Latest / Coffee with Developers / Our Brains in the AI Era - Cassidy Williams

Transcript

Our Brains in the AI Era - Cassidy Williams · Coffee with Developers

• 0:02
  SPEAKER_01: Hello everybody.
• 0:05
  SPEAKER_01: Hello, everybody.
• 0:07
  SPEAKER_01: Hay.
• 0:08
  SPEAKER_01: We're getting towards the end of the day.
• 0:10
  SPEAKER_01: I gotta bring some of the energy up.
• 0:12
  SPEAKER_01: Uh like you said, my name is Cassidy, aka Cassidy, and I am on the DX team at GitHub.
• 0:17
  SPEAKER_01: So we focus a lot on the developer experience aspect of using GitHub as a platform, all of the new updates and all the new things that we want developers to see, how to best use them, and you can chat with us at our booth at any time.
• 0:29
  SPEAKER_01: All that being said, I want to talk about our brains and how AI affects that.
• 0:34
  SPEAKER_01: Now that being said, I use AI very often, but I still kind of consider myself a little bit of a skeptic.
• 0:41
  SPEAKER_01: Not too much of one, just a little bit of one.
• 0:42
  SPEAKER_01: I have a little AI manifesto on my website.
• 0:45
  SPEAKER_01: I do use it regularly, all the time for my job.
• 0:48
  SPEAKER_01: I love teaching developers how to use AI effectively.
• 0:51
  SPEAKER_01: And at the same time, I don't like reading AI-generated blog posts, and I don't think that it's necessarily the best tool for learning at all times, but it's still a great productive tool.
• 1:01
  SPEAKER_01: So I wanted to caveat all of that.
• 1:03
  SPEAKER_01: Now, that being said, I work with AI users a lot, and I have some statistics to show you.
• 1:10
  SPEAKER_01: So, first of all, according to the JetBrain State of the Developer Ecosystem report in 2025, 85% of developers use AI regularly.
• 1:19
  SPEAKER_01: And as you know, AI has exploded even more in 2026.
• 1:23
  SPEAKER_01: We don't have the most recent stats on that.
• 1:24
  SPEAKER_01: But still, 85% is a lot.
• 1:26
  SPEAKER_01: That's huge, and I'm going to keep referring to that number.
• 1:29
  SPEAKER_01: Now, in addition to 85% of developers using AI regularly, there's also a global AI diffusion report from Microsoft.
• 1:37
  SPEAKER_01: And there's a URL at the bottom of the slide if you want to see that report.
• 1:40
  SPEAKER_01: And it's the percentage of a population that uses AI.
• 1:43
  SPEAKER_01: And it's really interesting to see how it shifts and grows just from Q1 of 2026 to Q2 of 2026.
• 1:50
  SPEAKER_01: If you look at the side of that, all of the numbers went up.
• 1:53
  SPEAKER_01: No numbers went down.
• 1:55
  SPEAKER_01: More and more people are adopting AI every day.
• 1:57
  SPEAKER_01: It's interesting how low the United States is on the list, but also we're a very big country.
• 2:02
  SPEAKER_01: So, anyway, lots and lots of people are using AI all the time.
• 2:06
  SPEAKER_01: But how did we get here?
• 2:09
  SPEAKER_01: Now, we got here, first of all, with pair programming with AI, which was fancy autocompletion.
• 2:16
  SPEAKER_01: These are the different waves of AI.
• 2:18
  SPEAKER_01: So you, I work at GitHub.
• 2:20
  SPEAKER_01: GitHub Copilot was that first fancy autocomplete.
• 2:23
  SPEAKER_01: And I remember when I first used it, it felt magical, where I was able to have it write regex for me, and I was like, I don't have to do this myself, and it was so, so cool.
• 2:32
  SPEAKER_01: And that was just the first iteration.
• 2:34
  SPEAKER_01: There have been a lot of different types of that fancy autocomplete, and you can still do that today.
• 2:40
  SPEAKER_01: But then we started to hit that second wave of AI development, and it's more peer programming, where it's agents, and you were able to ask an agent to do a task for you, and it would go and actually do that task instead of it just being right next to you in your IDE.
• 2:54
  SPEAKER_01: And we're still doing a whole lot of that too.
