The most controversial rewrite in history just shipped... — Transcript

Full transcript

1. 0:00Last week, Bun, the all-in-one
2. 0:02JavaScript toolkit that was acquired by
3. 0:04Anthropic late last year, announced that
4. 0:06they spent the equivalent of $165,000
5. 0:09running 64 parallel clot agents to
6. 0:12refactor their entire codebase from Zigg
7. 0:14to Rust in just 11 days. Usually, a full
8. 0:17rewrite to Rust is the swan song of any
9. 0:20startup about to get Rick James by the
10. 0:21invisible hand of the market, but in
11. 0:23this case, it may have actually worked.
12. 0:25The millionline port fixed 128 old bugs,
13. 0:28shrunk the binary by 20%, made
14. 0:30everything a few% faster, and has
15. 0:32already been powering Claude code since
16. 0:34June without anyone noticing. But not
17. 0:36everyone is celebrating. Andrew Kelly,
18. 0:38the creator of Zigg, responded claiming
19. 0:41the benchmarks are misleading, and the
20. 0:42Zig team is actually relieved to see Bun
21. 0:45go because internally they used its
22. 0:46codebase as an example of how not to
23. 0:48write Zigg and that Jared, Bun's
24. 0:50founder, was producing SLOP long before
25. 0:53LLM existed. In today's video, we'll
26. 0:55find out how the biggest AI powered
27. 0:57rewrite in history actually went down
28. 0:59and decide who should get custody of
29. 1:00public opinion in the divorce. It is
30. 1:02July 15th, 2026, and you're watching the
31. 1:05code report. That 26 years ago, when I
32. 1:07was throwing Japan airs in the
33. 1:08warehouse, Joel Spolski wrote one of the
34. 1:10most famous blog posts in software
35. 1:12history where he claimed that rewriting
36. 1:14your codebase is the single worst
37. 1:15strategic mistake a software company can
38. 1:17make. His example was Netscape, who
39. 1:19spent three years rewriting their
40. 1:21browser while Microsoft stole their
41. 1:22women and pillaged their village. So,
42. 1:24what led Bun to fall to the temptation
43. 1:26of the single word strategic mistake
44. 1:28they could make? Well, it's actually the
45. 1:30same thing that led you to fail CS 101,
46. 1:32switch majors, and become a JavaScript
47. 1:34developer, but poor memory management.
48. 1:36With Zigg, the language Bun was
49. 1:38originally written in, you manually
50. 1:40manage memory yourself without much help
51. 1:42from the compiler. That's usually a
52. 1:43feature of a low-level language because
53. 1:45Bun embeds JavaScript core, which is
54. 1:47Safari's garbage collected JavaScript
55. 1:49engine. Half of Bun's objects were owned
56. 1:51by the garbage collector, while the
57. 1:53other half lived in manually managed Zig
58. 1:55memory, and the two had to constantly
59. 1:57manage pointers to each other, which led
60. 1:59to a codebase that looked like it had an
61. 2:01explosive diarrhea parasite. But Bun's
62. 2:03change logs are full of bugs stemming
63. 2:04from this, where code was reading memory
64. 2:06that had already been freed, or freeing
65. 2:08the same memory twice, or just never
66. 2:10freeing it at all. At one point, the dev
67. 2:12server was leaking three megabytes every
68. 2:14time it rebuilt, all because a single
69. 2:16error path forgot to clean up after
70. 2:17itself. And though memory bugs alone
71. 2:19might not have justified a rewrite,
72. 2:21there was still a bigger problem. Once
73. 2:22Anthropic acquired Bun, the majority of
74. 2:24its future code was going to be written
75. 2:26by Claude. And Zigg is famously anti-
76. 2:29AI. So much so that they refuse LLM
77. 2:32generated pull requests and they'll even
78. 2:33close your security report if you admit
79. 2:35AI found the bug. On top of that,
80. 2:37because there's very little Zig code on
81. 2:39the internet to train on, and the
82. 2:40language itself hasn't hit 1.0 yet, so
83. 2:43it keeps introducing breaking changes,
84. 2:45models just aren't very good at writing
85. 2:46it compared to more established
86. 2:48languages. And so in early May, Jared
