AI Coding Weekly

Thorsten Ball: agents wrote 1,780 lines with no errors

Thorsten Ball is building a compiler one giant single attempt at a time, 1,780 error-free lines in one pass and then aiming at 5,000 to 8,000. He says that is why agents will not need the unit tests humans need.

Receipts

Ball one-shots 1,780 error-free lines, then aims at 8,000

Thorsten Ball is implementing Brezel with agents in single attempts, reporting 1,780 lines with no compilation errors, tokens streamed straight into the file. Level 2 added a bytecode compiler and stack VM, a disassembler, closures with upvalues, and a tracing garbage collector; level 3 targets 5,000 to 8,000 lines in one go.

Thorsten Ball
@thorstenball
X
And you think agents will need the same unit tests we need? Nah
Oct 8, 2026 · View on X

Read the full story

DHH let agents run a day on Campfire, C now leads

DHH says he let the agents run for a day on his Campfire rewrite, then reviewed every submitted PR, verified the work, merged what was ready, and asked for each architectural improvement to be carried across to every implementation. The C version is now in the lead.

Read the full story

How they actually work

Pocock: never start with /wayfinder, and never grill a tiny diff

Matt Pocock posted his rule for picking between grilling and one-shotting: if the expected diff is medium to large, start with a /grill-with-docs session, and if planning gets out of hand, escalate with "/wayfinder let's turn this into a map".

Read the full story

Built from 574 posts by the engineers, language designers and agent builders we follow on X over 1 days.

Get the next one by email

Coding with AI, read daily so you do not have to. The experiments, the receipts and the arguments from engineers shipping real software. Not a changelog.