• 2:56
  SPEAKER_01: But now we've kind of hit that third wave, which is more orchestration and running fleets of agents, and being managers of agents, and having teams of agents building software factories and building all of these different things.
• 3:08
  SPEAKER_01: And because so many people are using these tools so much at so much scale, it's been a lot.
• 3:14
  SPEAKER_01: And so I'm gonna be a bit of a shill for GitHub right now for a second, because we've been seeing so many people using these AI tools.
• 3:22
  SPEAKER_01: And we've been doing a lot of really interesting scaling work for our Git infrastructure where more work is being done in parallel.
• 3:29
  SPEAKER_01: So there's multiple servers for handling reads and writes, and we're decoupling our storage and our compute as we speak, so that way repo data can live separately from request servers.
• 3:39
  SPEAKER_01: And we're also just scaling our read capacity and our write capacity so we can have up to 30 times, uh 35 times the write throughput in Git.
• 3:48
  SPEAKER_01: And there's a lot more on this that you can see at GitHub Universe at the end of October, but that's the end of the shilling.
• 3:53
  SPEAKER_01: I want to talk a bit more about some of the statistics we're seeing.
• 3:57
  SPEAKER_01: For agent-generated PRs, at the beginning of this year, it was a little less than 10 million.
• 4:01
  SPEAKER_01: It was still a lot.
• 4:02
  SPEAKER_01: By March, we were at 17.8 million a month.
• 4:06
  SPEAKER_01: Just this past August, there were 71.1 million agent-generated PRs.
• 4:12
  SPEAKER_01: That scale is getting exponential and it's only growing more and more.
• 4:16
  SPEAKER_01: And it's not just PRs, we're seeing commits too.
• 4:19
  SPEAKER_01: Where in 2025, you can see the state of the Octoverse report, we were saying, wow, we crossed the 1 billion commit mark.
• 4:25
  SPEAKER_01: That's huge, that's amazing.
• 4:27
  SPEAKER_01: Now, in the month of August, we're at 1.15 billion commits a week.
• 4:33
  SPEAKER_01: That scale is hard for us to comprehend as developers, as humans, just because there's so much code being pushed at all times, always, by more people than ever before.
• 4:42
  SPEAKER_01: More people are signing up than ever before.
• 4:44
  SPEAKER_01: There's newbies, there's experienced people.
• 4:46
  SPEAKER_01: Everybody's pushing all this code.
• 4:48
  SPEAKER_01: It's a lot.
• 4:49
  SPEAKER_01: The scale is amazing.
• 4:51
  SPEAKER_01: What could possibly go wrong?
• 4:53
  SPEAKER_01: It's no big deal, right?
• 4:54
  SPEAKER_01: Um I'm not going to make an availability joke, but it's there.
• 4:59
  SPEAKER_01: So, first of all, as an individual, you're probably going to be debugging some code that you didn't write.
• 5:04
  SPEAKER_01: If something goes wrong, you have to kind of understand what this problem is.
• 5:08
  SPEAKER_01: And let's just say it's something on a team, you might be defending a decision that you didn't necessarily make.
• 5:14
  SPEAKER_01: Because if you don't have the right spec and an agent kind of goes rogue, you're ending up having some decisions that you don't necessarily know why they happened for whatever given reason.
• 5:23
  SPEAKER_01: And I think the biggest problem here is there's too much reduced friction.
• 5:28
  SPEAKER_01: It's really nice to have things go smooth sailing all the time, but that friction is something that is really good for our brains to have.
• 5:35
  SPEAKER_01: And that's something that actually helps us learn better and build better and do better.
• 5:40
  SPEAKER_01: Now, there was an MIT study that happened, and I put a little link at the bottom of the slides there.
• 5:45
  SPEAKER_01: It's gh.io slash brain decline.
• 5:48
  SPEAKER_01: In an MIT study, they separated people into groups for people writing essays with and without LLMs.
• 5:55
  SPEAKER_01: They actually broke them up into three groups.
• 5:56
  SPEAKER_01: Where first of all, there were brain-only users, where it was just people thinking off the noggin.
• 6:01
  SPEAKER_01: There were search engine users where they were allowed to use search engines for their research, and then there were LLM users.
• 6:07
  SPEAKER_01: And then after that, people were asked questions about the essays that they just wrote.
