Eduardo Ordax’s Post

AI won’t replace software engineers — but engineers who leverage AI will replace those who don’t. From bulky personal computers to today’s cloud-native ecosystems, we’ve consistently adapted and evolved with technology. The pattern is clear: those who stay two steps ahead — using AI to write smarter, faster code — will lead the future. Remember Nokia vs. Apple? Adapt to change — or change will change you. #AI #SoftwareEngineering #AdaptToChange #FutureOfWork #Claude4 #TechEvolution

I made with Claude the sleekest frontend redesign in just 2h. Then I've spent a whole week fixing all the logic it broke. True story.

Most importantly: it did it all with confidence, all the way to the "DROP TABLE" portion...

I’m finding that the frequency of … “None of it Worked”… is becoming less common. More emotional, less believable🙇🏼♂️ More-&-more people are trying AI code assistants to Vibe out entire codebases. More non-coders are trying to write code, where they previously would have abandoned their Killer App 💡🚀 The AI coding orchestrators are serious engineers, and contain serious tech inside them. They’re very smart & extremely well trained. What we’re starting to see is that… the hyped-up failures are now mostly the result of the Humans controlling them❗️ Eduardo Ordax

My golden prompt in Cursor: “Solve this (problem, bug) by making as minimal changes to the code as possible while following my codebase architecture (add the architecture documentation as context). Claude 4 is surprisingly good at following this. 

I cant tell you how many times I have totally nuked a feature branch with my tools. This was a lot more prominent an issue 6+ months ago but either I am getting better at this or the LLMs and Agents are. I suspect its a little of both. The industry leading AI IDEs are implementing robust memory managers, separate plan and build modes, and a more robust documentation and PRD process will help keep this issue at bay. But, this is also why I still manage my own git commits and PRs. If the agent or LLM fails me I can still restore the last commit I trust. Git blame is also very very effective with Agentic development.

Beautifully wrong 😂 I've covered this topic a few hours ago: https://www.linkedin.com/posts/terezijasemenski_ai-llms-generativeai-activity-7332677534066671618-ZFM5

Context windows matters. Claude 4 has 200k context window which is about 500 pages. That means it can read that much at a time. Depending on people’s monolith codebase, it can be larger. Googles Gemini 2.5 pro has a 1 million context window and soon expanding to 2 million. It also rivals Claude 4’s coding. They are getting better every release. It still boggles my mind on how if something doesn’t work 100% perfect on the first try some engineers immediately think it’s useless. Llms aren’t ready to take on your entire codebase. They are great at creating net new, or prototypes. They are good at looking at small samples of code, and when worded properly, can give you suggest on how to fix bugs or different wants to fix a problem. LLMs have their use cases and are rapidly expanding every few months. Is the hype sometimes overblown? Yep. But use it for what’s it’s good at. Not a 1 click magic button to remove all tech debt and whatnot.

It’s only a matter of time companies will start ditching ai and re-hiring humans again

In my opinion AI won’t replace software engineers ,it will empower them. It helps scale up their work, maybe automate repetitive tasks, and optimize development processes. Which certainly allowing engineers to focus on more complex problemsolving.