WEBVTT

00:00:00.060 --> 00:00:07.900
Speaker 0: Cloud providers like Vercel, Cloudflare but even AWS to some extent are changing rapidly.

00:00:08.060 --> 00:00:13.220
Speaker 0: At least some of them rapidly and I think that's good for us developers.

00:00:13.300 --> 00:00:15.720
Speaker 0: it's also worth a closer look now.

00:00:16.239 --> 00:00:21.040
Speaker 0: cloud providers are changing because we Are Changing as Developers.

00:00:21.220 --> 00:00:26.940
Speaker 0: Because Of course what's changing right Now is What We're Building And Also How?

00:00:28.620 --> 00:00:30.760
Speaker 0: things are changing right now.

00:00:31.200 --> 00:00:49.880
Speaker 0: What we're building is of course to some degree what More because AI agents make it easier to build stuff.

00:00:49.960 --> 00:00:59.400
Speaker 0: And let's be honest, there is a lot of applications and demos in there that will or at least should never see the light of day.

00:00:59.580 --> 00:01:00.820
Speaker 0: but we're building more.

00:01:04.640 --> 00:01:14.680
Speaker 0: is I think maybe not the thing you want to use primarily for serious projects, or a serious software which you plan on selling and distributing.

00:01:14.940 --> 00:01:17.740
Speaker 0: But of course Wipecoding can be absolutely fine!

00:01:18.140 --> 00:01:28.920
Speaker 0: For your own little internal tools some throw away software that solves specific problems, where you don't care about covering all niche cases or security.

00:01:30.120 --> 00:01:33.040
Speaker 0: So we got it and therefore are building more!

00:01:33.580 --> 00:01:41.160
Speaker 0: And then of course we're building AI agents and agentic apps whatever that means.

00:01:41.240 --> 00:01:55.920
Speaker 0: but in the end applications... Tools workflows, whatever you want to call it that a wrap large language models that rap AI and then do something potentially agenting where you Potentially are building.

00:01:56.000 --> 00:02:01.740
Speaker 0: Something wherein I can call tools dispatch tasks work with the file system And so on.

00:02:01.940 --> 00:02:03.540
Speaker 0: So that's the what part?

00:02:03.800 --> 00:02:14.760
Speaker 0: That is changing but of course The how is also changing because we are or at least many of us i guess no longer our writing the code by hand.

00:02:15.220 --> 00:02:20.440
Speaker 0: We're not doing that anymore, at least not one hundred percent of the code.

00:02:20.620 --> 00:02:32.540
Speaker 0: instead we are using AI agents like Claude or Codex or whatever you are using and Of course were letting those agents do more.

00:02:33.260 --> 00:02:47.120
Speaker 0: So you may use them to write code, but maybe also let them run all kinds of CLI commands and tools on your system starting with git for committing and pushing which is relatively trivial or easy.

00:02:47.720 --> 00:02:49.260
Speaker 0: All the way upto deployment!

00:02:49.880 --> 00:03:02.360
Speaker 0: You might not just be using those agents for generating a code, you may also generate docker images that push the applications to some hosting provider.

00:03:03.040 --> 00:03:37.500
Speaker 0: And that takes me back three times and they are clearly talking about being the platform for anything built.

00:03:37.980 --> 00:03:40.360
Speaker 0: by agents, including agents themselves.

00:03:40.540 --> 00:03:43.160
Speaker 0: Now it's pretty similar for Cloudflare!

00:03:43.220 --> 00:03:53.380
Speaker 0: If I visit cloudflare dot com right now i see this message where i learned that CloudflARE is one platform for my apps, agents and workforce.

00:03:53.840 --> 00:04:03.280
Speaker 0: so definitely a bit less agentic first or ai-first but we got the word agents right here in the main messaging as well.

00:04:04.540 --> 00:04:10.340
Speaker 0: We'll also see it when we take a look at the services offered by these cloud providers that they're changing quickly.

00:04:10.840 --> 00:04:17.480
Speaker 0: Now, Netlify is also mentioning AI but not the word agent here?

00:04:17.620 --> 00:04:30.720
Speaker 0: They are still mostly about serving your apps or web apps that may have been built with AI running AI agents in the cloud and I'll get back to why that matters.

00:04:31.140 --> 00:04:41.300
Speaker 0: And if we take a look at AWS as an last example here, also right now when i visit this half-the-word agent In the main message I see That can move it's speed of agents.

