430 — 00:25:04,831 --> 00:25:07,151
Right. Each one of these is a wait time, right?
Cue 431 — 00:25:07,191 --> 00:25:10,231
And then they would put up comments, and then you'd come back the next day and say,
Cue 432 — 00:25:10,271 --> 00:25:12,671
oh, oh yeah, I forgot I was working on 512.
Cue 433 — 00:25:12,711 --> 00:25:17,831
Hey agent, could you go get the comments and like update my PR based on the
Cue 434 — 00:25:17,871 --> 00:25:22,231
comments? And then someone else shows up with different review prompts and we just
Cue 435 — 00:25:22,271 --> 00:25:26,691
go around and around and around and it's like we're moving so fast to get nowhere.
Cue 436 — 00:25:28,671 --> 00:25:32,091
And then we said, well, what if we just didn't wait?
Cue 437 — 00:25:32,351 --> 00:25:36,911
What if we just had the same engine sort of agent take those prompts and run the
Cue 438 — 00:25:36,951 --> 00:25:41,751
reviews? And so we built this a, you know, parallel refinery reviewer process,
Cue 439 — 00:25:41,791 --> 00:25:43,271
a coordinator that executes all that.
Cue 440 — 00:25:43,311 --> 00:25:45,031
You probably hear this referred to as a factory.
Cue 441 — 00:25:45,071 --> 00:25:47,811
And all these conversations that you're hearing in the last day or two.
Cue 442 — 00:25:48,271 --> 00:25:49,131
And,
Cue 443 — 00:25:50,351 --> 00:25:55,351
and now we're sort of well, the next step after that, we write,
Cue 444 — 00:25:55,391 --> 00:25:56,431
we should probably build a factory.
Cue 445 — 00:25:56,471 --> 00:25:59,111
Like now it's September and we're back to February, but we fixed a lot of things
Cue 446 — 00:25:59,151 --> 00:26:05,391
along the way. Again, that that we couldn't really unblock without understanding and
Cue 447 — 00:26:05,431 --> 00:26:08,671
driving through the hurdles. We we put all this together.
Cue 448 — 00:26:08,711 --> 00:26:10,431
We built a first version of this that was public.
Cue 449 — 00:26:10,471 --> 00:26:17,231
We ran a workshop at SF Tech Week in San Francisco, and we ran into this little
Cue 450 — 00:26:17,271 --> 00:26:21,671
conundrum. So I'm running a workshop on building an autonomous software factory.
Cue 451 — 00:26:21,711 --> 00:26:24,711
Right. Common topic that you're all hearing about here today.
Cue 452 — 00:26:24,751 --> 00:26:28,191
There are 84 people there. And at some point I asked for a show of hands of how many
Cue 453 — 00:26:28,231 --> 00:26:34,511
people, when they merge to main, that software goes to production and four people
Cue 454 — 00:26:34,551 --> 00:26:37,791
put up their hands. Okay, well, I got one so far.
Cue 455 — 00:26:37,831 --> 00:26:40,351
I wasn't actually asking for a show of hands because I'm a little nervous to tell
Cue 456 — 00:26:40,391 --> 00:26:42,471
that story and then, like, call you all out.
Cue 457 — 00:26:42,511 --> 00:26:48,111
But if you build an autonomous software factory to get everything to the shipping
Cue 458 — 00:26:48,151 --> 00:26:51,971
dock and you don't ship it, that factory did nothing for you.
Cue 459 — 00:26:52,191 --> 00:26:55,991
You have to get the software into the hands of your customers.
Cue 460 — 00:26:56,031 --> 00:26:56,971
And so.
Cue 461 — 00:26:58,951 --> 00:27:05,311
We basically combine these pieces together and then, you know,
Cue 462 — 00:27:05,351 --> 00:27:09,511
oh, sorry. There were then we realized, speaking of people that were chomping at the
Cue 463 — 00:27:09,551 --> 00:27:12,911
bit, we realized there were a bunch of people internally who were building these
Cue 464 — 00:27:12,951 --> 00:27:15,111
single player software factories for themselves.
Cue 465 — 00:27:15,151 --> 00:27:19,331
We took all of that energy, reprinted it into one product that we're now working on.
Cue 466 — 00:27:19,631 --> 00:27:22,351
And then my last point as someone is involved is when you show up in a bunch of
Cue 467 — 00:27:22,391 --> 00:27:25,951
engineers to build something really awesome, you ask the most basic product
Cue 468 — 00:27:25,991 --> 00:27:30,131
questions like, what does the user know how to do when they arrive on this page?
Cue 469 — 00:27:30,191 --> 00:27:32,871
And it turns out that everybody forgot about that.
Cue 470 — 00:27:32,911 --> 00:27:35,531
So there's a little like bringing it back to the basics.
