Embracing the Future of Coding

Today marks a breakthrough for me as a software engineer, one of those rare moments where things finally click, and I feel myself leveling up in a meaningful way.

I’m writing this not just to share the technical insights, but to pause and acknowledge the moment. If I ever come back and reread this post in the future, I want to remember how it felt: the excitement, the clarity, the sense of progress. This isn’t just about code, it’s about acceleration.

Agentic coding is here

A couple of weeks ago, I tried Claude Code for the first time, and I was impressed. As a seasoned software engineer, I’ve tried a range of AI coding assistants, but Claude Code’s agentic capabilities took me by surprise. From the power of its Sonnet and Opus models to its newly released Plan Mode, it felt like I’d unlocked an entirely new workflow.

Discovering the Power (and Price) of Sonnet

When I first tried Claude Code, I was impressed by how intelligently the LLMs parsed and generated code. Back then, the Sonnet 3.7 model, in particular, excelled at understanding complex codebases. I could ask it to explore interdependent modules, and it would map out dependencies with accuracy.

That said, unlocking these advanced models comes at a premium. Token costs for Sonnet and Opus on Claude Code are higher than many other providers. After a few exploratory sessions, I found I already burned my 20$ of tokens. Soon after I switched to Cursor.io as the pricing was more predictable.

The “Plan Mode” Revelation

Fast forward to yesterday: I discovered Plan Mode, a recently launched feature in Claude Code that has completely shifted my perspective on AI-assisted development. Instead of sending prompts to an LLM and hoping for the best, Plan Mode lets you sketch out a multi-step strategy:

1. Define the Goal. Write down the end-state you want—a new feature, refactored module, or architecture upgrade.
2. Outline the Steps. Break the goal into logical sub-tasks or milestones.
3. Assign Context. Point the AI to specific files, functions, or architectural patterns it should honor.
4. Iterate on the Plan. Give feedback, refine priorities, and re-run the plan until it’s rock-solid.

This structured approach mirrors how I naturally think about engineering problems. Typically starting with a high-level design, then drill down into details. With Plan Mode guiding the AI, I felt like I was acting like a software architect and steward of the coding process, not just throwing prompts into the void. And I knew that when I plan it as accurate as possible, Claude Code will deliver create results.

A Full Day of Agentic Coding

Encouraged by Plan Mode, I upgraded to Claude Pro and devoted an entire day to agentic development. The results were nothing short of astonishing. By mid-afternoon, I had:

• Recreated a Task Manager Tool with Drupal integration in under an hour.
• Ensured every function passed the code sniffer and adhered to Drupal coding standards.
• Followed proper architectural patterns without lifting a finger on boilerplate.

The AI traversed my codebase, identified where to plug in new hooks, and even wrote documentation comments—all guided by the plan I laid out. Of course, I still performed final reviews and minor tweaks, but the heavy lifting was done by Claude Code. This functionality is now available as a beta feature in production!

From Ideas to Implementation

Previously, I had many ideas for AI-driven system coding. Some attempts succeeded, but most were cobbled together and fragile. Today, with the blend of Sonnet/Opus and Plan Mode, I can:

• Generate robust feature implementations that align with existing architecture.
• Maintain coding standards automatically, thanks to built-in linting and sniffers.
• Collaborate with the AI as a true development partner, steering it toward the exact behavior I need.

It’s a workflow I’ve long dreamed of: specify intent, let the tool explore the code, refine its understanding, and then produce production-ready code.

Why This Matters for the Industry

Claude Code isn’t just another AI assistant, it represents a shift toward fully agentic development. Engineers will soon expect their AI tools to:

• Autonomously analyze large codebases in minutes.
• Propose high-level architectures and scaffold entire modules.
• Learn project-specific conventions and enforce them consistently.

This level of assistance will dramatically speed up feature delivery, reduce errors, and free developers to focus on creative problem-solving rather than boilerplate.

Looking Ahead

On this fine day, June 17, 2025, I’m convinced we’re at the beginning of a new era in software engineering. Agentic coding platforms like Claude Code are more than gimmicks; they’re productivity multipliers. Yes, there’s an investment in both cost and learning curve, but the payoff is immense.

For fellow engineers: if you haven’t yet explored Claude Code’s Plan Mode, give it a spin. Sketch out a plan for a small feature, let the AI loose, and watch as it turns your designs into real code. It may just change the way you work, like it did for me.

Happy coding!

Photo credit: micheile henderson