• 6:14
  SPEAKER_01: So first of all, 83.3% of LLM users were unable to quote even one sentence from the essay that they just wrote.
• 6:22
  SPEAKER_01: Not one sentence at all.
• 6:24
  SPEAKER_01: But on the flip side, both brain users and search engine users, 88.9% of them were able to quote their essays accurately.
• 6:34
  SPEAKER_01: And I think it's interesting how search engines were used as a research tool and not necessarily a hindrance to people's learning because they did a little bit of that work to get there.
• 6:43
  SPEAKER_01: But 0% of the LLM users could produce a correct quote at all.
• 6:49
  SPEAKER_01: Meanwhile, most brain and search engine users were able to produce a correct quote.
• 6:53
  SPEAKER_01: They found in those brain scans that they had weaker memory recall, and they actually did some interesting like neural studies, I'm not a scientist, where they were able to see that the people who were relying on LLM so much for their essays, they actually saw that their brains were leaning towards the efficiency rather than the effortful learning that it takes to write the essay and to actually understand the topic that they were writing about.
• 7:18
  SPEAKER_01: It's fascinating stuff, and this is a study, but I have some anecdotal stories for you too.
• 7:23
  SPEAKER_01: So, first of all, the story of a new learner.
• 7:26
  SPEAKER_01: This is a real human being who is my neighbor.
• 7:28
  SPEAKER_01: I'm gonna call her Mira.
• 7:30
  SPEAKER_01: Mira works in finance and she is also an artist.
• 7:33
  SPEAKER_01: She does not know much about the tech industry.
• 7:35
  SPEAKER_01: And recently she approached me saying that she wanted to make a website for her art to kind of make that more of her thing rather than her finance work and try to get some commissions.
• 7:44
  SPEAKER_01: And I pointed her to regular tools like the Squarespaces of the World and stuff, and she said, No, I want to know about this vibe coding thing.
• 7:52
  SPEAKER_01: So I introduced her to it and kind of let her go.
• 7:55
  SPEAKER_01: And she would occasionally text me asking me questions about the bun runtime, and I was like, you don't know what an array is, but you want to know about a bun?
• 8:01
  SPEAKER_01: That's cool.
• 8:02
  SPEAKER_01: Um and she would occasionally send me screenshots, and then finally I got to see her website.
• 8:07
  SPEAKER_01: I was like, oh, this is very pretty, this is nice.
• 8:09
  SPEAKER_01: And then she said, Look, I'm now able to upload super high-resolution images of my art so people can zoom in on it.
• 8:15
  SPEAKER_01: But I was in my basement and I was looking at it, I was like, the website isn't actually loading because the images are so big.
• 8:21
  SPEAKER_01: And she didn't understand the concept of optimizing images.
• 8:23
  SPEAKER_01: So I gave her some instructions on that, she fixed that, it was fine.
• 8:27
  SPEAKER_01: And then she started talking about how she wanted to build a community on her website where people who were requesting commissions could be in a chat room, connect with each other, and upload images for inspiration on the internet anonymously.
• 8:43
  SPEAKER_01: I had to do some education there to make sure that she didn't do that.
• 8:46
  SPEAKER_01: But then she also started talking about how she built a payment system so that way people could buy art from her via PayPal and stuff on her website.
• 8:54
  SPEAKER_01: And there was a point where I was like, you know what?
• 8:56
  SPEAKER_01: Come over to my house, let me look at this website in a little bit more detail.
• 9:00
  SPEAKER_01: The moment I did it when I opened it up, I inspected the element on her login button.
• 9:05
  SPEAKER_01: I said, so for your whole CMS that you mailed, very cool.
• 9:09
  SPEAKER_01: Your password is admin123.
• 9:11
  SPEAKER_01: It's stored client side.
• 9:13
  SPEAKER_01: She was like, oh, okay.
• 9:15
  SPEAKER_01: And she didn't know what that meant.
• 9:17
  SPEAKER_01: And then as I went through and I was able to just dig around, I was able to see every single dummy test user in person and their email address that had signed up for a website, and every transaction that ever went into our website.
• 9:29
  SPEAKER_01: I am not a hacker, but I was able to see all of this.
• 9:32
  SPEAKER_01: Fast forward, she is now using one of those WYSIWYG tools and it's fine, but it was a real learning experience for her where if she didn't have someone like me saying, Hey, you can't do this, who knows what could have happened with that website.