00:04:41.400 --> 00:04:56.220
Speaker 0: Now what is important or interesting related To how we're building Is all these cloud providers nowadays have three main pillars.

00:04:56.860 --> 00:05:01.900
Speaker 0: The the first pillar is about apps and that's of course what they were always about.

00:05:02.000 --> 00:05:07.680
Speaker 0: now a little side note, of course aws Is a bit hard to compare to versel?

00:05:07.820 --> 00:05:09.660
Speaker 0: And so on because it's more complex.

00:05:09.760 --> 00:05:13.740
Speaker 0: It's more about having tons of primitives and managed services.

00:05:13.820 --> 00:05:16.180
Speaker 0: It's about all those enterprise customers.

00:05:16.300 --> 00:05:22.520
Speaker 0: not to say that versel doesn't have enterprise customers but AWS is all about giving you all the building blocks.

00:05:22.660 --> 00:05:44.600
Speaker 0: You could potentially ever want to build any kind of system in the cloud and therefore they were never primarily or only about running web apps, other pieces of software in a cloud... They were about giving in the cloud to store any kind of data there, run any workflow.

00:05:45.100 --> 00:05:53.000
Speaker 0: Vercel, Netlify and also Cloudflare were more about hey you got a web application?

00:05:53.200 --> 00:05:53.700
Speaker 0: Run it here!

00:05:54.020 --> 00:05:58.460
Speaker 0: Not just that but definitely their focus.

00:05:58.960 --> 00:06:05.660
Speaker 0: So I'll primarily talk apps here and when I mean apps, I mean of course web apps websites whatever you want to call it.

00:06:05.720 --> 00:06:11.020
Speaker 0: So that is what we use these cloud providers for And that of course has not really changed.

00:06:11.060 --> 00:06:13.120
Speaker 0: You can't will still used them for debt?

00:06:13.600 --> 00:06:16.600
Speaker 0: It doesn't matter how you build those apps right.

00:06:16.740 --> 00:06:18.780
Speaker 0: if you use clod code or codex.

00:06:19.340 --> 00:06:34.540
Speaker 0: the thing That maybe a bit easier now where i actually like AI alot Is deployment gets easier because you can point clodcode At the relevant documentation pages, you can install CLIs.

00:06:34.580 --> 00:06:39.120
Speaker 0: You can install the Vercel CLI, The Cloudflare Wrangler tool and so on.

00:06:39.420 --> 00:06:50.340
Speaker 0: And if got that installed If give the agent some hints at where to search for deployment details or instructions it could do the deployment for you Or at least it can prepare it For you.

00:06:50.440 --> 00:06:52.320
Speaker 0: It can write configuration files.

00:06:52.760 --> 00:07:01.780
Speaker 0: That all gets a bit simpler Because let's be honest Deployment Can Be A Bit Of Pain if it gets a bit more complex and AI can definitely help a bit there.

00:07:02.060 --> 00:07:03.600
Speaker 0: Now, justice with wipe coding.

00:07:03.740 --> 00:07:06.960
Speaker 0: obviously you still want to understand what's going on.

00:07:07.100 --> 00:07:10.020
Speaker 0: at least I want to Understand What's Going On but It Can't Help There.

00:07:10.220 --> 00:07:14.420
Speaker 0: so that pillar of course Still Exists now.

00:07:14.600 --> 00:07:22.440
Speaker 0: the second pillar which we know have however Which is new Is Agents or AI Agents That You Run Those in The Cloud.

00:07:22.540 --> 00:07:38.880
Speaker 0: because Of Course We're At this point in time, where we're for one still trying to figure out how exactly will work with AI as developers but... where all the companies are about building AI agents.

00:07:39.140 --> 00:07:56.540
Speaker 0: And it's sometimes hard to grasp what exactly you want to build and what those agents should be doing for you, but at least from a CEO perspective… You might have an agent that does customer support!

00:07:57.020 --> 00:08:07.500
Speaker 0: A finance agent who helps out in your finance department... And as a developer, you may have an agent that tackles open issues on your github repository.

00:08:07.960 --> 00:08:13.920
Speaker 0: You These agents need to run somewhere.

00:08:14.040 --> 00:08:15.940
Speaker 0: Now, obviously they can run on your system.

