Cloud Providers are changing RAPIDLY [REUPLOAD]

MMaximilian Schwarzmüller
Computing/SoftwareSmall Business/StartupsInternet Technology

Transcript

00:00:00cloud providers like versell cloudflare but even aws to some extent are changing rapidly at least
00:00:08some of them rapidly and i think that's good for us developers it's also worth a closer look now
00:00:16cloud providers are changing because we are changing as developers because of course what's
00:00:22changing right now is what we're building and also how we're building both things are changing right
00:00:30now what we're building is of course to some degree what we built before software web apps and so on so
00:00:40that of course to some degree has not changed but we're building more because ai agents make it easier
00:00:49to build stuff and let's be honest there's a lot of stuff a lot of applications and demos in there
00:00:55that will or at least should never see the light of day but we're building more because ai agents make
00:01:02it easier wipe coding of course is a thing maybe not the thing you want to use primarily for serious
00:01:09projects for serious software which you plan on selling or distributing but of course wipe coding
00:01:16can be absolutely fine for your own little internal tools some throwaway software that solves specific
00:01:24problems where you don't care about covering all niche cases or about security or anything like that
00:01:29so we got that and therefore we are of course building more and then of course we are building ai
00:01:36agents and agentic apps whatever that means but in the end it means applications or
00:01:45tools workflows whatever you want to call it that wrap large language models that wrap ai and then do
00:01:52something potentially agentic where you potentially are building something where an ai can call tools
00:01:58dispatch tasks work with a file system and so on so that's the what part that is changing but of course
00:02:05the how is also changing because we are or at least many of us i guess no longer are writing the code by
00:02:14hand we're not doing that anymore at least not 100 of the code instead we are using ai agents like
00:02:23clod or codex or whatever you are using and of course we're letting those agents do more and more so
00:02:33you may use them to write code but maybe you also let them run all kinds of cli commands and tools on
00:02:40your system starting with git for committing and pushing which is relatively trivial and easy all the
00:02:47way up to deployment maybe you may not just be using those agents for generating the code you may also be
00:02:54using them using them to generate docker images or to use cli tools that push the code that push the
00:02:59applications to some hosting provider and that takes me back to the cloud and how cloud providers are
00:03:07changing versell itself rebranded itself to be all about agentic infrastructure if you visit versell.com
00:03:17that is the first thing you read right now and on the right you see four coding agents to ship apps and
00:03:24agents automated by agents so here we got the word agents in that box three times and they are clearly
00:03:33talking about being the platform for anything built by agents including agents themselves now it's pretty
00:03:41similar for cloudflare if i visit cloudflare.com right now i see this message where i learned that
00:03:48cloudflare is one platform for my apps agents and workforce so definitely a bit less agentic first or ai first
00:03:58but we got the word agents right here in the main messaging as well and we'll also see it when we take
00:04:05a look at the services offered by these cloud providers that are changing quickly now netlify is
00:04:13also mentioning ai but not the word agent here they are still mostly about serving your apps your web apps
00:04:22that may have been built with ai but not so much about running ai agents in the cloud and i'll get back
00:04:29to why that matters and if we take a look at aws as a last example here i also right now when i visit this
00:04:35have the word asian in the in the main message i see that i can move at the speed of agents now what is
00:04:42important here or what's interesting and what is of course related to what we're building and how we're
00:04:48building is that all these cloud providers i would say nowadays have three main pillars the the first
00:04:57pillar is about apps and that is of course what they were always about now a little side note of course
00:05:03aws is a bit hard to compare to verse sell and so on because it's more complex it's more about having
00:05:10tons of primitives and managed services it's about all those enterprise customers not to say that
00:05:16verse sell doesn't have enterprise customers but aws is all about giving you all the building blocks you
00:05:22could potentially ever want to build any kind of system in the cloud and therefore they were never
00:05:29primarily or only about running web apps or other pieces of software in the cloud they were about
00:05:36giving you the tools you need to do anything in the cloud to store any kind of data there run any kind
00:05:43of workflow versell netlify and also cloudflare of course were more about hey you got a web application
00:05:53run it here not just about that of course but that was definitely more of their focus so i'll primarily
00:06:00talk about apps here and when i mean apps i mean of course web apps websites whatever you want to call
00:06:05it so that is what we use these cloud providers for and that of course has not really changed you can and
00:06:11you will still use them for that and it doesn't really matter how you build those apps right if you use
00:06:17cloud code or codex the thing that maybe is a bit easier now and where i actually like ai a lot is
00:06:24that deployment gets easier because you can point cloud code or whatever at the relevant documentation
00:06:32pages you can install the clis you can install the versell cli the cloudflare wrangler tool and so on
00:06:39and if you got that installed if you give the agent some some hints at where to search for for deployment
00:06:45details and instructions it can do the deployment for you or at least it can prepare it for you it can
00:06:50write configuration files that all gets a bit simpler because let's be honest deployment can be a bit of
00:06:57a pain if it gets a bit more complex and ai can definitely help a bit there now just as with wipe coding
00:07:03obviously you still want to understand what's going on at least i want to understand what's going on but
00:07:09it can't help there so that pillar of course still exists now the second pillar which we now have
00:07:16however which is new is agents or ai agents that you run those in the cloud because of course
00:07:24we're at this point in time where we're for one still trying to figure out how exactly we'll work
00:07:30with ai as developers but we're also at this point where we're all or where all the companies are about
00:07:37building ai agents and it's sometimes hard to grasp what exactly you want to build and what exactly
00:07:44those agents should be doing for you but at least from the perspective of a ceo you might have an a
00:07:51marketing agent you might have an agent that does customer support a finance agent that helps out in
00:07:59your finance department and so on and as a developer you may have an agent that tackles open issues on
00:08:06your github repository you may have another agent that does research and these agents need to run
00:08:13somewhere now obviously they can run on your system that will be a bit harder if you're working in some
00:08:18bigger company where you don't have unlimited privileges and can't install everything you want
