Our daily AI roundup highlights significant shifts in machine intelligence, expanding software acquisitions, agent workflows, and semiconductor design. Organizations worldwide continue integrating automated systems directly into production pipelines. Here are the key technical developments shaping the industry today.

Big Developments in Enterprise AI

Salesforce agreed to acquire Listen Labs to integrate research agents and customer simulations into its software suite, pending customary closing conditions and regulatory approvals, according to an announcement on X.

General Intuition secured $220 million in new funding at a $6.2 billion valuation, opening partner access to action models trained on synchronized video and recorded digital actions, as stated on X.

OpenAI launched Dots to maintain project continuity across sessions using persistent Astra agents, integrated cloud computers, and granular permission controls for connected apps, detailed on X.

Models and Evaluation Benchmarks

NVIDIA released Kumo Tabular across three sizes alongside an inference library, enabling classification and regression predictions directly from labeled rows without specialized task training, according to X. Users must still evaluate accuracy against their specific workloads before wide implementation.

ARC Prize verified GLM 5.3 Flash benchmark scores, recording 65.8% on ARC-AGI-2 at $0.09 per task and 91.0% on ARC-AGI-1 at $0.04 per task, while noting architectural differences between early endpoints and Baseten hosting on X.

Vals updated its model index to version 2.1, introducing Terminal Bench 4 and a separate GDP-weighted Tax Agent Benchmark, as noted on X.

Autonomous Software Agents in Production

Devin introduced an integration allowing eligible ChatGPT Plus and Pro subscribers to use existing OpenAI allowances across Devin Cloud, Desktop, and CLI environments, reported on X.

Cognition partnered with MongoDB to release a modernization workflow where Devin implements code modifications while MongoDB AMP manages data migration, as shared on X.

Sandbar linked its Stream voice application to external AI agents via authenticated Model Context Protocol connections, according to X. Users can track developments across our apps coverage.

Hardware Infrastructure and daily AI roundup Outlook

Prime Intellect received NVIDIA Vera CPU hardware to support execution sandboxes for automated agents, as announced on X. The team previously completed initial hardware benchmarks in March.

General Compute signed a multiyear agreement with Cerebras to secure dedicated inference compute capacity for agentic coding starting in early 2027, according to X.

AMD entered an agreement to acquire World Labs for $8.2 billion in stock to support spatial computing research, as confirmed on X. Tracking hardware innovation across computers and the broader digital economy remains essential as companies expand infrastructure.

Following each daily AI roundup provides engineering teams with structured visibility into operational model releases, infrastructure commitments, and specialized agent workflows.