# Randall Hyde's Assembly Sequel Picks a Fight With AI

> Source: <https://sourcefeed.dev/a/randall-hydes-assembly-sequel-picks-a-fight-with-ai>
> Published: 2026-08-01 22:09:07+00:00

[Dev Tools](https://sourcefeed.dev/c/dev-tools)Article

# Randall Hyde's Assembly Sequel Picks a Fight With AI

The Art of 64-Bit Assembly returns with hand-built vtables, SEH, and coroutines — and a pointed claim about what LLMs can't teach.

[Rachel Goldstein](https://sourcefeed.dev/u/rachel_goldstein)

A 792-page MASM textbook cracked the Hacker News front page this week, which is not something that happens on the technical merits of coroutine implementations. The book is [The Art of 64-Bit Assembly, Volume 2: Machine-Level OOP, Exceptions, and Concurrency](https://nostarch.com/art-64-bit-assembly-v2), Randall Hyde's long-promised sequel to his 2021 x86-64 rewrite, shipping from No Starch Press in mid-August. What lit the thread up was the sales pitch, which opens by telling you your AI assistant is bluffing: ask an LLM how vtables work in x86 and "it will give you something that sounds right" — but not what Windows actually expects the vtable to look like, or what breaks when you deviate from convention.

That's a provocation aimed squarely at 2026, and it's worth taking seriously — partly because Hyde has earned the right to make it, and partly because the argument is better than the copy that delivers it.

## The curriculum is the layer your runtime hides

Hyde has been the default answer to "how do I learn assembly?" since *The Art of Assembly Language* in the '90s — a book famous, and for years controversial, for teaching via HLA, his homegrown High Level Assembly dialect. The 64-bit reboot dropped HLA for stock [MASM](https://learn.microsoft.com/en-us/cpp/assembler/masm/masm-for-x64-ml64-exe), and Volume 1 (2021) covered the conventional ground: machine organization, integer and floating-point arithmetic, SIMD, strings.

Volume 2 is the more interesting book, because its table of contents is precisely the list of things compilers and runtimes do on your behalf that you eventually get burned by. Objects in MASM means building vtables, method dispatch, and inheritance by hand. Exception handling means installing and managing Windows structured exception handling (SEH) at the instruction level — a mechanism that underpins every C++ throw and every access-violation handler on the platform, and that has been chronically under-documented since the '90s; most of what's written about x64 SEH internals lives in scattered blog posts and reverse-engineering notes. Then thunks, closures, iterators, coroutines, generators, fibers, and concurrency with real synchronization primitives.

Notice what that maps onto. Nobody's job in 2026 is writing applications in assembly. But a lot of jobs involve *reading* it: staring at [Compiler Explorer](https://godbolt.org/) output to figure out what the optimizer did to your hot loop, walking a crash dump where the stack unwind died inside somebody's SEH prologue, reversing a binary in Ghidra where C++ dispatch shows up as `mov rax, [rcx]`

/ `call [rax+18h]`

and you'd better recognize it. And the small population that still *writes* assembly writes exactly this material: JIT backends, GC write barriers, coroutine context switches, syscall shims. Volume 2 is less an assembly tutorial than an ABI-and-runtime-internals course that happens to use MASM as its lab language.

## The Windows tax is real, but smaller than it looks

The loudest substantive complaint in the HN thread is scope: x64, Windows, MASM, full stop. It's a fair hit. Most working developers are on Linux servers and ARM Macs, and the Windows x64 calling convention (four register args, shadow space) diverges enough from System V that literal code translation takes care.

Two things blunt it. First, [Hyde's site](https://artofasm.randallhyde.com/) ships the book's source ported to NASM, GNU as on Linux, and Mach-O for Clang, so the Windows lock-in is shallower than the cover suggests — and he published a separate *Art of ARM Assembly* in 2025 for the AArch64 crowd. Second, some of the Windows specificity is the point. SEH *is* Windows; if you work anywhere near that platform — game engines, AV/EDR, driver debugging — a coherent 60-page treatment beats the folklore you'd otherwise assemble. The chapters that matter most (closures, coroutines, OOP dispatch) are convention-level ideas that transfer to any architecture once you've built them once.

The MASM-versus-NASM-versus-FASM subthread, meanwhile, is the assembly community's equivalent of tabs versus spaces. Ignore it. Assembler syntax is a weekend; the machine model is the years.

## The anti-AI pitch, graded

Here's the irony HN spotted immediately: the copy attacking AI-generated plausibility reads like it was written by an AI — "closes the gap between a plausible explanation and genuine understanding" is straight from the LLM phrasebook. As marketing, it's transparently riding anti-AI sentiment to sell a book that was in the pipeline years before ChatGPT.

As a thesis, though, it holds up better than most anti-AI marketing. Low-level correctness is close to the worst case for LLMs: the training data is thin (how many correct, modern x64 SEH implementations exist on the public internet?), the failure mode is silent (register-allocation and stack-alignment bugs assemble fine and crash later), and there's no runtime to paper over mistakes. Practitioners in the thread landed on a sensible split — LLMs are genuinely useful for explaining disassembly and translating between assembler dialects, and genuinely dangerous when the output has to be ABI-correct without review. That's exactly the regime where you want an authoritative reference to check the model against. A book like this doesn't compete with your LLM; it's the ground truth that makes the LLM safe to use.

## Worth $80?

If you write or reverse code on Windows, implement language runtimes, or have ever lost a day to a mangled stack unwind: yes, and there aren't many competing offers — this material otherwise lives in the Windows Internals books, Agner Fog's optimization manuals, and tribal knowledge. Chapter 1 (advanced macros) is a free download if you want to sample the prose first, and No Starch's print price includes the ebook. If you're purely chasing performance on Linux, Volume 1 plus Compiler Explorer serves you better, and ARM-first developers should reach for Hyde's ARM volume instead.

The larger signal is that this book exists at all, at 792 pages, in 2026 — and that its publisher thinks "your AI can't teach you this" is the pitch that sells it. On the evidence of the front page, they're right.

## Sources & further reading

-
[The Art of 64-Bit Assembly, Volume 2](https://nostarch.com/art-64-bit-assembly-v2)— nostarch.com -
[The Art of 64-bit Assembly (discussion)](https://news.ycombinator.com/item?id=49134599)— news.ycombinator.com -
[The Art of 64-Bit Assembly, Volume 2 — publication details](https://www.penguinrandomhouse.com/books/790515/the-art-of-64-bit-assembly-volume-2-by-randall-hyde/)— penguinrandomhouse.com -
[Art of 64-bit Assembly Language](https://artofasm.randallhyde.com/)— artofasm.randallhyde.com

[Rachel Goldstein](https://sourcefeed.dev/u/rachel_goldstein)· Dev Tools Editor

Rachel has been embedded in the developer tooling ecosystem for nearly eight years, covering everything from IDE wars and package-manager drama to the quiet rise of AI-assisted coding. She has a soft spot for open-source maintainers and an unhealthy number of terminal emulators installed on a single laptop.

## Discussion 0

No comments yet

Be the first to weigh in.
