Alright, buckle up, because the future of coding is becoming more expressive and creative! After listening to the discussion from Lex Fridman with the Cursor team, here’s where we’re headed:
What’s Possible Today:
- AI-Powered Code Editors: Tools like Cursor are already augmenting how developers write code, offering autocomplete on steroids, editing capabilities, and intelligent suggestions.
- Custom Models: Cursor uses a combination of frontier and custom models, specializing in the nuanced needs of code editing like understanding rough human intent and creating accurate diffs.
- Focus on Speed: Caching strategies, speculative edits, and optimized algorithms ensure the AI assistance doesn’t slow down the development process. It must be fast or it is not fun!
Where We’re Headed in the Future:
- AI as a Colleague: Picture AI not just as a code generator, but as a partner that can understand code, debug it, and suggest improvements.
- Context-Aware Programming: AI will be able to infer context from the entire codebase, imported libraries, and even the programmer’s recent actions, leading to more relevant suggestions.
- Abstraction Layering: Programmers might work at different abstraction levels, editing pseudocode or high-level designs and letting AI translate it into formal code.
- Code as Intent: Programming shifts from tedious text editing to expressing high-level intent, with AI handling the boilerplate and low-entropy tasks.
- Agents as Helpers: AI agents will automate repetitive tasks like setting up environments, solving dependency conflicts, and even running background tests.
- Formal Verification: Models help write specification and then verify code for correctness, and perhaps even prove that it’s aligned with human intent.
- Branching everything: Branching filesystems, the database, even the write ahead log to increase speed and reduce errors.
- Homomorphic Encryption: Solve data privacy concerns by keeping user data private while using models for inference.
- The Field’s Metal for AI: If AI can prove P doesn’t equal NP, then it will win the Fields medal.
Why Coding is Still Necessary:
- Human in the Driver’s Seat: Despite AI advancements, human programmers will remain essential for creative problem-solving, making critical design decisions, and guiding the AI.
- Expressing Intent: AI can help write code, but humans are still needed to express the desired functionality and overall architecture.
- Evolving Skills: Programming skills will shift from rote memorization to creativity, design thinking, and the ability to collaborate with AI effectively.
- Unspecified needs: Interview type problems have well specified needs unlike real coding where humans add half-broken English.
- The Vibe: there is some component of programming that benchmarks don’t solve which is “the vibe” and for this you will always need humans!
In a nutshell: Coding isn’t going away, it’s evolving. AI is set to become an indispensable tool, amplifying human capabilities and making programming more accessible and fun. The future programmer will be a hybrid—a blend of human creativity and AI assistance, capable of building systems that were previously unimaginable.