00:08:23on your system but even as a small developer running everything on your system has of course some
00:08:29security implications if you have an ai agent roam around on your system it can potentially do damage
00:08:35accidentally or because of prompt injections and so on so you may want an environment where it can
00:08:42safely operate or where it can't do as much damage now there are loads of different solutions out there
00:08:49you could look into tooling like just bash which is interestingly enough also by versell by the way
00:08:56they are releasing many tools uh that aim to make development with ai easier and just bash essentially
00:09:03is a sandboxing solution which you can run on your system to run an agent inside of a virtual bash with
00:09:10limited privilege and so on so you could use tools like that but of course you could also run those agents
00:09:16in the cloud or you could have a combination and have an agent that runs locally offload certain tasks
00:09:23to the cloud so the cloudflare sandbox sdk for example aims to simplify the process of running tasks
00:09:31agentic tasks with an ai agent doing its thing in the cloud where you can then potentially expose r2
00:09:39cloudflare r2 which is their s3 equivalent as a file system and so on so that you have a throwaway
00:09:46environment where an agent can do whatever you want to do if it breaks stuff it doesn't matter because you
00:09:51can just restart the sandbox and you don't expose your entire system and that of course therefore is a brand
00:09:57new category of things you want to deploy to the cloud or you want to run in the cloud and that is definitely
00:10:04where versell and sandbox are moving towards they want to not just make it easy for you and your
00:10:13agents to push web apps and apps to the cloud that's also something they want to do but not just but
00:10:19instead they also want to make it easy for you to run ai agents you may have built in the cloud and then
00:10:26there is a third pillar i think and that's the ai gateway pillar or however you want to call it you
00:10:34know we got all these ai providers we got proprietary providers obviously like open ai anthropic google and
00:10:42then we got some open source models which may be open source or open weight but running them on your
00:10:49system can be tricky depending on the size now i do have various courses obviously including on chat gpt
00:10:56the advanced guide cloud code but i also have a course about running local large language models because
00:11:03i'm a big fan of those open weight models where you can really just run them with olama or lm studio or
00:11:09other tools on your system but of course for most systems at least you are limited to the smaller
00:11:16models which can still be helpful as i show in this course but if we're talking about bigger models you
00:11:21will still likely want some provider that hosts them for you so that you can access them through the cloud
00:11:30and no matter if we're talking about those open models which you can't or don't want to run on your
00:11:35system or if we're talking about many of the proprietary models these providers like cloudflare but also
00:11:41versell and also aws they offer services that could be called ai gateway in the case of cloudflare or for
00:11:50aws it's amazon badrock but these are simply services that are all about accessing ai through one uniform api
00:12:00where you can basically just switch one value and switch to a different provider where billing is done
00:12:06through cloudflare or aws and where it's therefore easier to access a broad variety of ai models which
00:12:13are easy to monitor easy to switch where you don't need to create tons of accounts with open ai with
00:12:19anthropic with google with whatever other provider you want think of it as open router essentially open router of
00:12:27course is a service that makes it easy to switch between different models and cloudflare versell aws
00:12:33they essentially have their own open router like solution you could say there are some differences
00:12:40because open router is really just a router which is about funneling your request to different providers
00:12:47whereas cloudflare and aws for example they also provide hosting for these models so with amazon badrock
00:12:56for example you can access anthropic models that are hosted on aws infrastructure where as a customer
00:13:04you can control in which region it should be hosted where you have greater uh control over where your data
00:13:11gets stored or who gets access to your data and that of course can matter to enterprise customers so it's
00:13:17actually a combination of things here it's convenience obviously because it's easier to switch between
00:13:23different models you can even set up routing rules where different requests may get different models
00:13:29automatically you have a unified billing experience but besides the convenience it's also about more
00:13:36more control over where the model is hosted and so on at least for some of these providers
00:13:41and i think therefore these are the main pillars we see right now for these major cloud providers and i
00:13:50find it quite interesting that for example netlify is still mostly about that first pillar here it's
00:13:58their decision their strategy whereas cloudflare and bursell and aws and many others by the way are of
00:14:05course branching out and are trying to cover all three pillars now i think as a developer no matter if
00:14:12you're building ai agents or not or and no matter if you're working with ai or not these are great
00:14:19developments because one nice side effect is that these cloud providers in their effort to make it easier for
00:14:27ai agents to use them are giving us more ways of accessing their services they are improving their
00:14:36cli tools their documentation to provide all that information to agents which spoiler you as a human
00:14:43can also read so that's pretty nice and they're building more useful primitives i mean cloudflare
00:14:49alone and just to be super clear this is not some paid advertisement or anything like that these are
00:14:55really just services i'm using but cloudflare alone they released so many new nice services like dynamic
00:15:04workers where one worker which is like a serverless solution provided by them where you can execute code
00:15:10in the cloud where one worker can spin up more workers to offload certain tasks no matter if you're
00:15:16you're building ai agents or not this can be a useful feature if you're building any kind of service or
00:15:22tooling where your customers can maybe share some data or code with you or where you want to execute some code
00:15:30that involves customer data or customer code it's nice to have primitives where you can securely do so
00:15:37in throwaway environments and therefore these new capabilities are pretty nice to have even if you
00:15:44don't care about ai at all even if you're not building ai agents and that's why i'm pretty excited about
00:15:50these changes in cloud providers and why i think it's more interesting than ever to play around with different
00:15:57providers if you've used netlify forever maybe give vercel or cloudflare a try and dive back into their docs
00:16:05maybe together with ai to understand what they offer and how you can build on top of them
00:16:11anyways that's just something i'm observing something i'm pretty excited about and as always i'm interested
00:16:16in hearing your perspective on that and what i maybe missed or what you think about this transformation
00:16:22of cloud providers and where we're heading to.