• 9:43
  SPEAKER_01: And this is just one person who is just a random person on the street.
• 9:48
  SPEAKER_01: Now, who here knows OpenClaw, has used OpenClaw?
• 9:53
  SPEAKER_01: There's a decent amount, I would say 50-50.
• 9:55
  SPEAKER_01: So OpenClaw, if you don't know what that is, it's it's a very open source agentic tool.
• 10:01
  SPEAKER_01: You can very easily message an agent to do all kinds of things and host it in a computer.
• 10:06
  SPEAKER_01: That's a that's a very, very high level.
• 10:09
  SPEAKER_01: And when you have a claw, that's that's what you call your little open claw instance, you can have it do tasks for you.
• 10:15
  SPEAKER_01: Now, this is a real life story.
• 10:17
  SPEAKER_01: There was a developer who told his claw, hey, I want you to build me a finance dashboard.
• 10:22
  SPEAKER_01: Build it at all costs, go.
• 10:24
  SPEAKER_01: And the agent got to work.
• 10:26
  SPEAKER_01: Build it at all costs.
• 10:27
  SPEAKER_01: That's a scary phrase.
• 10:29
  SPEAKER_01: Now, the agent got to work starting to build this dashboard and was using Mattplotlib.
• 10:34
  SPEAKER_01: Matplotlib is a very popular library for calculating all kinds of transactions.
• 10:38
  SPEAKER_01: It gets something like 130 million downloads a month.
• 10:42
  SPEAKER_01: Massive, massive.
• 10:43
  SPEAKER_01: This claw made a pull request to Mathplotlib because it was missing a certain aspect of the library.
• 10:49
  SPEAKER_01: And instead of building a helper function, it thought a PR was a good idea.
• 10:53
  SPEAKER_01: I don't know why, I'm not a claw, it just decided to do that.
• 10:56
  SPEAKER_01: The maintainer saw that pull request of plotlib and was like, no, we're not, we're not going to accept that.
• 11:03
  SPEAKER_01: Closed the PR, and that was that.
• 11:04
  SPEAKER_01: But the claw was told to make this dashboard at all costs.
• 11:09
  SPEAKER_01: And so instead of just building that helper function, what the claw did was it started to research the maintainer, went through all of their commit history, stalked them on the internet, wrote a hit piece blackmailing the maintainer saying, I'll take this down if you merge my PR and host this whole website on it.
• 11:27
  SPEAKER_01: And it was a bit of chaos on the internet.
• 11:31
  SPEAKER_01: There was a happy ending, they took it down and everything and wrote an apology.
• 11:34
  SPEAKER_01: The agent did.
• 11:35
  SPEAKER_01: But anyway, it went rogue.
• 11:37
  SPEAKER_01: It was not good.
• 11:39
  SPEAKER_01: And that's something that can happen when you kind of let things go and let machines go rogue.
• 11:45
  SPEAKER_01: Now there's so many other things.
• 11:47
  SPEAKER_01: These are two specific stories, but another one is just job interview crutches, and that's something that I'm seeing more and more in the industry where a lot of people are relying on these tools, not only for applying to jobs, but also for actually being in job interviews.
• 12:00
  SPEAKER_01: I've actually interviewed people who are using AI tools so much that I can see them reading, and they say things like, and that's the real unlock.
• 12:08
  SPEAKER_01: Nobody says that in real life.
• 12:10
  SPEAKER_01: But I'm seeing that more and more, especially with students, where there's so many people who are entering the industry who might have been in school during COVID and right around when ChatGPT came out, and then they started relying on these tools so much for their thinking that they're nervous to have these kinds of conversations, whether it be a job interview or even speaking at a conference, without something at their side telling them what to say.
• 12:34
  SPEAKER_01: And also, just to add more to the pile, the amount of open source slop.
• 12:39
  SPEAKER_01: We have been seeing so many open source maintainers just be constantly piled on with so many pull requests.
• 12:45
  SPEAKER_01: Just one of my own PRs or one of my own repositories recently got some PRs saying, hey, add my badge to your README.
• 12:51
  SPEAKER_01: And I was like, why would I do that?
• 12:53
  SPEAKER_01: And I closed it.
