Forem: Gilder Miller

The Last Programming Language You'll Ever Learn Is English

Gilder Miller — Thu, 30 Apr 2026 14:17:02 +0000

There's a saying going around that hits a little too close to home.

2024's programming language is Solidity. 2025's is Python. 2026's is English.

If that doesn't make you pause, read it again.

We went from writing smart contracts in a niche language, to scripting AI models in Python, to literally just talking to machines in plain English. And somehow each jump happened faster than the last.

The Shift Nobody Prepared Us For

Here's what's actually happening. The bottleneck isn't code anymore. It hasn't been for a while.

The real challenge now is asking the right question.

Think about it. When you write Python, the syntax is strict. The interpreter doesn't care about your intent. Either it runs or it doesn't. But when you prompt an AI model, everything changes. The same request phrased three different ways gives you three different results. "Build me a dashboard" versus "Create a real-time analytics dashboard using React and D3 that updates every 5 seconds" versus "Make something that shows my data nicely."

Same goal. Wildly different outcomes.

The programming language is English now. But here's the uncomfortable truth most people aren't talking about. Most of us were never taught how to write well, let alone how to write precisely enough to steer a probabilistic language model.

We spent years learning syntax, algorithms, design patterns. Nobody taught us how to think in prompts.

The Prompt Is the New Code

I see this every day in my work as a data engineer.

Junior developers who can write clean Python struggle to get useful output from AI assistants. Not because the AI is bad, but because their prompts are vague, contradictory, or missing context that seems obvious to them but isn't obvious to the model at all.

Meanwhile, people with zero coding background but strong communication skills are building functional applications by just describing what they want clearly.

That's the inversion. The technical skill that matters most right now isn't loops or recursion or object-oriented design. It's clarity of thought expressed in natural language.

And that's weird. That's really weird for those of us who spent years mastering tools that are now becoming abstraction layers.

So What About 2027?

If 2026 is the year of English, what comes next?

Here's my guess. By 2027, the language won't be English anymore. It'll be intent.

You won't need to describe what you want in detail. The system will understand your intent from minimal context, your past behavior, your project structure, your team's patterns. You'll think "make that thing I was working on yesterday better" and it'll just happen.

Sounds great, right? Maybe.

But here's what worries me. When the distance between your intent and the output shrinks to almost nothing, what happens to understanding? When you don't have to think through the steps, when you don't have to break the problem down yourself, when the translation layer disappears, do you actually understand what you built?

Or are you just approving things you don't fully comprehend?

There's a difference between a senior engineer who uses AI to move faster and someone who relies on AI to think for them. The first one is augmented. The second one is replaced, they just don't know it yet.

The Uncomfortable Question

The real problem in 2026 isn't "how do I write good prompts?" That's a skill you can learn.

The real problem is: are you still thinking, or are you just approving?

Because the line between "I used AI to build this" and "AI built this and I said yes" is getting thinner every month. And somewhere around 2027, it might disappear entirely.

I don't have a clean answer here. Just a question that's been sitting with me.

As AI gets better at understanding us, are we getting worse at understanding what we're actually asking for?

Would love to hear what you think. Where does this go? Is intent-based development the dream, or should we be a little scared of how fast we're handing over the steering wheel?

Drop your thoughts below.

Direct Dive into the Data: A Beginner's Guide to Getting Started

Gilder Miller — Thu, 30 Apr 2026 13:46:12 +0000

I remember when I first heard the term Data Engineering. I nodded along like I knew what it meant, then immediately Googled it under the table.

If you're a software beginner, the data world probably feels the same way. Data Science, Data Analysis, Data Engineering... they all sound like secret societies with complicated handshakes and mandatory PhDs.

But here's the thing: data work is just work. Messy, frustrating, occasionally surprising work. And you can start way sooner than you think.

Why Data Feels So Intimidating
Let's be real. The data community has a problem.

You look at job postings and they want PySpark, Hadoop, Kafka, Airflow, dbt, Snowflake, and probably a sacrifice to the cloud gods. Meanwhile, you're just trying to figure out how to open a CSV file without Excel freezing.

But those fancy tools? They all do three simple things. Move data from here to there. Change data into something useful. Give data to people who need it.

That's the whole game. Everything else is just how you do it.