00:08:16.440 --> 00:08:30.260
Speaker 0: that will be a bit harder if you're working in some bigger company where You don't have unlimited privileges and can't install everything you want On your system but even as small developer running Everything on your System has of course Some security implications.

00:08:30.360 --> 00:08:38.980
Speaker 0: If you Have an AI agent roam around on Your system it Can potentially do damage accidentally or because of prompt injections and so on.

00:08:39.140 --> 00:08:46.580
Speaker 0: So you may want an environment where it can safely operate or where it can't do as much damage.

00:08:46.940 --> 00:08:49.700
Speaker 0: Now, there are loads of different solutions out there.

00:08:50.040 --> 00:08:56.020
Speaker 0: you could look into tooling like Just Bash which is interestingly enough also by Versel.

00:08:56.160 --> 00:09:06.600
Speaker 0: By the way they are releasing many tools that aim to make development with AI easier and just bash essentially Is a sandboxing solution Which You Can Run on Your System.

00:09:06.680 --> 00:09:23.800
Speaker 0: To run an agent inside Of A Virtual Bash With Limited Privilege And So On could use tools like that, but of course you can also run those agents in the cloud or have a combination and have an agent who runs locally offload certain tasks to the cloud.

00:09:24.240 --> 00:09:42.080
Speaker 0: So the Cloudflare Sandbox SDK for example aims an AI agent doing its thing in the cloud, where you can then potentially expose R two Cloudflare or to which is there as three equivalent.

00:09:42.540 --> 00:09:49.060
Speaker 0: As a file system and so on So that you have a throwaway environment Where an agent can do whatever we want to do.

00:09:49.280 --> 00:09:55.180
Speaker 0: if it breaks stuff It doesn't matter because you can just restart the sandbox And you don't expose your entire system?

00:09:55.860 --> 00:10:03.100
Speaker 0: In that of course they offer us a brand new category of things you want to deploy, or run in the cloud.

00:10:03.180 --> 00:10:08.680
Speaker 0: And that is definitely where Vercel and Sandbox are moving towards!

00:10:08.880 --> 00:10:16.800
Speaker 0: They want not just make it easy for you and your agents to push web apps & apps into the Cloud?

00:10:17.100 --> 00:10:25.220
Speaker 0: That's also something they wanna do but instead they also wanna make it easier for you to Run AI Agents You May Have Built In The Cloud.

00:10:26.000 --> 00:10:33.960
Speaker 0: And then there is a third pillar, I think and that's the AI gateway pillar or however you want to call it.

00:10:34.200 --> 00:10:37.040
Speaker 0: You know we got all these AI providers.

00:10:37.100 --> 00:10:52.180
Speaker 0: We've got proprietary Providers obviously like OpenAI and Thropic Google... ...and Then we got some open source models which may be open-source or open weight But running them on your system can be tricky, depending on the size.

00:10:52.440 --> 00:11:11.420
Speaker 0: Now I do have various courses obviously including on ChatGPT, The Advanced Guide, Cloud Code but i also have a course about running local large language models because i'm a big fan of those open weight models where you can really just run them with olama or lmstudio or other tools On Your System!

00:11:13.140 --> 00:11:18.780
Speaker 0: for most systems at least you are limited to the smaller models, which can still be helpful as I show in this course.

00:11:19.580 --> 00:11:30.300
Speaker 0: But if we're talking about bigger provider that hosts them for you, so that you can access them through the cloud.

00:11:30.680 --> 00:11:47.940
Speaker 0: And no matter if we're talking about those open models which you can't or don't want to run on your system Or If We're Talking About Many Of The Proprietary Models These Providers Like Cloudflare but also Vercel and AWS They Offer Services That Could Be Called AI Gateway.

00:11:48.320 --> 00:12:15.660
Speaker 0: in the case of cloudflare or for aws it's amazon bedrock but these are simply services that all about accessing a i through one uniform api where you can basically just switch to different provider Cloudflare or AWS and where it's therefore easier to access a broad variety of AI models which are easy to monitor, easy to switch.

00:12:15.940 --> 00:12:25.560
Speaker 0: Where you don't need to create tons of accounts with OpenAI, with Anthropic ,with Google, whatever other provider you want... Think about it as open router.

00:12:25.780 --> 00:12:26.260
Speaker 0: essentially!

00:12:26.660 --> 00:12:33.520
Speaker 0: Open Router is the service that makes it easily to switch between different models And cloudflare vs aws.