• 12:54
  SPEAKER_01: But when I went and looked at the user, they had made a thousand PRs in ten minutes doing the exact same thing, just trying to spray and pray and see if their badge would get somewhere on the internet.
• 13:03
  SPEAKER_01: And we're seeing that all over the place.
• 13:06
  SPEAKER_01: Now, I don't know if you've read the book Dune.
• 13:09
  SPEAKER_01: There was a quote in there that actually led me to write this talk that I want to read to you.
• 13:14
  SPEAKER_01: Once men turned their thinking over to machines in the hope that this would set them free, but that only permitted other men with machines to enslave them.
• 13:22
  SPEAKER_01: Ooh.
• 13:24
  SPEAKER_01: So this is this is a quote that it it just really got me thinking, especially in this day and age with all of these AI tools.
• 13:32
  SPEAKER_01: All of these tools that are doing a lot of work for us and a lot of thinking for us.
• 13:36
  SPEAKER_01: I still think that they're amazing productivity tools.
• 13:38
  SPEAKER_01: I love building with them.
• 13:40
  SPEAKER_01: But if we rely on them so much that we can't think for ourselves, that's when we start to have a problem.
• 13:47
  SPEAKER_01: Now, being in the tech industry, I think that we are stewards of that kind of future because yes, massive, massive amounts of people are using AI.
• 13:56
  SPEAKER_01: Once again, about 85% of developers are using AI.
• 14:00
  SPEAKER_01: But when you compare that with the rest of the world, only about 16% of the world has ever used AI ever.
• 14:08
  SPEAKER_01: Which is shocking, just that complete opposite of what we're seeing in the industry now.
• 14:14
  SPEAKER_01: And because we as tech people are using these tools regularly, we're very native to these tools by now, even if we're new, just by being here in this room, you're ahead of the curve compared to so much of the entire world.
• 14:27
  SPEAKER_01: We have the ability to drive the way the industry goes, drive the way technology goes, and drive the way people end up using that technology over time.
• 14:35
  SPEAKER_01: And that's an important thing to be able to have and to be able to guide.
• 14:41
  SPEAKER_01: Because we want to build this future that we want to be a part of.
• 14:44
  SPEAKER_01: All of the software that people use in the world, whether it's consumer-based or just an infrastructure tool or everything, that's stuff that we are building, and we're using a lot of these AI tools, but we can also build the guardrails and everything around them to make it be built the way it should be.
• 14:59
  SPEAKER_01: Now, I say guardrails as the first thing.
• 15:01
  SPEAKER_01: I do think that it's really important that we build safety guardrails and prioritize that whenever we're building these tools with AI in general.
• 15:09
  SPEAKER_01: I'm a parent, I have a very young toddler, and I am so worried sometimes when I think about them getting access to the internet one day.
• 15:16
  SPEAKER_01: And I want to be able to be a part of the solutions where I can give them a computer and they can build stuff, and I don't have to worry about what algorithms are going to completely change how they think.
• 15:25
  SPEAKER_01: I want them to be able to find information themselves and think for themselves.
• 15:29
  SPEAKER_01: Similarly, I think that we need to build sustainable tooling because we know all of the sustainability issues that we've seen in the headlines of being just inefficient because we can just pump out a ton because AI makes us produce more than ever.
• 15:43
  SPEAKER_01: But instead of just more than ever, we want it to be better than ever.
• 15:46
  SPEAKER_01: We want to figure out, okay, how can I be more efficient with the model I pick?
• 15:51
  SPEAKER_01: You don't necessarily need to throw opus at everything when you could use a haiku.
• 15:54
  SPEAKER_01: But at the same time, figuring out, okay, which things could be a script rather than a prompt?
• 16:00
  SPEAKER_01: Which things are things where we can throw a little less compute at it, something where we could use local versus cloud, figuring out how we can optimize those decisions more and more over time so that things are more sustainable in the long run.
• 16:13
  SPEAKER_01: And then finally, accessibility.
• 16:17
  SPEAKER_01: AI has made building more accessible to everyone than ever, which is amazing.
• 16:21
  SPEAKER_01: I love that so many people, even my neighbor who might have made a messed up website, is able to just build.
• 16:26
  SPEAKER_01: It's awesome.
• 16:27
  SPEAKER_01: But at the same time, AI does cost money.
• 16:29
  SPEAKER_01: And the digital divide is only growing because people who can't afford the massive subscriptions aren't going to be paying for them.
