I trust Claude Code.

Back in March, I wrote about why I pity the developers who haven’t yet jumped on the agentic coding bandwagon. I also pity the developers just starting out, who will never quite understand the power that they now have at their fingertips.

But most of all, I really pity the developers who refuse to use agentic development tools because they don’t trust AI agents. 

I understand that saying I trust Claude Code is a controversial statement. I know that the coding agent isn’t perfect, that it will make mistakes, that it will “hallucinate,” and that it will not always do what you want in the way you want it done. 

But you can fix that.

Start slow

But guess what? The same is true of every human developer on the planet. When you hire a new developer, especially a brand-new junior developer, they are going to make mistakes. They are going to misunderstand, and they are going to miss things that they shouldn’t.

Of course, you’ll take the time to teach this person, show them the error of their ways, and help them grow. You learn what they know and don’t know, what they are good at, and what they are bad at.

And admit it, even you sometimes take things in a bad direction and end up deleting half a day’s work, right?

The same is true for coding agents. Just like with a human developer, if you give your coding agent crappy instructions, if you don’t give it proper guidance, if you don’t take the time to point it in the right direction, you’ll get bad code. If you don’t keep an eye on things and continually nudge things in the right direction, you’ll end up in the wrong place. No surprise.

And trust isn’t binary. Imagine if you hired a junior developer, gave them a two-sentence instruction, got back something that wasn’t quite right, and then fired them because of it. That would be silly, right?  Well, that is what a lot of developers are doing with coding agents. Saying “I tried agentic coding, and it hallucinated something, so it is clearly worthless” isn’t any different.

Gain speed

Trust takes time. You get out what you put in.

My practice has been to make sure those guardrails are in place, carefully track what Claude Code is up to, and continue to guide its responses and improve its instruction set. The more I do that, the better things go. I find that now I seldom have to steer things back onto the correct path.

Even better, you can leverage solid guardrails and guidance from the very start. Tools like gstack and Superpowers can turn Claude Code and a single developer into a whole company’s worth of coding intelligence.

I’ve had great results. Claude Code gets my work done at least 10 times faster because I trust it. I’ve taken the time to teach it and provide it clear instructions, and — surprise! — it follows those instructions. And if it doesn’t, I can refine the guidelines and make things better. 

And just as it has always been among humans, trust is the key element that will separate those who plod along typing their own code, and those who move at the speed of light with a coding agent.