English Is the New Programming Language
Explore how programming has transformed from precise coding to navigating the unpredictability of AI-driven English commands in 2026.
I learned to program in a language with a grammar. A compiler that told me no in under a second. A tight loop between my brain and the machine.
In 2026 I program in English.
The new toolchain
English is the source language now. The AI is the compiler.
Except it's a compiler like no compiler you've ever used. It's non-deterministic. It's slow. It hallucinates. It compiles your intent into something, and that something is different every time you run it.
Imagine if go build produced a different binary every invocation. Sometimes correct. Sometimes almost correct, which is worse. Sometimes it quietly rewrites half your codebase because it "noticed an opportunity to refactor."
We would have thrown that compiler in the sea.
Instead we made it the center of our profession.
Amazon Prime development
Here is my job now:
I place an order. I describe what I want in careful, hedged, defensive English. I hit enter.
Then I wait.
Same-day delivery, if I'm lucky. The package arrives. I open it. It's almost what I ordered. The color is wrong. It's the EU plug instead of the US one. There's a free bonus feature I never asked for, bolted on, untested.
So I file a return. I re-describe what I wanted, more carefully this time, like I'm arguing with customer service. I hit enter.
Then I wait again.
This is not engineering. This is procurement.
The feedback loop is dead
Programming used to be a conversation measured in milliseconds. Type, compile, fail, fix. The machine pushed back instantly and honestly. That loop is where the learning lived. That loop is where the fun lived.
Now the loop is measured in minutes. Prompt, wait, review, sigh, re-prompt, wait.
The waiting is the worst part. Too short to context-switch. Too long to stay in flow. So you sit there, busy waiting, a human process blocked on I/O, refreshing a chat window like it's a tracking page. Your code is out for delivery.
I used to be the CPU. Now I'm the spinner.
Prompting is not a grammar
People say "English is the new programming language" like it's a triumph.
It's a downgrade.
A programming language is a contract. Precise semantics. If it compiles, you know what it does. English is the opposite: ambiguous by design, interpreted by a statistical model in a mood you can't inspect.
We spent seventy years building languages that removed ambiguity from the human-machine interface. Then we shipped the most ambiguous language humanity has ever produced and called it progress.
Specs written in English were always the problem. Now they're the whole job.
Reviewer of the machine
I don't write code anymore. I read it.
All day. Diff after diff of plausible, competent, soulless code that I have to own without having made. The AI is the author. I'm the guy signing for the package.
The productivity gains are real. I won't pretend otherwise. The output per week is absurd. But nobody asked what got optimized away.
What got optimized away was the part where I was a programmer.
Nobody's home
Here's the part nobody wants to say out loud: we're losing touch with the internals.
Everyone in the org uses the AI now. All of them. Every service, every module, every migration — machine-written, human-approved. Approved by humans who read the diff but never walked the code. There is a difference between reading a map and having walked the terrain. We are all map-readers now.
Ask anyone on the team how the retry logic actually works. Watch them open a chat window to find out.
The codebase has become a rental property nobody lives in. It's maintained. It passes inspection. But nobody knows where the pipes run anymore, because nobody laid them.
The guidelines are the job
Our defense against this? Guidelines. Markdown files full of rules for the machine. Preferred stack, preferred patterns, what not to touch. Guardrails for a driver that isn't us.
Congratulations. We reduced software engineering to a config file.
And here's the punchline: if the job is writing English guidelines and approving deliveries, why does it need us? Management can write English. Management loves writing English — they've been doing it in Confluence for decades. The moment the guardrails are good enough, the person placing the order doesn't need to have ever laid a pipe.
We're not being replaced by AI. We're being replaced by whoever's cheapest at supervising it. We spent two years training our own substitutes, one carefully written guideline at a time.
The last commit from a human won't be code. It'll be a markdown file explaining how to work without us.
Order confirmed
So this is development in 2026. Describe. Submit. Wait. Inspect. Return. Repeat.
The craft became a checkout flow.
And the industry's answer to every complaint is the same: get better at ordering. Prompt engineering. Context engineering. Agent orchestration. An entire discipline dedicated to talking more clearly to a machine that used to just do what I told it, deterministically, in a language built for exactly that purpose.
I still have that language. It still compiles in under a second. It still tells me no, instantly and honestly.
Some evenings I open a terminal and write Go by hand, just to remember what it feels like when the machine answers immediately.
No tracking number required.