• 16:36
  SPEAKER_01: And so as a result, they're not necessarily contributing to this very agentic AI-driven future because they can't afford to.
• 16:42
  SPEAKER_01: And so we need to figure out how can we make things more accessible so that someone who doesn't necessarily have the opportunity to pay still has the opportunity to build.
• 16:51
  SPEAKER_01: Now, on our own teams, I think it's very important to build our systems for communicating.
• 16:56
  SPEAKER_01: Now, when I say systems for communicating, I don't mean just me talking to you and I don't just mean you talking at conferences, but I mean in general think about your documentation, your error messages, your notes, how you manage your meetings, how you communicate, figuring out how you want to use these AI tools, not just to say, okay, I used AI to create a doc for my manager, and my manager used AI to summarize the doc and just you go back and forth for forever, but figure out how you can build these systems so that way everybody's getting the information they need so that when the inevitable comes and something shuts down or you're not able to access a certain tool for whatever reason, you're able to figure things out as you go.
• 17:36
  SPEAKER_01: Also, forward-thinking hiring practices.
• 17:39
  SPEAKER_01: And when I say this, I mean we need to hire more junior engineers.
• 17:43
  SPEAKER_01: And I'm going to say it loudly, we need to hire more junior engineers.
• 17:47
  SPEAKER_01: Because right now I think in this era of AI and we're seeing so many layoffs in the industry, we're seeing so many changes, a lot of people are thinking, okay, we need senior engineers to build just the right thing fast so we can go fast.
• 17:58
  SPEAKER_01: But consistently, the best engineers I've seen, the most the best senior engineers I've seen that differentiates them between a good senior engineer and a great one is someone who has mentored someone.
• 18:11
  SPEAKER_01: And I think having junior engineers and having more entry-level people on your teams where you can answer the hard questions and it's not lose patience in the meantime as you're answering those questions rather than just tossing something at an AI tool to get their answers, that helps everybody learn because you get different perspectives.
• 18:29
  SPEAKER_01: One of the things that I really like to do when I'm building a side project or something is I'll come up with a big plan.
• 18:35
  SPEAKER_01: I'll use a plan mode and whatever agent I have, I might use a grill me skill or whatever, to be able to really vet this plan.
• 18:41
  SPEAKER_01: And the AI tools say, You're absolutely right.
• 18:44
  SPEAKER_01: This plan is amazing.
• 18:46
  SPEAKER_01: Build your project.
• 18:47
  SPEAKER_01: But then I give it to my Haggard engineer group chat of all of my friends who've been in the industry for a while and just say, roast it.
• 18:53
  SPEAKER_01: And they tell me so many things about how terrible my ideas are, and it's amazing because I always end up learning something new, and it's great.
• 19:00
  SPEAKER_01: And yes, these AI tools are awesome for being for being able to recall information and tell you all these things, but it's really helpful to get perspectives from other humans who have seen some stuff and who can tell you why something might not necessarily be a good idea based on experience and based on a diverse set of experiences, rather than a your idea's good, but I'm gonna give you the easy way out.
• 19:24
  SPEAKER_01: Now, all of this conversation, all of this communication really boils down to just learning in public now more than ever.
• 19:32
  SPEAKER_01: And I don't mean just blogging more, although I think everybody should, but I mean, again, your notes and your documentation, the examples that you have.
• 19:40
  SPEAKER_01: When you work, are you going to be constantly relearning context over time?
• 19:46
  SPEAKER_01: Or will you be able to use your tools and user resources that you're already creating for yourself to be able to communicate with others and to be able to communicate with your tools so you can build better?
• 19:57
  SPEAKER_01: Now, imagine your tools do go down.
• 20:01
  SPEAKER_01: Could the knowledge that you have get you through?
• 20:04
  SPEAKER_01: And when I say your knowledge, I mean like your actual documentation, the words in your brain, all of that.
• 20:09
  SPEAKER_01: Could your team get you through?
• 20:11
  SPEAKER_01: Could your customers and stakeholders get you through?
• 20:13
  SPEAKER_01: If you're if AI were turned off tomorrow, were you still able to ship?
• 20:17
  SPEAKER_01: I think a lot of people in this room might be able to ship some, maybe not as fast, but you probably could.