87. 2:50decided to fix both problems at once by
88. 2:53porting all 535,000 lines of Zigg to
89. 2:56Rust, whose borrow checker would move
90. 2:58memory management into the type system,
91. 3:00which made most memory mismanagement a
92. 3:02compile time error rather than a runtime
93. 3:04error. But the most interesting part is
94. 3:06how they did it. They first had Claude
95. 3:08spend hours studying the codebase to
96. 3:10produce a porting guide. Then ran a
97. 3:12workflow that traced the lifetime of
98. 3:13every struck field into a giant
99. 3:15spreadsheet, essentially documenting
100. 3:17years of tribal knowledge about who
101. 3:19frees what and when. From there, they
102. 3:20used 64 parallel CLA agents across four
103. 3:23git workrelate
104. 3:25all, 1448 files, at one point cranking
105. 3:28out 1300 lines of Rust per minute. And
106. 3:31to keep the agents honest, every
107. 3:32implementer was paired with two
108. 3:34adversarial reviewer clouds running in
109. 3:36separate context windows whose only job
110. 3:38was to assume the code was wrong and
111. 3:40find out why. 11 days, 6,52 commits and
112. 3:44what would have been $165,000
113. 3:47in spend if they didn't own the token
114. 3:48casino later. Bun's entire test suite
115. 3:50was passing on every platform. The port
116. 3:53fixed 128 long-standing bugs. The dev
117. 3:56servers memory leak issues resolved, but
118. 3:58binaries got 20% smaller. and Andrew
119. 4:00Kelly, the creator of Zigg, lost his
120. 4:02mind. In his response, Andrew makes it
121. 4:04clear that this divorce was years in the
122. 4:06making. According to him, the Zigg team
123. 4:08spent years watching Bun embarrassed
124. 4:10their language with a codebase they
125. 4:11privately used as the example of how not
126. 4:14to write Zigg. He also takes some
127. 4:16personal swings, pointing out that Jared
128. 4:17skipped college at 18 to take Peter
129. 4:19Teal's money and Chase startup glory,
130. 4:22which Andrew blames for his beginner
131. 4:23energy and passing along secondhand
132. 4:25reports that Jared wasn't exactly a
133. 4:27great manager to work under. But buried
134. 4:29under the resentment are some legitimate
135. 4:31technical gripes. He argues the
136. 4:32performance gains mostly came from link
137. 4:34time optimization which Zig has
138. 4:36supported the whole time that the binary
139. 4:38size reduction had nothing to do with
140. 4:40Rust and that Jared conveniently left
141. 4:42out any reference to compile times which
142. 4:44Zigg almost certainly wins. So the
143. 4:46question is who gets custody of public
144. 4:48opinion? Well, probably nobody. Zigg
145. 4:50lost its most famous user. Andrew lost
146. 4:52his cool and Jared got publicly
147. 4:54diagnosed with beginner energy by a
148. 4:56compiler engineer. But if you're looking
149. 4:58to rewrite your own codebase in Rust or
150. 5:00just write better code in general, then
151. 5:02you need to know about Code Rabbit, the
152. 5:03sponsor of today's video. Writing
153. 5:05thousands of lines of code with AI feels
154. 5:07great until you have to review all that
155. 5:09unholy slop by hand. Code Rabbit Review
156. 5:11turns your pull requests into guided
157. 5:13walkthroughs you'll actually finish. It
158. 5:15groups related changes into cohorts,
159. 5:17orders them into layers, and adds AI
160. 5:19summaries so you can move through all
161. 5:21the changes quickly in the right order.
162. 5:23Instead of one giant flat diff, it feels
163. 5:25like an IDE for your pull requests. Your
164. 5:27comments and approvals still go back to
165. 5:29GitHub natively. And when it helps, the
166. 5:31Code Rabbit can generate inline diagrams
167. 5:33for your call flows, state changes and
168. 5:35schema updates. Over 100,000 open-
169. 5:38source projects like Bun and Nex.js use
170. 5:40Code Rabbit, and you can try it out for
171. 5:42free at the link below. But this has
172. 5:43been the Code Report. Thanks for
173. 5:45watching, and I will see you in the next
174. 5:46one.