Public story · 2026-03-17 · source-backed

Code-A1: Adversarial Co-Evolution Eliminates Test Suite Dependency.

Issue

2026-03-17

Confidence

source-backed

Sources

Redaction

redacted

Story

A code model and test model trained adversarially via RL — each model's failures become the other's training signal. This self-improving loop could break the benchmark dependency bottleneck for training coding agents. Source

RSIBench-Data: agents improve on their first attempt 58% of the time but finish below their own peak in 78% of runs.
arXiv 2607.25886 isolates data-centric research capability by fixing the entire post-training stack so only the agent's data strategy varies. Four frontier agents across six benchmarks. Among searches that continued past the best observed score, 78.26% ended on a lower-scoring...
Engram: a lean retrieved context beats stuffing the full history into the window.
Engram, a bi-temporal memory engine, scored 83.6% versus 73.2% for a full-context baseline on the 500-question LongMemEval_S benchmark, a statistically significant +10.4 points, while using ~9.6k tokens instead of 79k (arXiv 2606.09900). Roughly 8x fewer tokens and more accura...
Benchmark scores AI loop controllers apart from coder
The benchmark has a Controller model receive a structured summary after each coding round and tell a separate fixed Worker agent what to do, verify, or when to stop, which isolates loop guidance from coding ability. Across Controllers the paired reduction in estimated inferenc...
Airbnb says AI writes 60% of its new code and cut concept-to-launch by 60%
Four stories about things going wrong. Here's one about something working, with actual numbers attached. In an August 7 disclosure covered by TechCrunch, Airbnb said AI now writes 60% of its new code, that concept-to-launch time on key initiatives has dropped by as much as 60%...
Microsoft's MAI coding models reach Copilot and VS Code.
MAI-Code-1-Flash, a 5B-parameter coding model, is in GitHub Copilot and VS Code, and Microsoft says it beats Claude Haiku 4.5 across core coding benchmarks, +16 points on SWE-Bench Pro at 51.2% versus 35.2%, using up to 60% fewer tokens. MAI-Thinking-1, a 35B-active MoE with a...
GitHub Copilot Switches to Token-Based Billing Today. Developers Report 25x Cost Increases.
The flat-rate era for AI coding tools ended today. Not with a whimper. With invoice shock. GitHub Copilot officially moved from fixed monthly subscriptions to usage-based "AI Credits" billing on June 1, 2026. Code completions remain free, but agent mode, chat, and premium mode...

Source trail

Source More from arxiv.org

Entities

Adversarial Co Code Evolution Eliminates Test Suite Dependency

Provenance

Canonical issue: Ramsay Research Agent — 2026-03-17
AI generated: no
Story unit: 2026-03-17-code-a1-adversarial-co-evolution-eliminates-test-suite-dependency
Labels: source-backed, canonical briefing excerpt

Fetching from the wire…

Code-A1: Adversarial Co-Evolution Eliminates Test Suite Dependency.

Story

Related stories

RSIBench-Data: agents improve on their first attempt 58% of the time but finish below their own peak in 78% of runs.

Engram: a lean retrieved context beats stuffing the full history into the window.

Benchmark scores AI loop controllers apart from coder

Airbnb says AI writes 60% of its new code and cut concept-to-launch by 60%

Microsoft's MAI coding models reach Copilot and VS Code.

GitHub Copilot Switches to Token-Based Billing Today. Developers Report 25x Cost Increases.

Source trail

Entities

Provenance

Code-A1: Adversarial Co-Evolution Eliminates Test Suite Dependency.

Story

Related stories

RSIBench-Data: agents improve on their first attempt 58% of the time but finish below their own peak in 78% of runs.

Engram: a lean retrieved context beats stuffing the full history into the window.

Benchmark scores AI loop controllers apart from coder

Airbnb says AI writes 60% of its new code and cut concept-to-launch by 60%

Microsoft's MAI coding models reach Copilot and VS Code.

GitHub Copilot Switches to Token-Based Billing Today. Developers Report 25x Cost Increases.

Source trail

Entities

Provenance

Story

Related stories

RSIBench-Data: agents improve on their first attempt 58% of the time but finish below their own peak in 78% of runs.

Engram: a lean retrieved context beats stuffing the full history into the window.

Benchmark scores AI loop controllers apart from coder

Airbnb says AI writes 60% of its new code and cut concept-to-launch by 60%

Microsoft's MAI coding models reach Copilot and VS Code.

GitHub Copilot Switches to Token-Based Billing Today. Developers Report 25x Cost Increases.

Tomorrow's issue, in your inbox.

Source trail

Entities

Provenance

Story

Related stories

RSIBench-Data: agents improve on their first attempt 58% of the time but finish below their own peak in 78% of runs.

Engram: a lean retrieved context beats stuffing the full history into the window.

Benchmark scores AI loop controllers apart from coder

Airbnb says AI writes 60% of its new code and cut concept-to-launch by 60%

Microsoft's MAI coding models reach Copilot and VS Code.

GitHub Copilot Switches to Token-Based Billing Today. Developers Report 25x Cost Increases.

Tomorrow's issue, in your inbox.

Source trail

Entities

Provenance