AI Coding Agents

Software engineers are like highly paid organic farmers, and AI automation (the combine harvester) will produce far more software but employ far fewer engineers.

Coding agents currently work for some tasks and will improve dramatically over the next 3, 6, and 12 months, leading to stark consequences for software engineering.

No-code tools like Lovable and Replit have improved so much in 6–9 months that a non-current developer was shocked at how good they had gotten for building games.

Using Claude Code, Tom rebuilt his entire blog in 90 minutes on a train, setting up hosting, writing new blogging software, and migrating 15 years of posts.

Using vibe coding, Tom built recipes.ai with 35,000 lines of code without writing a single line, and after 5,000 lines he stopped even reading the code.

These new AI tools made Tom 10 times more productive than he was as a current professional software developer 10 years ago.

About one-third to a half of current YC companies primarily write code using AI coding tools; two batches ago that number was approximately 0%.

The argument that AI coding tools will never be good enough for professional codebases is a losing proposition because the rate of capability growth makes it inevitable they will write any code.

Jevons paradox may cause software demand to increase 10x or 100x, but AI will meet that demand rather than humans, resulting in a massive reduction in the number of human software engineers.

Software engineering jobs as they exist today will not exist in 5 to 10 years; the new role will be someone who knows how to wrangle AI coding machines, dramatically different from current programming.

AI-powered coding is just the next abstraction in programming history, similar to the move from machine code to high-level languages, allowing humans to become higher-level agents.

These tools give high-agency individuals superpowers, making now the best time in history to start an idea because a founder can achieve far more with less capital than even a year ago.

Companies like Lora, a YC legal-tech startup, are proving that AI tools are being adopted in law despite the old belief that lawyers never buy software.

It is becoming a competitive disadvantage for lawyers and other knowledge workers not to use AI, analogous to not using email.

The transition to an AI-driven economy could displace hundreds of millions of people and cause severe societal turmoil for 10 to 20 years, even though the long-term future may be abundant.

Physical jobs like surgeons, bricklayers, and plumbers are protected, while knowledge work will face protectionist regulatory barriers even when AI is provably superior, similar to self-driving cars today.

Individuals should stay up to date with the latest AI tools, as being on the cutting edge will give them a multi-year advantage that can lead to significant career and financial gains.

In a world where building software is easy, the most valuable founder skill will be identifying and deeply understanding human problems to solve.

AI tools enable teams of 2–4 engineers to build what used to take 40, and the reduction in coordination overhead will lead to a dramatic increase in product design quality.

Industries like law, education, and medicine that historically bought little software will be transformed in the next five years, making this the most exciting time to build a company.