Schematical

GitHub Copilot is coming for your data.

GitHub, which used to be the knight in shining armor leading the open source movement, now wants to use all your code to train their AI agents to someday take your job.



It all starts on April 24th, 2026.



In their own words:



From April 24 onward, interaction data—specifically inputs, outputs, code snippets, and associated context—from Copilot Free, Pro, and Pro+ users will be used to train and improve our AI models unless they opt out.


I feel like I wouldn’t be as disgusted with this if the user had to opt in instead of just having them opt in by default.



I suppose this is part of a bigger trend that has been happening for a while.



Part of me wants to move everything from the cloud to hardware I self-host out in a cabin in the woods while forging my own food and growing a big beard like Ron Swanson, but that doesn’t seem practical either.



How do you feel about these big companies using your hard work to train their models?

12 hours ago | [YT] | 0

Schematical

If you didn't catch our post yesterday, you're in luck. On today's CTO Coffee Hour episode, Matt & Dom dive into AWS Agent Registry that is now in preview.

1 day ago | [YT] | 0

Schematical

AWS Agent Registry is now in preview

Is your organization jumping on the Agentic AI bandwagon?



If so, chances are your various teams could be creating redundant or overlapping tools.



AWS wants to solve that with AWS Agent Registry.



Agent Registry stores metadata on every MPC Server, Tool Call, Skills, and even Agents.



This way, your various agents can query the registry and see what tools to call via MCP or what agents to collaborate with via A2A.



This all sounds like a great way to burn cash on tokens or the start of Skynet.



But seriously, while I am getting a bit of “Agentic” fatigue from people sticking chatbots into everything, I do think the tech is here to stay. The question is, what is the appropriate use case for it?



It’s no different than how in the late 2000s everyone was creating mobile apps for every use case when a simple website would do just fine.

2 days ago | [YT] | 0

Schematical

How do you determine legitimate bot traffic from malicious bot traffic?

After analysing literally billions of requests at this point via various tools like Cloud Watch Insights, I have found a convenient way to determine good traffic from bad.



Legitimate crawlers like Meta’s link checkers put the link to their documentation right in the User agent:



meta-externalagent/1.1 (+developers.facebook.com/docs/sharing/webmasters/cr…)


Google, OpenAI, and Amazon all do the same thing:



Mozilla/5.0 (compatible; Googlebot/2.1; +www.google.com/bot.html)
Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko; compatible; GPTBot/1.3; +openai.com/gptbot)
Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko; compatible; Amazonbot/0.1; +developer.amazon.com/support/amazonbot) Chrome/119.0.6045.214 Safari/537.36


Heck, Anthropic even gives you an email address you can contact:



Mozilla/5.0 AppleWebKit/537.36 (KHTML, like Gecko; compatible; ClaudeBot/1.0; +claudebot@anthropic.com)


Is it computationally effective to spam out the same link over and over again in your user agent?



Probably not, but it makes it convenient for someone like me to figure out where the traffic is coming from and better decide if it is legitimate or not.



If you are getting spammed, try a thorough inspection of the User Agent to see if you can get a link straight to the source.



If you have a crawling service, consider putting a link to your docs in your user agent to better communicate to the websites you are crawling a reason not to block you.



I will point out that spoofing user agents is not rocket science, so don’t just trust the user agent. Ideally, these docs would give you a way to verify the user agent is legitimate.



For example, Google gives you a really simple way to verify their bot via DNS.



Question for you:

How are you determining what traffic is malicious or useless bot traffic vs legitimate traffic?

Skip the Carousel

1 2

5 days ago | [YT] | 0

Schematical

S3 Express One Zone

Want to speed up access to AWS S3 up to 10x faster while saving up to 80% compared to standard S3?



Then you need to check out S3 Express One Zone.



You might be wondering, “10x faster? 80% cheaper? Surely there must be a catch, right?”



Of course there is. When you set up your bucket, you need to select a specific Availability Zone where the bucket will live. No replication, or any of the multi-AZ/Region replication you see in normal S3 buckets.



This means a massive reduction in redundancy, but for the right use case, that might be a small price to pay.



The best use cases I have seen for this are high-throughput background tasks like training a model.



You can collocate the hardware you train on and the S3 bucket in the same AZ to decrease latency.



If the AZ falls over, no, it can’t fail over to another AZ, but it's a background worker, not your production API server.



You will lose some training time, but your end customers will never see the site flicker.



The “up to” 80% savings is huge, too. From what I can tell, this is because AWS doesn’t need to replicate your data out to other AZs and Regions. They save compute time and disk space by replicating your data, and you save big $$$.



