TUESDAY · SEPTEMBER 29, 2026 · ISSUE 20 · SINCE SATURDAY No. 20
Daily News — 2026-09-29
1,926 papers indexed on arXiv·~12,000 packages released on PyPI·33 papers surfaced by Hugging Face
None of it is in your agent's weights.
Spotlight
Randomizing which encoder layers form the latent during training turns a fixed architectural choice into a regularizer that benefits both reconstruction and generation without extra cost.
Representation autoencoders face a structural tension: the encoder layers that best preserve pixel detail are not the same ones that make a diffusion model's job easy. RAEv2 documented this as a Pareto frontier —…
The Crews-plus-Flows split is the right architectural instinct — deterministic control and agent autonomy belong in separate primitives, not one leaky abstraction.
CrewAI draws a sharp line between two modes of working: Crews, where agents operate with genuine autonomy and negotiate task delegation among themselves, and Flows, which impose event-driven, decorator-based control…
Papers
Randomizing which encoder layers form the latent during training turns a fixed architectural choice into a regularizer that benefits both reconstruction and generation without extra cost.
FuseReg trains over random encoder-layer subsets so one decoder handles any fusion without retraining, cutting unguided gFID by up to 29%.
Phase-aware quantization that treats prefill and decode as distinct hardware problems — and solves both without requiring a single unified compromise.
Disaggregated quantization uses separate formats for prefill and decode, yielding 32+ point accuracy gains and 1.78x faster first-token latency on 27B models.
Solves the overlooked quadratic bottleneck inside block-sparse attention — block *selection* itself — with a log-depth pyramid that keeps overall complexity at O(N log N).
PISA cuts block-sparse attention's quadratic block-selection bottleneck to O(N log N) via a coarse-to-fine pyramid Top-K strategy with fused Triton kernels.
A disciplined architecture that turns future-video supervision into inference-free predictive representations, backed by the largest open robot training corpus assembled to date.
InternW0-Δ unifies visual dynamics, 4D geometry, and action generation in a Mixture-of-Transformers WAM trained on 20K+ hours of open robot and egocentric data.
Failure-driven tool synthesis with a paired admission gate that counts harm directly beats every human-curated and self-refinement baseline across all thirty task-backbone combinations.
TimeEvo grows a time series agent's tool library from scratch—clustering diagnosed failures into capability gaps—and improves accuracy on every task tested.
A week-one ecosystem census showing Jev is already a multi-purpose judgment layer, not just a router, with attention and adoption pointing in opposite directions.
A data-driven analysis of 2,170 GitHub projects examining how the Jev decision model is used for judgments, scoring, and routing across application domains.
7 more tool picks in this edition
Every pick in the Wire gets the same treatment: read, verified, and given a written verdict.
A disciplined adaptation of Stable Diffusion 3 to elevation refinement that halves Dense Urban DSM error in held-out tests — with honest accounting of where it fails.
Modified Stable Diffusion 3 cuts satellite DSM elevation error nearly in half (6.00→3.45 m RMSE) by conditioning on both DSM and optical imagery.
Routing tool and summary advantages to separate token segments before the backward pass measurably fixes a real failure mode in tool-calling RL, with honest bounds on what it cannot fix.
SLCA-GRPO routes execution advantages to tool tokens and preference advantages to summary tokens to fix cross-segment credit misattribution in tool-calling RL.
Tools & packages
A local proxy plus config manager that makes the multi-agent mess on your machine actually navigable, with unusually careful file-editing discipline.
A menu bar app that routes multiple AI coding agents to different model backends from one place.
A focused cost-reduction tool for coding agents: skip speculative file reading by asking what code does, not where it lives.
CLI built for coding agents that uses Jev to locate relevant files and source context by describing what code does.
The Crews-plus-Flows split is the right architectural instinct — deterministic control and agent autonomy belong in separate primitives, not one leaky abstraction.
CrewAI orchestrates role-playing autonomous agents that collaborate on tasks by assigning each agent a defined role.
A practical, well-documented system for running a 125B MoE model on a single gaming GPU by distributing experts across VRAM, RAM, and SSD in parallel.
A one-click inference engine that runs a 125B MoE model on an 8GB+ NVIDIA GPU with a local OpenAI/Anthropic-compatible API and optional image input.
A Claude agent skill that enforces visual continuity through automated scoring, not taste — and the rejection log proves the oracle has teeth.
Generates product launch and feature demo films as a single uncut sequence, with an oracle enforcing beat-to-beat continuity throughout.
A structured design methodology baked into an agent skill — the mandatory checkpoint and automated SVG audit are what make it more than a prompt.
A logo-design reference skill for AI agents covering principles, SVG craft, and a 1,400+ logo library.
8 more paper picks in this edition
Every pick in the Wire gets the same treatment: read, verified, and given a written verdict.
A 3D shared office where coding agents sit at desks and wave when they need you — a spatial bet on how small teams coordinate autonomous work.
Cartoon 3D office where you seat Claude Code agents at desks, share live terminals, talk over voice, and track GitHub issues and PRs.