Stop Planning, Start Diving
The biggest mistake beginners make is trying to find the perfect learning path. You know the one. It starts with a linear algebra textbook, moves through a statistics course, somehow involves six months of Python tutorials, and ends with you giving up before you ever touch actual data.

Skip all that.

Here's what worked for me, and what I recommend to anyone starting out.

Pick a dataset you actually care about.
Not some boring sample dataset about iris flowers. Find something that makes you curious. Your Spotify listening history. Your city's restaurant inspection scores. Your favorite game's player stats. Your own screen time data.

Why? Because curiosity starts with a question that bugs you. Why do I always skip songs after 30 seconds? is a real data question. Which restaurants keep failing inspections? matters to someone.

Download it. Open it. Stare at it until it stops being scary.

Break something on purpose.
Don't read the documentation first. Don't watch a four-hour tutorial where someone explains pandas for the 400th time.

Open a Jupyter notebook or a Python script and just try. Load the data. Count the rows. Find the weirdest value.

You will get errors. !Good!. Errors teach you more than tutorials ever will.

Ask one dumb question.
Not What insights can I derive from this dataset? That's not a question. That's what managers say in meetings.

Ask something specific and slightly embarrassing. How many times did I listen to the same song in a row? What's the worst-rated restaurant that's still open? Which day of the week am I most productive?

One question. One answer. One tiny win that makes you want to keep going.

Share your mess.

This is where most people freeze. They think their analysis needs to be polished, their code needs to be clean, their charts need to be beautiful.

It doesn't. Share the messy version. The one with TODO comments everywhere. The one where your chart labels overlap. The one where you're not sure if your conclusion is right.

Post it on Dev.to. Put it on GitHub. Tweet it. Because someone will correct your mistake, and that's free learning. Someone will relate to your struggle, and that's community. Someone will build on your work, and that's collaboration.

The Three Flavors of Data Work
Once you start, you'll notice the field splits into different paths. Here's the honest breakdown from someone who's been around the block.

Data Analysis is for the what happened people. You'll use SQL, Excel, some Python or R, and visualization tools like Tableau. It's the easiest entry point, but the hardest to show real impact. And you'll spend most of your time cleaning data. Like, 80% of it. Get comfortable with that.

Data Science is for the what if people. You'll use Python or R, scikit-learn, statistics, and a lot of trial and error. The math can be intimidating, but you don't need to be a genius. Your models will be wrong more than they're right, and that's completely normal.

Data Engineering is for the how do we make this work reliably people. This is where I spend most of my time. You'll use Python, SQL, cloud platforms, Airflow, dbt, and whatever tool gets the job done. The learning curve is steeper, but the demand is huge. Fair warning though: you're the plumber. Nobody notices you until something breaks.

You don't have to pick one forever. Start anywhere. The paths cross constantly.

Stuff I Wish I Knew Earlier
SQL is your best friend. Learn it before anything fancy. It's the language of data. Every data job uses it. Every single one. I use SQL almost every day, and I've been doing this for years.

The cloud isn't scary. AWS and GCP have free tiers. Spin up a database. Break it. Delete it. Do it again. The fear of cloud platforms is way worse than the platforms themselves.

Documentation is a superpower. Write down what you did, even if it's just for yourself. Future you will thank present you. Trust me on this one.

"Production" isn't a dirty word. There's a big gap between "works in my notebook" and "works when I'm not watching." Bridging that gap is where the real learning happens. It's also where the jobs are.

Your software background is an advantage. If you're coming from software, you already understand version control, testing, and system design. Most data people don't have that foundation. That's your edge. Use it.

Your First Week Plan
If you want a concrete starting point, here's what I'd suggest.

Day one, download a dataset that interests you. Fifteen minutes, tops. Day two, load it into Python or a SQL database. Day three, ask and answer one question about it. Day four, make one chart, even if it's ugly. Day five, share what you found somewhere public. Day six, read someone else's data project and leave a comment. Day seven, start your second project, slightly harder than the first.

That's maybe five hours total. Less than a weekend of your time.

The Real Talk
Data work isn't glamorous. Your pipeline will break at 2 AM. Your model will confidently predict complete nonsense. Your stakeholder will ask for "just one more metric" for the fifth time this week.

But there are moments that make it worth it. When your data pipeline finally runs clean. When your model catches something no one else saw. When your analysis actually changes a real decision.

You don't need permission to start. You don't need a degree. You don't need to know everything before you begin.

You just need to dive in.

The data's waiting.