Cue 471 — 00:27:36,111 --> 00:27:39,051
I see my timer guy lining up. So to wrap it all up.
Cue 472 — 00:27:40,791 --> 00:27:42,271
I think it's important to lead.
Cue 473 — 00:27:42,311 --> 00:27:45,111
I think it's important to be at a high level, but in order to do that,
Cue 474 — 00:27:45,151 --> 00:27:46,431
you need to understand a few things.
Cue 475 — 00:27:46,471 --> 00:27:48,631
You need to know where your company sits in the market.
Cue 476 — 00:27:48,671 --> 00:27:52,311
You need to understand where you can have impact, and then you need to be able to
Cue 477 — 00:27:52,351 --> 00:27:55,991
drive and execute. And in order to do that, you have to have ground truth.
Cue 478 — 00:27:56,031 --> 00:27:58,751
And there's only so many ways to get ground truth when everything is changing.
Cue 479 — 00:27:58,791 --> 00:28:01,051
And the one thing I would call them, they're going to read this.
Cue 480 — 00:28:01,871 --> 00:28:04,191
I say this probably every day about myself.
Cue 481 — 00:28:04,231 --> 00:28:07,791
I know that I'm wrong. I don't know how wrong or why I'm wrong,
Cue 482 — 00:28:07,831 --> 00:28:10,431
but the faster we move, the faster I'm going to find out why we're,
Cue 483 — 00:28:10,471 --> 00:28:11,631
why I'm wrong, why we're wrong.
Cue 484 — 00:28:11,671 --> 00:28:17,120
And then we can go fix it. So stay humble, keep learning and just execute like mad.
Cue 485 — 00:28:17,471 --> 00:28:19,211
All right. Thank you so much.
Cue 486 — 00:28:23,311 --> 00:28:25,911
Awesome. Thank you. Thank you.
Cue 487 — 00:28:25,951 --> 00:28:29,651
We have one minute. If anyone has any final questions.
Cue 488 — 00:28:36,851 --> 00:28:40,731
So you've talked about this kind of difficult mutual trust situation where you've
Cue 489 — 00:28:40,771 --> 00:28:44,891
got engineers that maybe are a little suspect of how you're going to roll this out,
Cue 490 — 00:28:44,931 --> 00:28:49,171
and you are sitting there thinking, I'm not sure how to do my job anymore.
Cue 491 — 00:28:49,211 --> 00:28:53,451
That's why I've got to, you know, go, you know, play in the in the fields of code
Cue 492 — 00:28:53,491 --> 00:28:59,531
and all that. I would love to, to be the sort of developer in a company that if the
Cue 493 — 00:28:59,571 --> 00:29:03,371
CTO felt like they didn't really know what was going on on the floor,
Cue 494 — 00:29:03,411 --> 00:29:05,851
that I'd be able to help them.
Cue 495 — 00:29:05,891 --> 00:29:10,971
But there is that kind of intrinsic difficulty where an executive can't exactly say
Cue 496 — 00:29:11,011 --> 00:29:13,791
the words, hey, I don't know what's going on.
Cue 497 — 00:29:14,491 --> 00:29:20,211
How would you advise someone like me to, to kind of be able to,
Cue 498 — 00:29:20,251 --> 00:29:24,531
to, to make those inroads and be a reliable source of information for someone like
Cue 499 — 00:29:24,571 --> 00:29:27,091
you and you're a senior developer.
Cue 500 — 00:29:27,131 --> 00:29:30,171
What role are you in? I'm typically a tech lead.
Cue 501 — 00:29:30,211 --> 00:29:33,971
Okay, okay. Be honest, number one.
Cue 502 — 00:29:34,011 --> 00:29:35,851
I mean, your mileage may vary.
Cue 503 — 00:29:35,891 --> 00:29:37,531
Your company varies. Like, there's all these things.
Cue 504 — 00:29:37,571 --> 00:29:40,781
I think that, you know, I glossed over my history, but I've.
Cue 505 — 00:29:40,821 --> 00:29:43,191
I've been at CircleCI for 12 years.
Cue 506 — 00:29:43,331 --> 00:29:46,611
You know, when I started, I was like, kind of junior developer or whatever.
Cue 507 — 00:29:46,651 --> 00:29:49,831
I mean, I was a CTO within seven months of getting there, but,
Cue 508 — 00:29:50,011 --> 00:29:53,491
so I have a lot of history with a lot of those people, and I invest a lot in that
Cue 509 — 00:29:53,531 --> 00:29:57,611
history. And I think that the or in that trust, and I think one of the biggest
Cue 510 — 00:29:57,651 --> 00:30:00,231
things that now I'm going to give you an answer from my perspective,
Cue 511 — 00:30:00,931 --> 00:30:05,251
but one of the biggest ways to build and maintain that trust is to ask for feedback
Cue 512 — 00:30:05,291 --> 00:30:06,971
and then to act on the feedback.