00:12:34.040 --> 00:12:38.800
Speaker 0: they have their own open router-like solution, you could say.

00:12:39.060 --> 00:12:50.760
Speaker 0: There are some differences because OpenRouter is really just a router which is about funneling your request to different providers Whereas Cloudflare and AWS, for example.

00:12:51.100 --> 00:12:54.260
Speaker 0: They also provide hosting for these models.

00:12:54.360 --> 00:13:13.660
Speaker 0: so with Amazon Badrock For example you can access enthropic models that are hosted on AWS infrastructure whereas a customer You can control in which region it should be hosted where you have greater Control over Where your data gets stored or who gets access to your data?

00:13:13.740 --> 00:13:16.840
Speaker 0: And then of course Can matter to enterprise customers.

00:13:16.900 --> 00:13:19.360
Speaker 0: So it's actually a combination of things here.

00:13:19.740 --> 00:13:30.100
Speaker 0: It's convenience obviously because its easier to switch between different models, you can even set up routing rules where different requests may get different models automatically.

00:13:30.460 --> 00:13:32.920
Speaker 0: You have unified billing experience.

00:13:33.220 --> 00:13:41.160
Speaker 0: but besides the convenience Its also about more control over Where are the models hosted and so on at least for some of these providers.

00:13:41.940 --> 00:13:50.400
Speaker 0: And I think therefore these are the main pillars we see right now for these major cloud providers.

00:13:50.460 --> 00:13:57.380
Speaker 0: and i find it quite interesting that, for example Netlify is still mostly about that first pillar here.

00:13:57.480 --> 00:14:09.060
Speaker 0: It's their decision-their strategy whereas Cloudflare and Vercel and AWS many others by the way Are of course branching out in R trying to cover all three pillars.

00:14:09.960 --> 00:14:16.300
Speaker 0: Now, I think as a developer no matter if you're building AI agents or not and no matter If your working with

00:14:16.440 --> 00:14:16.780
Speaker 0: A.I.,

00:14:18.260 --> 00:14:37.560
Speaker 0: these are great developments because one nice side effect is that These cloud providers in their effort to make it easier for AI agents To use them giving us more ways of accessing their services, they are improving their CLI tools.

00:14:37.640 --> 00:14:44.300
Speaker 0: Their documentation to provide all that information to agents which spoiler you as a human can also read.

00:14:44.560 --> 00:14:45.640
Speaker 0: so that's pretty nice.

00:14:46.120 --> 00:14:48.760
Speaker 0: and They're building more useful primitives.

00:14:48.880 --> 00:14:50.280
Speaker 0: I mean cloudflare alone.

00:14:50.820 --> 00:14:54.700
Speaker 0: Just to be super clear, this is not some paid advertisement or anything like that.

00:14:54.980 --> 00:14:58.340
Speaker 0: These are really just services I'm using but Cloudflare alone.

00:14:58.800 --> 00:15:10.920
Speaker 0: they released so many new nice services Like dynamic workers where one worker which Is a serverless solution provided by them Where you can execute code in the cloud?

00:15:11.220 --> 00:15:15.420
Speaker 0: Where OneWorker Can spin up more workers To offload certain tasks.

00:15:15.800 --> 00:15:32.020
Speaker 0: No matter if You're building AI agents Or Not This Can Be A Useful Feature If you're building any kind of service or tooling where your customers can maybe share some data or code with you, where you want to execute some code that involves customer data.

00:15:34.260 --> 00:15:45.900
Speaker 0: It's nice to have primitives so you can securely do it in throwaway environments and therefore these new capabilities are pretty nice even if you don't care about AI at all!

00:15:48.320 --> 00:15:58.680
Speaker 0: That is why I'm excited about the changes in cloud providers More interesting than ever to play around with different providers.

00:15:58.920 --> 00:16:10.420
Speaker 0: if you've used netlify forever Maybe give we're sell or cloudflare a try and dive back into their docs maybe together With AI to understand what they offer in how you can build on top of them.

00:16:11.420 --> 00:16:15.380
Speaker 0: Anyways That's just something I'm observing, something i'm pretty excited about.

00:16:15.440 --> 00:16:25.520
Speaker 0: And as always...I am interested in hearing your perspective on that and what maybe missed or you think of this transformation of cloud providers?