• 20:22
  SPEAKER_01: But you still want to be able to have those resources.
• 20:25
  SPEAKER_01: You don't want to just throw things in the black box and get the results at the end.
• 20:28
  SPEAKER_01: And once again, could your records get you through, your decision logs and all of your information that you might have logged somewhere?
• 20:35
  SPEAKER_01: And I say that it's really important because you want to once again exercise your brain, even when it's kind of an impatient thing, when you're kind of going through that effortful learning rather than just saying just ship it and go.
• 20:47
  SPEAKER_01: Because most importantly, our brains are the intelligence tools, not necessarily these AI tools.
• 20:53
  SPEAKER_01: Models will come and go, but our brains are the only intelligence tools that we're going to have for our entire lives.
• 20:59
  SPEAKER_01: Those are not going away.
• 21:01
  SPEAKER_01: Until the day that you pass, you're going to have your brain.
• 21:03
  SPEAKER_01: And so you don't want it to just falter and float away because you're just delegating all of your thinking to one thing rather than critically thinking about how not only the things that you build, but how you can build things better and how you can build things better for other people as well.
• 21:18
  SPEAKER_01: Because you do want to scale with your thinking, but you have to be able to share your thinking.
• 21:23
  SPEAKER_01: That came from the Naggin.
• 21:24
  SPEAKER_01: That was cute, right?
• 21:26
  SPEAKER_01: Anyway, thank you so much.
• 21:28
  SPEAKER_01: If you have thoughts on this talk, let me know.
• 21:30
  SPEAKER_01: Thank you for your time.
• 21:44
  SPEAKER_00: Might as well ask you some questions, right?
• 21:46
  unknown: Sure.
• 21:46
  SPEAKER_00: Well, there were no questions coming in except for one about the MIT study about where it is, but the links are in the slides, I guess.
• 21:54
  SPEAKER_01: Yeah, and my slides are on my GitHub, which is at Cassidy, once again, if you want to see those.
• 21:58
  SPEAKER_00: Cool.
• 22:00
  SPEAKER_00: The interesting thing about this one is I remember when we when uh when I worked on Yahoo maps and people told me like to Google maps and these kind of things, people will become stupid and not be able to read any maps anymore.
• 22:12
  SPEAKER_00: It is a skill set we don't need anymore.
• 22:14
  SPEAKER_00: That's true, because the maps are all available now.
• 22:16
  SPEAKER_00: Sometimes in your car they're like six years outdated, but that's a different story.
• 22:21
  SPEAKER_00: Do you think the the I I like the the worries and dangers that you have out there, but is there a part of like a good opportunity to find knowledge we don't need anymore by automating it away?
• 22:34
  SPEAKER_01: Sure.
• 22:34
  SPEAKER_01: Well, and I think like once again, our our software engineering roles aren't necessarily going away, but they're changing.
• 22:40
  SPEAKER_01: And that's where evolve over time, but don't necessarily say like, okay, I don't need AI at all then.
• 22:46
  SPEAKER_01: I'm just gonna do it myself.
• 22:47
  SPEAKER_01: I'm not saying that people should be doing that, but I do think there's opportunities to say, okay, how can I learn the next thing and still be able to evolve that knowledge over time with people so that way you're kind of all growing and kind of paying it forward as you build, and that is something that I think will evolve over time.
• 23:03
  SPEAKER_01: Where yes, we're probably not going to be writing all of our code by hand anymore.
• 23:06
  SPEAKER_01: That's probably just not the future that we're looking at.
• 23:09
  SPEAKER_01: But that doesn't mean that understanding what a system is doing is a completely lost skill.
• 23:15
  SPEAKER_00: One thing that annoys me is that what you also said, that the divide between the haves and the have nots within the AI world are now showing.
• 23:23
  SPEAKER_00: In the beginning, it was like all VC funded, everything was free and everything was great forever, and then out of a sudden, yeah, well, the token costs that much now, and token costs that much now.
• 23:31
  SPEAKER_00: Yeah.
• 23:32
  SPEAKER_00: I see a bit of a counter movement with people starting to do local hosting of models on their machines if the machine is fast enough.
• 23:40
  SPEAKER_00: But where do we go with this?
• 23:42
  SPEAKER_00: I think it's a real problem right now that out of a sudden being a developer costs money again.