🤖Being Interviewed by a Robot: My Honest Take

Gilder Miller — Thu, 30 Apr 2026 03:12:14 +0000

You click the link. The camera turns on. A voice says hello.
But nobody is there. It is just an AI interviewer.
Welcome! and Welcome to the most awkward half hour of your life.

🧱Talking to a Wall That Talks Back
The weirdest part is the silence. When you finish answering a question, there is no nod. No smile. No "tell me more about that." Just a pause, then the next question.

You start overthinking everything. Did I ramble? Was I making sense? Without human feedback, you feel like you are shouting into a void.

🙁It Messes With Your Head
I noticed myself changing how I spoke. With a real person, I tell stories. I joke around. I am myself.

With the AI, I caught myself listing facts like a resume. Speaking in neat little bullet points. I literally became a robot to talk to a robot. It was depressing.

🫡Sure, It Has Perks

🚫No judgment about your hoodie
🗓️No scheduling nightmares
🧠No bias

But you lose so much. You cannot ask real questions. You cannot read the company vibe. You walk away with zero clue if you did well.

⏰The Waiting Sucks
After a normal interview, you have a gut feeling. You connected, or you did not.

After an AI interview, you just wait. Did the algorithm like my keywords? Did I speak at the right pace? You are left guessing what a machine thought of you.

💡Be Human Anyway
If you have to do one of these, just be yourself. Tell your messy, human stories. Do not optimize for the algorithm. The companies worth joining will have real people review your actual answers, not just a computer score.

Have you done one of these yet? Did it feel as weird as I think it does? Let me know below.🤷🏻‍♂️

When Your Data Pipeline Becomes a Coffee Pipeline

Gilder Miller — Wed, 29 Apr 2026 01:32:53 +0000

I'm at my favourite coffee shop, laptop open, terminal humming. The barista asks what I'm working on. "Just setting up a real-time data pipeline," I say, trying to sound casual.

She nods knowingly. "Like for... coffee orders?"

I laugh. "No, for business intelligence. Using Python, PostgreSQL, and AWS Lambda to process streaming data."

Next thing I know, I've been roped into building their new "smart ordering system." Suddenly my data pipeline is literally a coffee pipeline. My Jenkins CI/CD is now triggering espresso shots. My MongoDB is storing latte preferences. My GCP functions are calculating optimal milk-to-coffee ratios.

Three weeks in, the system works perfectly. Too perfectly. It's now predicting when customers will arrive based on their previous order patterns. The barista calls it "creepy but helpful." I call it an unintended consequence of saying "yes" too quickly.

How would you handle it? If you built something that worked too
well in an unexpected context, would you embrace it or pull back?

Your Data Doppelgänger is Already Here.

Gilder Miller — Sun, 26 Apr 2026 21:41:25 +0000

You're sitting in a café, talking to a friend about maybe, just maybe, taking up gardening. You haven't Googled it. You haven't liked a single plant picture on Instagram. The next day, your feed is a lush jungle of ads for potting soil, ergonomic trowels, and beginner-friendly tomato plants.

Spooky, right? Our first instinct is to think our devices are eavesdropping on us. But the truth is both more complex and, in a way, more invasive.

What's actually happening is that data science has moved on from simply tracking your clicks. It's now in the business of creating a data **doppelgänger**—a detailed, predictive model of you.

These AI models are voracious. They don't just care about what you do online. They're obsessed with the patterns of how you live. They analyze your location data to know you drive past a specific gardening store every Tuesday. They note you linger on a friend's post about their new balcony garden. They see you're part of a demographic that's recently shown a spike in home improvement.

Using a technique called behavioral clustering, the system then finds thousands of other users who match this pattern. It creates a digital "you" and places it in a cluster with all your data twins. When enough people in that cluster suddenly buy gardening supplies, the model's conclusion is simple: "You're next."

My take is that this convenience is a Trojan Horse. We happily trade the raw data of our lives for a smoother, more "magical" user experience. But the real issue isn't just that the AI is smart; it's that it's a complete black box. We can't see inside it. We don't know what assumptions it's making or which of our habits it's weighing most heavily.

This creates a ghost in our machine—a silent, predictive entity that knows our habits and desires, sometimes even before we do. It's not just showing us ads; it's subtly shaping our choices by curating the reality we see, one personalized suggestion at a time. And that's a power we should be a lot more curious about.