The Big Picture
In these conversations, software engineering is mostly a story about AI changing the job without removing the need for skilled people. The Cursor team describes the code editor as a souped-up word processor that understands code's structure, and expects it to change a lot over the next decade [2]. Their tool already tries to predict not just the next characters but the next edit and the next place you'll jump [2]. Dylan Patel expects software engineering to be the first domain where AI agents succeed, because code can be checked, noting that SWE-bench scores rose from about 4% to 60% in a year [3].
Most voices here stop short of predicting the end of programmers. ThePrimeagen (Michael Paulson) tells Lex Fridman he doesn't expect programmers to be replaced anytime soon, noting that no computing advance so far has reduced jobs [1]. Nathan Lambert expects the job market to flatten rather than collapse suddenly [3]. The Cursor team believes humans should stay in control [2]. Mark Zuckerberg, speaking with Joe Rogan, takes the widest view: like the shift from most people farming to about 2% farming, AI will free people for creative work so that "we'll all just kind of have superpowers" [4].
Where the Experts Agree
- AI is becoming part of everyday programming. Fridman says AI speeds up tasks like wiring up the YouTube API and makes programming more fun and less lonely [1]. A Stack Overflow survey the two discuss shows most developers use or plan to use AI [1]. The Cursor team is building its editor around AI features such as predicting the next edit, custom models and agents [2].
- Deep skills still matter. Prime and Fridman both agree AI is poor at hard debugging and that deep, hard skills will remain essential for steering it [1]. Patel advises programmers to become domain experts and ride the rising tide [3].
- Humans should supervise. Fridman advises programmers to embrace supervising AI [3], and the Cursor team argues humans should stay in control of how programming changes [2]. Zuckerberg says guardrails matter, especially for new reasoning models [4].
- Code is a good fit for AI because it can be checked. Patel's reason for expecting agents to succeed in software first is that code is verifiable [3]. The Cursor team discusses bug finding, dangerous code and formal verification as part of their agent work [2].
Where They Differ
- How much to trust the output. Prime is skeptical: only about 2.7% of developers in the survey highly trust AI output [1]. Fridman argues the opposite side, that working with AI is a skill worth building, including learning what context the model is missing [1]. The difference is partly about the task. Both agree AI struggles with hard debugging, while Fridman's examples are routine integration work [1].
- How far agents can go. Patel is optimistic for software specifically, citing rapid benchmark gains [3]. Yet he also compares chaining tasks to semiconductor yields, where even high reliability per step compounds into failure over many steps, and Lambert calls the term "agents" overblown [3]. The Cursor team notes that benchmarks can mislead [2].
- What natural-language programming becomes. Prime suspects programming in plain language will evolve into another precise language [1], which implies the core skill of precise specification won't disappear. The Cursor team's vision focuses on the editor anticipating what the programmer wants [2].
- Open versus closed platforms. Zuckerberg argues AI won't become a god-like power if it's widely distributed, which is why Meta champions open source [4]. He also criticizes Apple for squeezing developers with its 30% fee [4]. The Cursor team raises a related concern: the privacy risks of centralized AI [2].
- The economics of software. Patel says cheap software could let every company build custom tools instead of buying SaaS, as is common in China, and says businesses built on GPT-3- or GPT-4-level capability are effectively dead as costs collapse [3]. That points to more software being written, which fits Prime's view that computing advances haven't reduced jobs [1].
Practical Takeaways
- Learn to work with AI tools. Treat prompting and giving the model the context it's missing as a skill worth building, as Fridman argues [1].
- Keep sharpening hard skills. Debugging and deep understanding remain the areas where AI is weakest and where you'll be needed to steer it [1].
- Verify what AI produces. Given how few developers highly trust AI output [1], and how errors compound over multi-step agent tasks [3], review and test generated code.
- Become a domain expert. Patel's advice is to know a field deeply and ride the rising tide [3].
- Practice supervising, not just typing. Fridman's advice is to embrace supervising AI [3], and the Cursor team wants humans to stay in control [2].
- Be skeptical of benchmarks. The Cursor team warns they can mislead, so judge tools by how they perform on your own work [2].
- Protect your life outside code. Prime tells people in their twenties to find the kind of programming that brings them back every day, but not to sacrifice their health, relationships or morals for the next big startup, as he did [1].
Sources
- The Primeagen: Programming, AI, ADHD, Productivity, Addiction, and God · Lex Fridman Podcast
- Cursor Team: Future of Programming with AI · Lex Fridman Podcast
- DeepSeek, China, OpenAI, NVIDIA, xAI, TSMC, Stargate, and AI Megaclusters · Lex Fridman Podcast
- Mark Zuckerberg · The Joe Rogan Experience