I love that S3 is offering up such a variety of storage options on all sides of the spectrum. On one side, you have AWS S3 Glacier for infrequently accessed data, but highly redundant, and on the other side, you have S3 Express One zone being accessed super frequently with no latency but no redundancy.



People complain about cloud storage costs a lot, but those costs can be avoided if you choose the right tools for the job.



If you need help choosing the right tools for the job, feel free to reach out.

6 days ago | [YT] | 0

Schematical

AWS just dropped ECS Daemons

The concept of Daemons has been in computing for a long time.



AWS just brought it to the world of cloud computing in the form of ECS Daemons.

This feature appears to be specific to Amazon ECS Managed Instances, not Fargate which makes sense once you hear what it does.



One of each of the daemon tasks would get booted up per managed ECS Managed Instance before any of the other tasks get booted up.



From there, the Daemon task can do advanced orchestration and/or specialty monitoring of the other tasks that spin up on that instance.



This new functionality should open up some interesting use cases/infrastructure designs.



Question for you:
What use case do you have for ECS Daemons?

1 week ago | [YT] | 0

Schematical

On today's episode, Matt is mostly riding solo and runs through Axios, JavaScript’s most popular library.

1 week ago | [YT] | 0

Schematical

AWS finally rolls out account/region-specific namespaces for S3 buckets.

I am amazed this feature wasn’t rolled out years ago.



I can’t imagine how many S3 buckets were misconfigured and crawled because of their global namespaces.



Not only was it kind of annoying needing to find an S3 namespace that wasn’t taken for some S3 bucket you never planned to make public, but there were a ton of security implications there.



I have heard of popular projects shutting down their S3 bucket and other people grabbing the bucket's newly released global namespace.



This can be done for malicious purposes, like hoping the users of the old popular project will try to grab some files from the original S3 bucket, only to end up pulling a new malicious payload.



On the other side of that, I have heard of new non-malicious parties grabbing the global namespace, not knowing about the namespace being used for the now-defunct popular product, and then getting spammed into oblivion by people still running the software pointed at that namespace.



Basically, there are a lot of things that can go wrong if the bucket isn’t configured perfectly.



So I am glad they are allowing you to create buckets that are unique to your account.

It only took them 20 years!



If you want to know more about how to secure your entire AWS account, much less your S3 buckets, you should check out my On-Demand Video Course on O'Reilly Zero to Hero on AWS Security: An Animated Guide to Security in the Cloud. Link in comments.

1 week ago | [YT] | 1

Schematical

TechDebt is coming to Steam!

For those of you who play games on PC at all, you will likely be familiar with Steam, which is the biggest marketplace for PC games.



After years of dabbling with publishing a real game on something like ItchIO I decided to take a swing at the big leagues.



Now I aim to keep the scope down. From what I hear, spending 5 years building the game of your dreams is a sure way to shoot yourself in the foot.



Small games iterated on quickly are the smart move.



I still have plenty of content planned for Tech Debt, but this should keep me focused on content related to the core mechanics that already exist and prevent me from expanding the core mechanics into a full-out colony sim.



We just got the Steam page cover art back from the artist I commissioned for the project, and I love it, they did a great job. The next step is to get a trailer cut so we can launch our Steam page.



This is really exciting for me to publish a game on Steam, and hopefully not the last.



With all this said, I had a client pop up with an interesting and urgent project that needed me to go hands-on with, so you won’t hear as much about Tech Debt in April.



The good news is that I will give the video editor I am working with time to get the trailer up and edited. So that is it. Let me know what you think of the cover art. Any feedback is appreciated!

1 week ago | [YT] | 3

Schematical

The US Government seems to think AI won’t replace software engineers. Are they right?

Digging deeper into the BLS Data, it seems like the “Job Outlook” for 2024–34 will be 15%, which is “Much faster than average”.



While I am quite certain that commercially available LLMs that exist in the market today are in no way capable of replacing a solid senior-level engineer, I am less confident in these job growth numbers.



For starters, as I write this in March of 2026, those numbers seem to be based on 2024.



Second, I found a conflicting or possibly revised report on the same website from Aug 2025 that puts the 10 year growth projection numbers closer to 6.5% to 7.5%, but that groups SWE jobs in with a few other areas.




And if that wasn’t enough, I found I could easily trigger a 500 error on the BLS website simply by omitting a query string var.



Not that the error reflects on the quality of data, but you would think with the amount of tax dollars that go into the BLS, they would know how to return a 400 status code when appropriate.



What do I think? I think software engineering will look very different in 10 years. New jobs will emerge that we couldn’t have imagined. It will be like what a “Mobile App Developer” would look like to someone in the 80s/90s.



Where do you think the Software Engineering field will be in 10 years?

1 week ago | [YT] | 1