Company@X — 2026-04-28#

Signal of the Day#

AWS has officially partnered with OpenAI to natively integrate OpenAI models, Codex, and Managed Agents into Amazon Bedrock. This bridges the gap for massive enterprises, allowing them to build and run OpenAI-powered agents directly inside AWS’s auditable, private infrastructure without data ever leaving their VPCs.

Key Announcements#

[AWS] · Amazon Bedrock OpenAI Integration Amazon Web Services announced three new OpenAI-powered offerings for Amazon Bedrock: access to the latest OpenAI models, Codex for enterprise development, and OpenAI-optimized Managed Agents. The Managed Agents feature includes memory, per-agent identity, and full auditability, fundamentally shifting how large organizations deploy autonomous systems in production. This gives AWS customers the ability to leverage frontier intelligence while maintaining the rigid security and governance standards of their existing AWS environments.

[Poolside AI] · Laguna Open-Weights Release Poolside released its first public open-weight models, Laguna M.1 and Laguna XS.2, under an Apache 2.0 license. Laguna XS.2 is a 33B total / 3B active parameter Mixture-of-Experts (MoE) model specifically optimized for agentic coding and long-horizon tasks that can run locally on a single GPU. The models, which deliver near-SOTA results comparable to Qwen-3.5, were trained entirely in-house using Poolside’s proprietary data pipelines, training infrastructure, and reinforcement learning stack.

[Hugging Face] · ml-intern Launch Hugging Face launched ml-intern, an open-source autonomous agent designed to replicate the workflow of human machine learning researchers. The agent natively browses arXiv, pulls datasets, writes training scripts, provisions GPU sandboxes, and executes supervised fine-tuning runs. In internal benchmarks, ml-intern autonomously trained a scientific reasoning model that achieved a 32% on GPQA—significantly outperforming Claude Code’s 22.99%—by self-directing dataset curation and executing 12 iterative training loops.

[AWS] · Amazon Quick Desktop App AWS launched the Amazon Quick desktop application, a personalized AI productivity assistant that connects directly to local files, calendars, email, and enterprise tools like Slack. The agent sits locally on the user’s machine, learning from daily workflows to preemptively retrieve information, draft communications, and flag shifting priorities. CEO Andy Jassy highlighted the app’s capability to let users create custom micro-agents that automate undifferentiated, repetitive tasks out of their inbox.

[OpenClaw] · OpenClaw 2026.4.26 OpenClaw shipped version 2026.4.26, a robust release focused heavily on local model operations, matrix encryption, and migration pipelines. The update introduces improved Ollama context handling, a constrained credentials pathway for Google Live Talk, and bundled migration tools to import configurations from Claude Desktop and Hermes. The release also reduces setup friction by enabling a one-command Matrix E2EE configuration without manual file editing.

Also Noted#

• [Google Wallet] (Digital ID Expansion): Google Wallet rolled out selective disclosure digital IDs, enabling Aadhaar credentials in India and secure passport-based ID passes in Singapore, Taiwan, and Brazil.
• [Google] (Translate Pronunciation Practice): To mark its 20th anniversary, Google Translate launched an AI pronunciation feature on Android that analyzes user speech to provide instant feedback on delivery.
• [Waymo] (Portland Groundwork): Waymo announced its expansion to Portland, Oregon, beginning manual driving operations today to map the city in preparation for a future robotaxi service rollout.
• [Y Combinator] (Agent Ecosystem Focus): YC is heavily signaling to founders that the next generation of enterprise software must be rebuilt with AI agents as first-class citizens, rather than relying on legacy software designed for human clicks.
• [Tesla] (Actually Smart Summon): Tesla rolled out an update to “Actually Smart Summon,” notably improving the autonomous retrieval system’s reaction speed and overall usability.