Mastering Your Developer Workflow: How AI is Changing Software Engineering

In the fast-paced world of software development, a developer's workflow is never static - it's a constantly evolving beast. In a landscape now dominated by AI, a hybrid approach combining traditional tools with powerful new agents is proving to be the most effective way to drive both quality and speed. We've been experimenting with a new setup that's delivering incredible results, and here's a look at how we're building, testing, and shipping code in the era of AI.

The Right Tools for the Job: A Hybrid Approach

Our setup has shifted from an all-in-one editor to a more specialized toolkit. While VS Code is still a go-to for quick code lookups, we've found our main productivity comes from using Ghostty as our primary terminal. Its stability, even when handling large amounts of text, is unmatched.

For AI-assisted coding, we've found a sweet spot by using a combination of models for specific tasks. While Gemini's editing tools can be messy, GPT-5 is our go-to for reviewing plans and providing a solid foundation. Claude, on the other hand, is a champion at refactoring and code cleanup, helping to keep technical debt at bay. This strategic use of different models for their strengths is a game-changer.

Our current hardware setup, a Dell UltraSharp monitor, is a testament to the need for a wide workspace. The massive screen real estate allows us to have multiple AI instances and a browser open simultaneously, reducing the friction of context switching and keeping all the information we need right in front of us.

The Art of Context and Planning

Our new workflow is built on a "plan mode" philosophy. For complex tasks, we outline a plan in a file and have GPT-5 review it for efficiency and foresight. For smaller tasks, we simply use a series of small, focused prompts. It's about giving the AI enough context to work effectively without cluttering the conversation.

Managing context is crucial. By having the initial topic in a status line and a session ID, we can easily jump back into a project without losing a train of thought. This approach has proven to be far more efficient than the "worktree" setup, which we've found can slow us down. The key is to carefully choose work areas to minimize cross-pollination.

The Hard Parts: Custom Infrastructure and Smart Testing

While AI can automate a lot, the truly challenging parts of a project - like designing distributed systems, picking the right dependencies, and building a forward-thinking database schema - still require a human touch. we've also found that building custom infrastructure, such as admin pages and command-line interfaces (CLIs), has dramatically accelerated our workflow.

Our testing strategy has also evolved with AI. we've learned that having the model write tests within the same context as the code is the most effective approach. This allows the AI to catch issues it might otherwise miss. Automated tests still have their place, but the most valuable insights come from models that can analyze the code in real time. This focused approach on context helps prevent the AI from "spinning up" unnecessary processes and ensures a faster, more efficient workflow.

The Future is Collaborative, Not Automated

Our current setup has allowed us to achieve an incredible amount of work, but it's not a set-it-and-forget-it system. The most crucial part of this new workflow is the constant steering of the models. The moment you notice an AI agent drifting off course, you need to be ready to step in and correct it.

As new AI models and rate limits emerge, the cost of this work will surely increase. However, the value provided by this hybrid, human-in-the-loop workflow is undeniable. The future of software engineering isn't about fully automated background agents - it's about a dynamic partnership between the developer and AI, where both sides play to their strengths to build better, more reliable software, faster than ever before.