Cue 513 — 00:30:07,011 --> 00:30:10,531
And when people don't give me honest feedback, I can't do anything with that.
Cue 514 — 00:30:10,571 --> 00:30:14,131
Right? So the best thing that you can do in your situation is give that honest
Cue 515 — 00:30:14,171 --> 00:30:16,211
feedback, and you're going to learn a lot about your leaders,
Cue 516 — 00:30:16,251 --> 00:30:18,891
right? If you're afraid to give honest feedback to your leaders,
Cue 517 — 00:30:18,931 --> 00:30:24,131
like that's a signal to you and you could sort that out, I am I think the how can I
Cue 518 — 00:30:24,171 --> 00:30:29,011
get to this quickly? Every conversation that I have ever thought was going to be
Cue 519 — 00:30:29,051 --> 00:30:32,731
awful, that I put off one one, don't put it off, because then you're just going to
Cue 520 — 00:30:32,771 --> 00:30:35,291
carry the weight of it. Just go have the conversation.
Cue 521 — 00:30:35,331 --> 00:30:39,491
And two, within 10 seconds of starting that conversation, it turns out to just be
Cue 522 — 00:30:39,531 --> 00:30:41,491
like a regular human conversation.
Cue 523 — 00:30:41,531 --> 00:30:45,731
I have like the most terrifying of the conversations that I've had in my career have
Cue 524 — 00:30:45,771 --> 00:30:49,571
gone from absolutely terrifying to totally fine within 10 seconds because I just
Cue 525 — 00:30:49,611 --> 00:30:54,411
went and had them right. So if you feel like and if you work at a place where being
Cue 526 — 00:30:54,451 --> 00:30:56,451
really honest with your boss is going to get you thrown out the door,
Cue 527 — 00:30:56,491 --> 00:30:58,171
like ask some questions about where you work.
Cue 528 — 00:30:58,211 --> 00:31:00,411
But I think people build up things in their head.
Cue 529 — 00:31:00,451 --> 00:31:02,251
The best thing you can do and do it slowly.
Cue 530 — 00:31:02,291 --> 00:31:04,291
Find maybe easier. Don't pick the hardest topic.
Cue 531 — 00:31:04,331 --> 00:31:06,771
If you've never had an honest conversation, pick something simpler,
Cue 532 — 00:31:06,811 --> 00:31:12,491
but start building that trust the other way and I to the best that I can commit.
Cue 533 — 00:31:12,531 --> 00:31:15,891
Your mileage may vary. Would say that you will become a more valuable member of the
Cue 534 — 00:31:15,931 --> 00:31:18,631
organization if you give that honest feedback to your leaders,
.tessl-plugin
skills
talk-arcolano-autonomy-and-trust
talk-aroussi-decision-context
talk-bevin-agent-assisted-instrument-design
talk-boeckeler-context-lifecycles
talk-braun-context-graphs
talk-carmont-evolving-agent-harnesses
talk-chanezon-gordon-agent-experience-evaluation
talk-cox-secure-agentic-workflows
talk-cunniffe-reusable-agent-platform
talk-davis-agent-tool-policies
talk-debois-concurrent-delivery
talk-dzialets-team-automation
talk-fahmy-harness-api-design
talk-finster-continuous-delivery
talk-fowler-regenerative-software
talk-gavrilescu-factory-architecture
talk-graciolli-amin-verification-aware-programming
talk-gruenbaum-reverse-graph-engineering
talk-hassan-agentic-engineering
talk-hossain-enterprise-agent-adoption
talk-huntley-patch-ready-factories
talk-jain-verifying-intent
talk-katsioloudes-montei-agentic-lifecycle
talk-knox-harness-engineering
talk-kovac-deterministic-memory
talk-lee-perkins-code-readiness
talk-lorenc-machine-speed-security
talk-lubken-software-of-one
talk-meijer-software-specialization
talk-meppiel-versioned-agent-contracts
talk-moss-concurrent-agent-risk
talk-myers-understanding-agent-generated-code
talk-ocytko-scaling-agentic-engineering
talk-oza-automating-design-quality
talk-panzica-agentic-sdlc
talk-pari-patel-agent-awareness
talk-podjarny-trustworthy-software-factories
talk-poulter-factory-improvement
talk-pratik-patel-refined-rag-data
talk-re-ferre-development-platforms
talk-sadogursky-factory-of-factories
talk-shah-training-agent-skills
talk-sharma-mohapatra-evolving-intent
talk-singhal-operating-agents-at-scale
talk-stack-incident-dossiers
talk-tyler-team-coding-benchmarks
talk-wotherspoon-agent-execution-environments
talk-yegge-ai-powered-factories
talk-zuber-leadership-through-coding