Frontpage / Guides / The Ultimate Guide to Coding with AI
Coding

The Ultimate Guide to Coding with AI

What 36 videos from Fireship, Modern Software Engineering and Y Combinator actually say about AI writing code — where it helps, where it costs, and what it does to learning the job.

Last updated · by Jacob S. Olsen

Something unusual shows up when you count what developers are actually talking about. Tech Feed Watch has covered 36 videos on coding, from Fireship, Modern Software Engineering, Y Combinator, NetworkChuck and others.

16 of them are about careers, skills and learning. More than any specific technology. Eleven are about AI writing code.

That ratio is the story. The dominant question in software right now is not which framework to use. It is what the job becomes when the typing is handled for you — and whether the path into it still works.

Where AI genuinely helps

The archive is not sceptical about the tools. It is specific about when they work.

Code you could have written yourself, faster. Boilerplate, a familiar pattern in an unfamiliar language, the fiddly bits of a form. You know what correct looks like, so you catch the wrong version immediately. This is where nearly all the reported gain comes from.

Understanding code someone else wrote. Asking what a function does, or why a build fails, is often faster than reading it cold. The answer is a starting point rather than a verdict, but it is a good starting point.

The first draft of something tedious. Tests, a migration script, converting data from one shape to another. Work that is well-defined and boring is exactly what these tools are good at.

Getting unstuck. Not the answer — a direction. Several sources describe this as the underrated one, because being stuck costs far more time than typing does.

Where it costs

Code you do not understand. This is the whole risk in one sentence. Generated code that appears to work is not the same as code that works, and the difference surfaces later, usually under pressure. The archive has a name for accepting it anyway: vibe coding.

Security. These models reproduce patterns from what they were trained on, including the insecure ones, and they do it with total confidence. Generated code deserves the same review you would give code from a stranger — because that is what it is.

Confident wrongness. A model does not know when it is guessing. It will invent a function that does not exist, an option that was never in the library, an approach that has been deprecated for years, and describe all of it in the same tone as the parts that are right.

Debt that arrives quietly. Faster shipping with less understanding is a trade, not a win. It comes due when something breaks and nobody in the room can explain the code — including whoever accepted it.

Working with it sensibly

Six habits come up repeatedly across the sources.

Read everything before you accept it. If you cannot say what a block does, you are not reviewing it, you are hoping.

Give it the context it needs. Most poor output is a poor question. The context window is finite, so what you leave out matters as much as what you put in.

Ask for the approach before the code. A wrong plan produces a hundred lines of wrong code, and the plan is faster to check.

Keep the tests yours. Tests generated from the same misunderstanding as the code will confirm the misunderstanding.

Use it hardest where you are strongest. That is the opposite of the instinct, and it is what the archive supports: you catch the mistakes fastest in your own territory.

Commit in small pieces. When something breaks, you want to know which suggestion caused it.

The career question

Sixteen of the 36 videos are about careers, skills and learning — more than any technical subject. Worth saying plainly that the people answering have interests in the answer, on both sides.

The pattern the archive actually supports is narrower than either headline.

The typing was never the hard part. Deciding what to build, breaking it into pieces, noticing that something is subtly wrong — that was always the job. Generating syntax is the part that got cheap.

Which makes review the core skill. If a model produces code faster than you can read it, the reading becomes the bottleneck. The developers described as getting the most out of these tools are the ones who read fastest and trust least.

The path in got stranger, not closed. The traditional route was to write a lot of simple code until the patterns became instinct. If that code is now generated, the practice has to come from somewhere else: reading, breaking things on purpose, and building without help until it hurts. Several sources worry about this openly, and none of them claim to have solved it.

Specifying is becoming its own skill. Describing what you want precisely enough that a machine produces it is close to programming, just in a different language. That is not a smaller job — it is the same judgement with different typing.

A note for people learning right now

If you are starting out, the archive points at one uncomfortable piece of advice: turn the assistant off sometimes.

Not out of principle. Because the skill you need is recognising when something is wrong, and you cannot build that by watching correct-looking code appear. You build it by getting it wrong yourself and finding out why.

Use the tools for work you understand. Do the learning without them. It is slower, and that is the point — the slowness is where the understanding comes from.

Small things that save time

Not everything needs a model. A few of these come up constantly in development work, and they run entirely in your browser:

Reading a machine-generated response that arrived as one unbroken line — the JSON formatter. Encoding or decoding a token, a config value or a small file — the Base64 and URL encoder. Changing one recurring string across a block of text — find and replace. Splitting a file too large to send — split a file, and the pieces rejoin with cat or copy /b without any tool at all.

The short version

The loudest question in software is not technical. Sixteen of 36 videos are about careers and skills, eleven about AI writing code.

The tools genuinely help with work you already understand, and genuinely cost you when they are used on work you do not. Review became the core skill, because generation stopped being the bottleneck. And if you are learning, the fastest route still runs through doing it the slow way often enough to recognise wrong when you see it.

Every article behind this guide links to its original video, with the creator credited. The Coding tag has all 36.


This guide draws on 36 videos covered on Tech Feed Watch, from channels including Fireship, Modern Software Engineering, Y Combinator and NetworkChuck. The counts quoted come from the archive itself, not from an industry survey. Written and maintained by Jacob S. Olsen. If something here is wrong, the corrections policy explains how to tell me.

