I've been using GitHub Copilot for a while now, and while it's decent, my company is really diving headfirst into AI, with everyone in the team getting access to cursor licenses. I appreciate the model as a way to brainstorm or 'rubber duck' ideas, and I think the agent mode that allows browsing through application files to answer questions is pretty cool. However, I have some serious reservations about the industry's push toward more agentic implementations. I'm in my mid-30s and have spent years in this field, so it's not just a case of 'get off my lawn'—I genuinely want to create something I won't regret six months down the line.
1) I recently watched a video featuring Bedrock, where someone built a customer service agent to handle returns. The concept seems straightforward: connect some models and lambdas with basic instructions. But what's stopping the model from hallucinating when processing data or interacting with customers? We lack control over the outputs. Sure, I can feed information back in, but that means pushing more requests to a black box. When we pay for a service, we expect reliability, and it feels risky to be outsourcing so much to an uncertain 'magic box.'
2) I've also been digging into the potential for model poisoning. While nation-states typically manipulate models for specific narratives, I'm concerned about the risk of my code becoming outdated if I rely on a third-party model that could evolve unpredictably based on new training data.
3) And seriously, why are we trying to automate everything? Sure, cutting costs is appealing, but running AI models all the time can be much more expensive than simply having a web form that connects back to the same lambdas.
I have a few more concerns, but am I wrong to think that even if these models are impressive, they may not be the best path forward? Plus, announcements from companies like Shopify suggesting they won't hire unless candidates can prove they can't accomplish tasks with AI only adds to the pressure to adapt, but I want to face that future with my eyes wide open instead of falling for marketing hype.
5 Answers
The real challenge comes from ongoing support and maintenance. If your role revolves around basic tasks, you risk being replaced, especially as firms look for cost-efficiencies. However, technical roles focused on infrastructure will always be essential. AI doesn’t magically fix everything.
LLMs (Large Language Models) are still carving out their niche, but the hype often oversells their usefulness. There’s potential, for sure, but don’t depend solely on them—apply them wisely and they can add value without compromising job security.
True! Misguided applications of AI can definitely jeopardize our roles in the long run.
There’s been a lot of over-investment in AI hype. Management is clinging to this hope that it will reduce labor costs, but the reality is, it may not pan out as expected. Everyone's afraid of being left behind, but deep down, many know the hype is outpacing reality.
That’s true, but if implemented wisely, AI can make developers more efficient in their routine tasks.
Many companies are making the mistake of thinking they need to apply AI everywhere. They should focus on mundane tasks that employees dislike, like data entry or processing receipts. That’s where AI can truly shine: automating boring stuff instead of trying to make everything AI-driven.
Exactly! Focus AI where it can actually solve problems rather than forcing it into areas where it doesn’t fit.
The AI hype is definitely driven by financial gains as much as the technology itself. A lot of execs seem to believe AI is a cure-all for their problems. While it's exciting tech, there's a distortion in how it's perceived due to the money involved. It’s cool with some real-world applications, but we need to keep a level head about it.
Agreed, the enthusiasm from the top brass often overlooks the practical limitations of AI.
Right? They just hop on the latest trend without understanding the implications.
For sure. It seems like AI can streamline processes, but it also needs real human oversight.