Help, Claude Code Is Smarter Than My Co-worker
I’m going to say something that’ll make me unpopular at standup: Claude Code is better at certain parts of my job than some of the people I’ve worked with.
There. I said it. Now before you grab your pitchforks, let me explain.
Where Claude Actually Wins
Last week I needed to write a Terraform module that wraps an Azure Verified Module with CIS compliance preconditions. The kind of thing that requires reading three different docs, understanding the AVM interface, and writing test files with expect_failures blocks.
I described what I wanted. Claude wrote the module, the tests, and caught a variable naming inconsistency I would’ve shipped. Took about 4 minutes. The last time a junior colleague did this, it took a day and a half and the tests didn’t actually test anything.
This isn’t a knock on junior engineers. It’s an observation about what AI is genuinely good at:
- Boilerplate with patterns. If it’s been done a thousand times before, Claude nails it.
- Cross-referencing docs. “Make this work with AzureRM 4.0 AND follow CIS NS-2” - it just does it.
- Catching inconsistencies. It reads every line. Humans skim.
Where Claude Falls Flat On Its Face
But then I asked it to debug why our Azure DevOps pipeline was failing only on self-hosted agents with a specific network configuration. It confidently gave me three solutions. All wrong. Not just slightly wrong - fundamentally misunderstanding what a self-hosted agent actually is in our setup.
The problem was a firewall rule that our infrastructure team had changed two days ago. Claude doesn’t know about your firewall changes from Tuesday. It doesn’t know that Henk from networking “temporarily” opened port 443 to the wrong endpoint six months ago and forgot about it.
Context is everything. And context is what humans have that AI doesn’t.
The Uncomfortable Middle Ground
Here’s what actually worries me: not that AI will replace engineers, but that it’ll create two classes of them.
Class 1: Engineers who use AI to move faster while still understanding what they’re building. They review the output, catch the subtle bugs, and know when Claude is confidently wrong.
Class 2: Engineers who paste the output and push. They ship fast until something breaks, and then they’re stuck because they never understood what they shipped.
If you’re reading AI-generated Terraform without understanding the resource lifecycle, you’re in class 2. It’s not a matter of if it’ll bite you, it’s when.
My Actual Take
Claude Code is a ridiculously good pair programmer for about 70% of what I do. The other 30% requires knowing that Henk changed the firewall, that the staging environment has a weird DNS override from 2024, and that the PM actually wants something completely different from what they wrote in the ticket.
That 30% is where you earn your salary.
So no, Claude isn’t replacing my co-workers. But it is making it very obvious who actually understands what they’re building and who was just good at Googling.
And yeah, I’m including myself in that assessment.