October 22, 2026/4 min read
A grader can be wrong. So can an agent's summary of its own change. Give coding agents checks that can prove them wrong while they can still fix the change.
Zerg
AI Research Team
Blog
FEATURED POST
October 14, 2026/3 min read
How we built a CS 1.6-inspired shooter in 4 days with Zergboard
I narrated one playtest. Zergboard made 13 tickets; the first fix was live at 1:51 AM ET.
Idan Beck
September 28, 2026/5 min read
Keep your coding tools. Choose your models.
Run DeepSeek in Codex, give each agent a budget, and see which account has room. ZergRouter brings model choice, explicit fallbacks, and usage into one place.
Matthew Eisner
September 15, 2026/5 min read
Zerg Migrated a Key Durable Module for Under $500 in Model Tokens
A 21-day agent run migrated a key legacy module to TypeScript and React for under $500 in model tokens, while exposing the visual checks needed before production.
Matthew Eisner
July 2, 2026/5 min read
Nobody Reads Code Anymore
Code review used to mean reading the diff carefully. Now it means skimming a diff an agent wrote and clicking approve. The bugs that get past skim-review are different — and the review surface has to change.
Matthew Eisner
May 21, 2026/4 min read
The Gigacontext Threshold
Context windows are scaling toward a billion tokens. The retrieval-vs-window architecture call is already being made.
Matthew Eisner
May 12, 2026/6 min read
Most AI Agents Have No Memory
Most AI coding agents treat failure as a signal to retry. A new paper from CMU, the University of Washington, and Arm shows what changes when agents actually remember and distill their failures: 3.12× speedup on a standard GPU kernel benchmark, and on one operation (RMSNorm), a kernel 1.75× faster than expert-written CUDA.
Idan Beck
Have a system AI could finally make possible?
Talk to us about connectors, hardware-facing software, migrations, or operational systems where the validation loop matters.