Episode Summary
Executive Summary: Chris Lattner discusses his post-Apple career path through Tesla, Google, and SciFive, focusing on Swift’s role in machine learning, Python interoperability, MLIR, and the future of language design. The conversation also covers RISC-V, chip-design tooling, ABI stability, SwiftUI/Combine, and how COVID-19 reshaped work, schooling, and community events like WWDC.
Main Topics: Chris Lattner’s career path after Apple: Lattner explains why he moved from Tesla to Google to SciFive, emphasizing choosing hard, meaningful problems over easy roles and learning from each environment. Swift for TensorFlow and machine learning: He details how Swift was adapted for differentiable programming, better deployment, and Python interoperability to make ML more productive and less fragmented. MLIR and compiler infrastructure: Lattner describes MLIR as a general compiler construction toolkit intended to unify repeated compiler problems across domains like ML, Swift, and accelerators. SciFive, RISC-V, and chip-design tooling: He explains SciFive’s mission to build tooling and infrastructure for custom silicon and RISC-V-based designs, improving chip development workflows. Swift language evolution and future features: The discussion covers ABI stability, ownership, variadic generics, concurrency, regular expressions, and concerns about adding too much syntactic sugar too soon. SwiftUI, Combine, and the balance of APIs vs language: The speakers debate SwiftUI’s strengths and rough edges, how reactive programming fits into Swift, and whether new frameworks are outpacing language standardization. COVID-19, WWDC, and remote life: The end of the episode shifts to the pandemic’s impact on work, schooling, teachers, family life, and the likely future of WWDC and in-person conferences.
Key Arguments: Swift should be compelling enough that people want to use it, rather than being forced into it. Machine learning tooling improves when the language itself supports differentiable programming and interoperability, instead of relying on layered hacks. Python interoperability in Swift is valuable because it lets users access the ecosystem while gradually moving toward safer, more expressive APIs. MLIR matters because compiler infrastructure should be reusable across many domains instead of rebuilt from scratch for each project. SwiftUI shows the power of Swift’s type system, but the language should focus on big foundational features before adding more syntactic polish. RISC-V is important because open, non-proprietary instruction sets reduce vendor lock-in and enable customizable silicon. WWDC and conferences may need to evolve into more scalable, less exclusive formats, even if the loss of in-person community is significant. The pandemic exposed the importance of teachers and the difficulty of remote schooling for families and educators.
Data Points: Swift for TensorFlow weekly public video chat: roughly weekly - Lattner says the team held recurring public calls to discuss technical work and community feedback Python module size in Swift: about 1,200 lines of Swift code - He notes the interoperability layer is surprisingly small MLIR age: about 18 months - At the time of the conversation, he says MLIR is still very young but already widely adopted Swift age: about 5.5 years - Used to argue that Swift is still early in its evolution Swift 5.2 timing: shipped yesterday (relative to recording) - Referenced while discussing dynamic member lookup and callable features Google Swift/TensorFlow public release cadence: new release a week ago / two weeks ago - Lattner references recent project releases Tesla autopilot team leaders: several leaders with short time horizons - Used to characterize instability before his departure WWDC demand vs supply: too much demand and too little supply - Explains why in-person labs and tickets are scarce Apple device management offer: first 3 devices free - Jamf Now sponsor mention Linode promo credit: $20 credit with ATP2020 - Sponsor offer for new accounts
Pivotal Quotes: "My goal is for Swift to be such an amazing thing that people want to use it." — Chris Lattner: He explains his philosophy for spreading Swift beyond Apple and into machine learning, servers, and other domains "We can take this idea of taking a function and calculating its derivative. Build just that piece into the Swift language." — Chris Lattner: He summarizes the differentiable-programming idea behind Swift for TensorFlow "I think that if they do WWDC virtually this year and it sucks, then it's probably going to go physical again." — Chris Lattner: His view on whether WWDC would remain online after the pandemic
Implications: Swift’s future depends on finishing foundational language features and proving itself across more ecosystems. MLIR and RISC-V suggest a broader shift toward reusable, open tooling. COVID may permanently change how tech communities, conferences, and education operate.