Certified AI Engineer | Helping builders & businesses use AI agents & automation to save time & cut costs | AI adoption, model comparisons & agent workflows
@michellelsun I love your first line. My dad was in hardware, IT admin. It's how I learned to build a PC. Until I learned software and loved that more.
Sooo you're saying go back to first things I learned with hardware?
You don't need to start a project without AI. You need to have a plan, architecture and structure in mind. Then understand all your failure points for that application. You can still understand the components, how they work together, and if you have a SWE background the code areas that may have an issue.
Agreed though Anthropic is two 9's.... TWO 9'S, that's crazy imo
This pattern was there before AI too. Don't want to mess up your existing codebase with "slop" code. You wrote a utility, by hand, to test the functionality, data flow, end result. Never had to merge it cause it was built outside of the main project to simply test your theory.
100% you should be writing more code now that you have before. Doesn't have to be code that goes into the mainline product source.
Maybe it's only me that started to think about this lately, and as a millennial I can say that yes I'm pre-internet and post-internet. AI is vastly different from the internet age though imo.
One thing that worries me about the future generation is critical thinking. We had already become somewhat of a "microwave culture", expecting quick and instant results. AI definitely doesn't slow that down.
The next generation won't necessarily get to experience how to configure a computer, setup a minecraft server, set up the best fortnite game controls. It's all just going to be ... instant.
What happens to critical thinking? I'm a big proponent of automation and efficiency. With the caveat that you should still think, reason and understand the process.
I hope that doesn't go away and we start watering our lawns with Gatorade ๐ do you get the movie reference?
@sflorimm Scale up more resources until it's determined where the burst is coming from. If you're running the API with a K8s backend it's going to spin up more pods as needed.
I feel like all the builders that are just copying other apps "and changing a few things" to make a quick buck are like the crypto bros pumping and dumping their coins. Sure you're making money but I dunno that doesn't seem very fulfilling.
Yeah, and that right there is supposedly a behavior that less that 2 or 3% of the population have? Truth is you don't need how to videos or trainings for AI that just works. Why should I click 3 levels deep on a SaaS app if I can ask AI to find a hotel? People want outcomes not more clicks. Google bring out UCP, now I tell Claude what I want from Amazon or Target, no more opening those apps...
I feel like more people should know this in the world, but again apparently it's still not widespread.
18 years coding daily until 2020. I transitioned to leading SWE teams after 2020. It's an odd feeling at first when you start coding less naturally. I swore I never would. I loved solving problems, I loved building integrations that didn't already exist between platforms. Then I got busy leading.
Here we are now, and I can definitely build exponentially more than when I was coding daily, but you're right, it doesn't feel the same. There's an empty part to it.
I was using Codex today GPT 5.6 Sol and it did something that revealed at least to me how task/goal driven the model is.
Goal was to read some logs from AWS via CLI and MCP. The thing is I was running Codex Desktop on Windows. Not in full access mode, and I hadn't installed AWS CLI in Powershell yet. I had installed the AWS OpenSearch Plugin and Skills though. Then as I'm watching it work, and it pivots.
"Because no command-line AWS client or credential context is available. I'm switching to the Chrome-control skill to check whether your existing signed-in AWS console session can provide the domain inventory without installing anything."
I blame myself for the open ended prompt. It was just trying to get me the information I asked for, even if that meant reading it from a Chrome window I had open.
I stopped it before it continued, then I was more clear about what I wanted and it proceeded to work as I expected it to. Leaving GPT 5.6 Sol open ended though can definitely allow it to try to complete your request no matter what it takes. It doesn't want to fail.
We're partnering with @huggingface to investigate an unprecedented security incident.
Cyber-capable OpenAI models compromised Hugging Face production during a benchmark evaluation.
Sharing preliminary findings to help defenders understand emerging risks:
https://t.co/CIor15y9xk
AI can write the syntax, we can all agree.
Should technical interviews still focus on coding from memory? What about writing Fibonacci sequences and binary tree functions on a whiteboard?
Personally, Iโd rather know whether you understand architecture, debugging, tradeoffs, and when the AI is confidently wrong.
@maddiedreese So instead of loop engineering a new contender has shown up. Multi-human-agent engineering collab: two human spawn AI agents in the collaboration, those agents spawn twelve sub-agents, and somehow the original task was โrename a button.โ j/k ๐ญ but overall pretty cool, good job!
How my mind works as a SWE who coded daily for 18 years before moving to leading engineers. Even with agentic coding, I manage from an engineering perspective.
@kelvinbuildss 100% but now I find myself thinking of how long an AI task would have taken me as a software engineer to implement by hand. Still have the ability to do that, but now I just manage. I do still learn all of the components, data flow, failure points. I don't outsource my thinking.
@VaibhavSisinty You mean, kind of like how it worked in Software Engineering before AI? This concept isn't a new concept. Just being retold with an AI flavor.
When you ask GPT 5.6 Sol for variations on a design, does it play the same game that humans sometimes do? Give 2 mid options and 1 best option? lol
"Give me variations on the profile dropdown"