The Shift: When the Tools Finally Caught Up

Building has never been easier than it is right now.

The latest AI coding tools like Claude Code and Cursor have enabled developers to build with rapid speed and branch out into new domains without skipping a beat. I've been using AI coding tools for a couple of years and coding agents since they came out earlier in 2025, but it just feels different now.

When agents first came out, they were helpful, but if you tried to build an entire project from start to finish with the agent writing the majoirty of the code you'd end up with a mess of AI slop. A couple of months ago, I started to use agents to write a good bit of my code at work and drastically increased my velocity. Agents have been a great additional set of "eyes" used to review code and catch mistakes that I would have missed otherwise. And now, I use agents to build entire applications. When these agents first came out, it was fun but not a reliable way to develop a new feature, let alone an entire app.

But now it feels different. Opus 4.5 is the new frontier.

The Spark

About a month ago something changed. I'd been working on a side project for some time and stalled out a bit. I had reliled on early models too much and was left with a codebase of slop. New features and refactors were almost impossible to complete without creating a new problem. While I didn't want to abandon the side project completely, I knew that I'd likely need to start over from scratch to build it the right way. I had lost too much context to the agents.

I was so infatuated by coding with agents that I had to build something to completion. At the same time, I needed something new to work on. Something that I would use on a daily basis.

A little defeated, my ADHD brain kicked in. I began thinking of other problems in my life that I could solve with an app. That's when I thought of Tally, a budgeting app made simple.

From Idea to Prototype

My wife and I had tried all of the big brand budgeting apps and none of them felt right. We'd tried spreadsheet templates and eventually ended up with a simple custom-built spreadsheet for our budget. It worked, but it made budgeting feel like a chore. We'd often dread doing it.

So I spun up a chat with Claude and began brainstorming. I shared a screenshot of our spreadsheet and wrote a brief explanation of what I was looking for to give Claude some context. This included must have features, tech stack, and the purpose of the app. The original output was immediately helpful. I was able to quickly judge how feasible building a prototype would be with just a few prompts.

Later that night I had a working prototype I could start using.

Ship Fast, Ship Often

I read about the culture at Cursor and how they ship at an insane speed. This article resonated with me. Shipping products is ultimately why I got into software engineering and why I love it to this day. I want to build products people want to use and continue to make them better.

I brought this strategy to Tally. I'd use the app, find something wrong or think of an improvement, spin up an agent, test it, and ship the new feature.

This felt different.

Not different like "oh, this is a nice productivity boost." Different like I was finally building the way I'd always imagined building. Ideas to features in minutes. Bugs squashed before they had time to frustrate me. The gap between thinking "this should work differently" and deploying the fix collapsed to almost nothing.

Nearly 100% AI Generated

Nearly 100% of the code written in Tally is AI generated, and it's better code than I could have written in a fraction of the time.

I was blown away. Just before Opus 4.5 came out, I was struggling with a pile of AI slop. Now, I've guided AI to generate high quality output and a working app.

When building, I start in plan mode in Cursor. Before agents, that's how I worked too. Starting with a plan before jumping into coding. When a bug arises, I switch over to debug mode and let the agent get to work.

Typically what has worked for decades in software engineering tends to work well with agents too. As I'm writing this, Cursor dropped a detailed post about coding with agents best practices.

What Changed

The models got better. That's the obvious answer. But the tooling caught up too. Claude Code running in my terminal, Cursor with its plan and debug modes, the seamless context switching between brainstorming and implementing. The friction disappeared.

When I first started using AI for coding, I spent more time correcting mistakes than I saved. The code was functional but brittle. It didn't understand my codebase. It didn't understand my intent.

Now it does. Or at least, it understands enough that the collaboration feels productive rather than adversarial. I'm not fighting the AI to get what I want. I'm working with it.

The New Bottleneck

Here's the thing nobody talks about: when building becomes this fast, the bottleneck shifts. It's no longer "can I implement this?" It's "should I implement this?"

Product thinking becomes more important. Design decisions carry more weight. The speed amplifies both good judgment and bad judgment. If you're building the wrong thing, you'll build the wrong thing faster.

This is where taste matters. Taste in what features to build. Taste in how the UI should feel. Taste in knowing when something is "good enough" versus when it needs more polish. Agents can execute on your vision, but they can't define the vision for you. The developers who will thrive in this new era are the ones with strong opinions about what makes a great product.

For Tally, that meant being ruthless about scope. Every feature had to earn its place. The agents could build anything I asked for, but that didn't mean I should ask for everything.

Try It Yourself

If you've been hesitant to dive into AI coding tools, or if you tried them a year ago and bounced off, now is the time to try again. The experience is genuinely different.

Start with a small project. Something you've been meaning to build. Let the agent take you further than you expected to go.

A word of caution: agents still require guidance. They make mistakes. Every line of code should be reviewed before pushing to production. Think of them as a brilliant junior developer who works at superhuman speed but still needs oversight. The magic happens when you combine their raw output with your experience and judgment.

That's what makes this moment so exciting. Agents don't replace developers—they empower us in ways we couldn't have imagined a year ago.

Try Tally Budget: tallybudget.app