The Modern Engineering Stack: Moving Beyond the Search Bar

AI is quickly becoming an accelerator for more things than just google searches. While that is a fundamental use case, there are a few things that help quite a bit with actual work.

I am a user of Copilot, and while it’s not the fastest AI on the block, it does the absolute best job of keeping business data inside your business. Of course, this relies on using a business account configured with the appropriate permissions. The real value here is that the model isn’t trained on your business data. If it were, your unique ideas or proprietary data could easily be leaked by someone else using the chat functionality in the personal versions of Copilot, which is a worst-case scenario. That security aside, the integration with Excel, Word, PowerPoint, Teams, GitHub, and Azure is fantastic. It leverages your internal data to not only understand your specific workflows but to actually do the work for you. Features like CoWork, for example, deliver much of the agentic functionality you see in tools like OpenClaw, but with a fraction of the risk. It might have its limitations but putting security first rather than as an afterthought is a great way to protect your digital assets.

Anthropic is my absolute go-to for vibe coding. It does have a tendency to run out of tokens fairly quickly if you are throwing big development projects at it… which is still a weird thing for me to say, given that I am not a traditional developer. I’m an engineer who has wanted these kinds of tools for years but could never explain exactly what I wanted to a programmer who didn’t understand my operational needs, and I didn’t want to waste time on the disconnect. With Claude, I have successfully built API Proxies, File Sync Utilities, and UniFi Management tools, among others. Knowing the foundational process of how to do something and having AI handle the execution makes things that were previously impossible for me completely achievable. I can knock out a working prototype from scratch in just a couple of days. I prefer Claude’s GUI features over the rest, and I like to use it in conjunction with other AIs, like OpenAI inside of VS Code, so I get a highly comprehensive application that balances performance, usability, scalability, and stability.

This is the OG of AIs, and the dynamic right now is a lot like the classic battle between Intel and AMD processors as Claude and OpenAI fight for relevancy. In my testing, Claude is currently the better choice for front-end work and mapping out the initial visual application, but Codex reigns supreme when it comes to backend services. I use OpenAI to build most of my containers and core backend logic. It does a pretty mediocre job at visual design, but when it comes to optimization and making things work efficiently under the hood, it’s the model I call on once Claude has the application laid out and running. I also really value the voice response feature, being able to chat with an AI agent to run through complex architectural ideas while I’m driving is a huge time-saver.

Gemini is the fastest of the AIs in my testing, making it my primary thought partner for bouncing ideas around. The sheer volume of data it has real-time access to far outweighs what OpenAI, Claude, or Copilot can reach (outside of the internal business data I feed into Copilot). For example, it can write functional OpenSCAD scripts to create 3D printing models that I can drop straight into my Bambu printer. The initial responses are incredibly solid out of the gate, and they get significantly sharper if you start the conversation by establishing the full context of what you are trying to achieve. This is a universal truth with AI: you have to know what you want, ensure you provide the data needed to answer the scenario, and then refine the output through ongoing conversation with the model.