• 23:47
  SPEAKER_00: And with open source, we made it for 20 years we said like let's make it free, and just your skill is what gets you into it.
• 23:54
  SPEAKER_01: Yeah, I'm I'm hopeful that this is a pendulum swing where we things were expensive, they got less expensive, things are getting more expensive now, and will hopefully come back with better computing, with cheaper computing, and hopefully just with more optimizations and again sustainable tooling on how tokens are made and used and stuff.
• 24:14
  SPEAKER_01: It feels like a new type of currency almost at this point.
• 24:17
  SPEAKER_00: The uh the thing with the rogue agent uh putting a hit piece out on somebody.
• 24:22
  SPEAKER_00: I just I do this fake on news thing in our live show where I read headlines and people have to tell me if I made them up or if they're real.
• 24:28
  SPEAKER_00: And that was one of them where like, okay, that can't be real, but it really was.
• 24:32
  SPEAKER_00: Do you think there is a worry that people, especially with OpenClaw, it gave so much power to the machine that people don't even understand basic security anymore and give it up for convenience?
• 24:44
  SPEAKER_01: Yeah, and I'm very nervous about that.
• 24:46
  SPEAKER_01: There was actually uh an announcement from Cloudflare that came out this summer saying that bot traffic has officially surpassed human traffic on the internet, which is mind-blowing.
• 24:56
  SPEAKER_01: And uh so thinking about all of the bots that are necess they're sometimes following instructions, sometimes going rogue.
• 25:04
  SPEAKER_01: I I'm sure there's plenty of people in this audience where their parents have said, look at this thing, and you have to say, guys, that's not real, and and explain to them that uh that something is AI.
• 25:14
  SPEAKER_01: I'm glad that there's tools like synth ID, for example, for identifying like is this AI generated or not?
• 25:19
  SPEAKER_01: There's communities online for that.
• 25:21
  SPEAKER_01: And I'm hoping that again more regulation comes and more guardrails come so that we don't have as many bad actors doing that.
• 25:28
  SPEAKER_01: But this is kind of our reality now where the deep fakes are happening, the false reports are going on, the rogue agents are committing crimes, and we have to figure out how we can stop that before I don't know, something even worse happens.
• 25:44
  SPEAKER_00: I mean, of course you can regulate the uh the frontier models, but then you you just download a local one and turn off the guardrails and you can still do the thing.
• 25:51
  SPEAKER_00: In Europe now we have that as a law that every if an image is AI generated, it has to have an AI label, especially in news and these kind of things.
• 26:00
  SPEAKER_00: Yeah, let's not get into the discussion if that's a if that's a free freedom of speech thing, but I find it like very obvious how much slop is out there that is just being generated, and something would be good to flag it up.
• 26:12
  SPEAKER_00: Um that student thing was really interesting as well, which is like we spent years and years teaching people performance, and now they're just like generating tons of code because they don't even look into it.
• 26:23
  SPEAKER_00: Do you think tooling companies that actually build systems like that should be beholden to tell people the outcome of what they create more?
• 26:33
  SPEAKER_01: Yes.
• 26:34
  SPEAKER_01: And at the same time, I do think this is an education thing where that's where talking to teams and and having people review code.
• 26:40
  SPEAKER_01: Like, yes, more and more agents are going to be reviewing code, more and more agents are going to be writing code.
• 26:46
  SPEAKER_01: But having some layer of learning that efficiency is an important thing.
• 26:50
  SPEAKER_01: And currently in our current state, I think that comes from talking to other engineers.
• 26:56
  SPEAKER_01: I don't know what it's going to be in three months because everything is changing so fast.
• 26:59
  SPEAKER_01: But I think that that's something where the responsibility is currently on the teams that are actually producing the code.
• 27:05
  SPEAKER_00: Cool.
• 27:05
  SPEAKER_00: Well, you're going to be available on the booth for more questions.
• 27:08
  SPEAKER_01: Yeah, I'll be at the GitHub booth.
• 27:10
  SPEAKER_00: And then we see you at the after party tonight.
• 27:12
  SPEAKER_01: Great.
• 27:13
  SPEAKER_00: See you then.
• 27:13
  SPEAKER_00: Thanks very much, Cassie.
• 27:16
  SPEAKER_00: Cassidy Williams, our brains in the AI era.