The numbers in this guide are a snapshot of the archive as of July 2026; the archive itself keeps growing.

Articles in this guide

What AI Coding Tools Mean for Healthcare Documentation

What AI Coding Tools Mean for Healthcare Documentation

Coding is Not Dead, but Rapidly Transforming by 2026

Coding is Not Dead, but Rapidly Transforming by 2026

How Coding Harnesses Transform LLMs into Agentic Systems

How Coding Harnesses Transform LLMs into Agentic Systems

How Can AI Help Game Development for Solo Developers?

How Can AI Help Game Development for Solo Developers?

Do AI Code Agents on Gemini Platform Reshape Developer Roles?

Do AI Code Agents on Gemini Platform Reshape Developer Roles?

What Is the Primary Goal of AI Agents in Software Development

What Is the Primary Goal of AI Agents in Software Development

Are AI coding skills transforming software development?

Are AI coding skills transforming software development?

GStack: Garry Tan's Toolkit Redefines AI Software Development

GStack: Garry Tan's Toolkit Redefines AI Software Development

Parallel AI Coding Agents: Git Worktrees and Docker Prevent Conflicts

Parallel AI Coding Agents: Git Worktrees and Docker Prevent Conflicts

How Does TDD AI Improve Software Design?

How Does TDD AI Improve Software Design?

How Good Are AI Coding Tools for Software Architecture?

How Good Are AI Coding Tools for Software Architecture?

Open-Source AI Tools Empower Developers for Faster AI Apps

Open-Source AI Tools Empower Developers for Faster AI Apps

Rust SQLite Turso Improves Database Memory Safety and Concurrency

Rust SQLite Turso Improves Database Memory Safety and Concurrency

AI Front End Development Tools Transform Web UI Creation

AI Front End Development Tools Transform Web UI Creation

Jane Street's Custom AI: OCaml Coding Assistants Boost Productivity

Jane Street's Custom AI: OCaml Coding Assistants Boost Productivity

What Is Smart Contract Escrow on Solana, and How It Works

What Is Smart Contract Escrow on Solana, and How It Works

What Do Open Weight Models Mean for AI in 2026?

What Do Open Weight Models Mean for AI in 2026?

Quantum Computing for Developers Demands New Programming Logic

Quantum Computing for Developers Demands New Programming Logic

AI Requires Programmers to Develop Architectural Thinking Skills

AI Requires Programmers to Develop Architectural Thinking Skills

Remix Ethereum IDE Simplifies Smart Contract Coding

Remix Ethereum IDE Simplifies Smart Contract Coding

Can Compact Language Build Private Smart Contracts Midnight ZK?

Can Compact Language Build Private Smart Contracts Midnight ZK?

Project-Based Skills Build Machine Learning Careers Without Degrees

Project-Based Skills Build Machine Learning Careers Without Degrees

What AI Engineer Skills Are Needed for System Deployment?

What AI Engineer Skills Are Needed for System Deployment?

Foundational ML Neural Networks and LLMs for AI Engineering

Foundational ML Neural Networks and LLMs for AI Engineering

How to Use Etherscan to Inspect Smart Contracts and NFTs

How to Use Etherscan to Inspect Smart Contracts and NFTs

How Do Smart Contracts Work on Ethereum? Code Your First One

How Do Smart Contracts Work on Ethereum? Code Your First One

Website visibility: how HTML tags affect SEO rankings?

Website visibility: how HTML tags affect SEO rankings?

How to Use yfinance in Python for Quant Investing

How to Use yfinance in Python for Quant Investing

What Is Smart Contract Development in Web3 with AI Tools?

What Is Smart Contract Development in Web3 with AI Tools?

Technical SEO: Developers Ensure Rankings and Fast UX via Rendering

Technical SEO: Developers Ensure Rankings and Fast UX via Rendering

How to Make Claude Code Websites Look Better

How to Make Claude Code Websites Look Better

How AI Changes Lorem Ipsum Generator for WordPress Fast Design

How AI Changes Lorem Ipsum Generator for WordPress Fast Design

Frequently Asked Questions

Does AI actually make developers faster?

For code you already know how to write, usually yes. For code you do not understand, it produces something that looks finished, which is a different thing from being faster. The archive is clear that the gains are real and narrower than the marketing.

Should a beginner learn to code now?

The archive leans yes, with a change in emphasis. Typing code was never the hard part; deciding what to build and recognising when something is wrong is, and that is now the whole job rather than half of it.

What is vibe coding?

Accepting AI-generated code without fully understanding it because it appears to work. It ships quickly and creates debt that lands on whoever maintains it — often the same person a few months later.

Is AI-generated code secure?

Not by default. It reproduces patterns from its training data, including insecure ones, and it will do so with complete confidence. Generated code needs the same review as code from a stranger.

Will AI replace programmers?

The archive's clearest pattern is that the work shifts rather than disappears — from writing code to specifying, reviewing and orchestrating it. That is a real change in what the job feels like day to day.

Jacob Olsen

Jacob S. Olsen

Runs Tech Feed Watch, from Denmark

Guides are written and maintained by hand, unlike the article summaries on this site — how that works. If something here is wrong, tell me.