Key Takeaway

Major cloud providers are evolving into AI-first platforms by offering three core pillars: standard web app hosting, secure sandbox environments for AI agents, and unified AI gateways for model management.

Highlights

  • Cloud providers like Vercel, Cloudflare, and AWS are shifting their core focus to support agentic infrastructure and AI-native applications.

  • Developers now frequently use AI agents to automate tasks ranging from generating code to running CLI commands and managing deployments.

  • Vercel, Cloudflare, and AWS currently offer integrated AI gateway services that provide unified APIs for accessing multiple proprietary and open-source models.

  • Cloudflare's Dynamic Workers enable the creation of serverless environments that can spin up additional workers to execute tasks in isolated, secure sandboxes.

  • Cloud providers are improving their documentation and CLI tools specifically to make their infrastructure more accessible and interoperable with AI agents.

Timeline

Drivers of Cloud Infrastructure Evolution

  • Software development processes are changing due to the increased use of AI agents for coding and task automation.
  • Cloud providers are rebranding to highlight agentic infrastructure in response to these shifts in developer behavior.
  • Vercel, Cloudflare, and AWS now feature “agents” prominently in their primary marketing and messaging.

Developers are shifting from manual coding to using AI agents for generating applications, managing file systems, and handling deployment workflows. Cloud providers are responding to this trend by updating their core platforms to better support agent-driven infrastructure, moving beyond simple application hosting.

Three Pillars of Modern Cloud Services

  • The first pillar is traditional web application hosting, which remains a primary service.
  • The second pillar involves secure, throwaway cloud environments for running AI agents to avoid local system risks.
  • The third pillar is the AI gateway, providing a unified API for accessing various AI models while handling billing and routing.

Cloud providers have expanded from hosting web applications to offering specialized services for AI. This includes secure sandbox environments, such as Cloudflare's SDK, which allow agents to operate without compromising the developer's local machine. Additionally, AI gateways act as a central interface for managing different AI models, simplifying the developer experience by consolidating model access and billing.

Benefits of the AI-Cloud Shift for All Developers

  • Improved CLI tools and documentation designed for AI agents are also more accessible and useful for human developers.
  • New primitives like Cloudflare's Dynamic Workers provide better security for executing untrusted code or handling customer data.
  • These developments offer value even to developers who do not actively build AI agents.

The push to make cloud infrastructure machine-readable results in better documentation and more powerful, flexible tools for human developers. Features like dynamic serverless execution allow for more robust and secure application architectures. These advancements demonstrate that current platform updates are broadly applicable to modern software engineering.

Community Posts

View all posts