AI is not your engineer, but it is the best junior you've ever had
The AI-will-replace-engineers crowd and the AI-is-useless crowd are both selling you something. Neither is telling you the truth.
Modern AI tools are excellent junior engineers. Not excellent senior engineers, which is where the hype machine keeps getting caught lying. Excellent juniors. That’s the smaller, more useful claim, and I’m making it from actual experience: real code, real small businesses, right now.
What a good junior actually does
A good junior engineer is fast, tireless, and willing to do the boring work. They’ll write the CRUD endpoint. They’ll build out the test harness. They’ll scaffold the new React component. They’ll read the docs you don’t want to read and summarize the three options. They do not, however, decide what should be built, or why, or in what order, or whether the thing you asked for is actually what you need.
A good senior engineer, by contrast, is the person who says “we should not build that.” Or “if we build it the way you’re describing, it will cost us three months six months from now.” Or “there’s a simpler version of this that gets 80% of the value.” That judgment call, whether something should exist at all, whether it should exist in the form being requested. That’s where senior engineering earns its keep.
AI does the first job very well. It does the second job poorly, in a confident voice. The trap is that the confident voice makes people think the second job is being done.
How I actually use it
I use AI every day. What it’s actually like, versus the hype version:
Scaffolding new code, writing boilerplate, generating tests from specs, translating between frameworks: all of this used to be slow, careful work. Now it takes hours instead of days. Writing the fifth draft of copy when I already know what the first four should avoid: genuinely useful. Every one of these is real output.
Architecture decisions, choosing between two approaches, debugging something weird: I use it here too, but differently. It will cheerfully tell me the wrong answer in a plausible tone. So I treat it like a junior: listen, use it to prompt my own thinking, verify before trusting. The rubber duck now talks back. The judgment stays with me.
Where it actively hurts: anything requiring actual business context. Saying “no” to a client. Reading between the lines of a vague requirement. Knowing what broke last time, or that the CEO’s favorite feature is politically untouchable. AI doesn’t know your customer. It doesn’t know your history. These moments need a real person who’s been in the room.
What this means for you
If you’re running a small business and someone is telling you AI will let you skip hiring engineers entirely, they’re lying. If someone is telling you AI is a toy that doesn’t do real work, they haven’t actually used it in the last twelve months.
The actual reality is that AI has lowered the cost of a lot of engineering work by an order of magnitude, which means the economics of what’s worth building have shifted. Things that didn’t pencil out as projects six months ago do now. Integrations you put off because they’d take three weeks of someone’s time now take a day. The custom internal tool you bought a SaaS subscription for, because building it was too expensive, is suddenly worth building.
This is the real revolution. It’s not “the engineers are replaced.” It’s “the floor for what’s worth building has dropped by 10x.” Which means the backlog of “someday we should automate that” just became your competitor’s Tuesday.
The trick is knowing which job the junior is good at, and which job still needs the senior. Get that wrong and you get the confidently-broken code that’s all over the internet right now. Get it right and you move at a pace that used to require a team.