OpenAI Establishes DeployCo, a Company to Support Enterprise AI Adoption

OpenAI has launched DeployCo, a new company dedicated to supporting enterprise AI adoption. The firm specializes in helping organizations deploy frontier AI models into production and turn them into measurable business impact.

While OpenAI's main battleground has so far been model development, the establishment of DeployCo signals a clear pivot from the "selling models" stage to the "supporting adoption" stage. With many enterprises still stuck at the PoC (proof of concept) stage, the move appears aimed at encouraging the transition to production use.

In the enterprise AI market, Google Cloud, AWS, and Microsoft Azure already offer comprehensive services, so how DeployCo plans to differentiate itself will be closely watched.

JetBrains Announces Junie, an LLM-Agnostic AI Coding Agent

JetBrains, a major developer tools vendor, has announced Junie, an LLM-agnostic AI coding agent. Whereas GitHub Copilot, Cursor, and others depend on specific models (primarily OpenAI-based), Junie is designed to work with multiple LLM backends.

The AI coding tool market is expanding rapidly, with Claude Code, Codex, Cursor, and others competing in the space; JetBrains' strength lies in deep integration with its existing IDE ecosystem. Its LLM-agnostic design philosophy could prove an attractive option for enterprise users who want to avoid lock-in to a specific vendor.

AI Replaces Management, Not Engineers

A noteworthy Substack article, "AI replaced the management, not the engineers," has sparked discussion on Hacker News. The article points out that while much of the debate over AI's impact focuses on "replacing engineers," it is actually middle management and coordination work that are the most promising targets for AI-driven automation.

The analysis goes like this: engineering work is inherently creative and full of non-routine tasks, making complete automation difficult, whereas much of management consists of routine processes—aggregating, disseminating, and coordinating information—which overlap with what LLMs do best. This perspective suggests that when it comes to AI's impact on the labor market, it may be managers, rather than engineers, who face the greater changes.

Unsloth Releases MTP GGUF Models for Qwen3.6

Unsloth, a project drawing attention in the local LLM community, has released Multi-Token Prediction (MTP) GGUF models for Qwen3.6-27B and Qwen3.6-35B-A3B on HuggingFace. MTP is a technique that predicts multiple tokens simultaneously at inference time, promising significant speedups compared to conventional token-by-token generation.

The Reddit LocalLLaMA community has been actively evaluating Qwen3.6-35B-A3B. Users running it via llama.cpp report that it delivers nearly the same intelligence as Gemma4 26B-A4B while offering faster inference with no slowdown even on long contexts. It's clear that MoE (Mixture of Experts) models for local LLMs are steadily maturing.

SmolLM4-750M Proposal Sparks Community Discussion

A proposal for a "fully open SmolLM4-750M" is under discussion on Reddit. The argument is that HuggingFace's existing lineup (135M, 360M, 1.7B, 3B) has a size gap between 360M and 1.7B, and the 750M parameter class could fill that "missing middle ground."

The proposal sets out goals including a 16K context, Apache-2.0 license, and fully open weights and data recipe. As a size class well suited to low-RAM CPU inference, educational use, and small-scale fine-tuning, it is attracting interest from the community.