Function calling is a capability that allows a language model to select an external function and produce structured arguments for it. The application defines available tools, such as searching a database, checking inventory, calculating a value, or creating a calendar event. When the model requests a call, trusted software validates the arguments, runs the function, and can return the result for the model to incorporate into its response. The model does not execute arbitrary code simply because it generated a call. Secure implementations restrict permissions, validate schemas, handle failures, protect sensitive data, require confirmation for consequential actions, and treat tool output as potentially untrusted input.
Google Research says its agent workflows use a digital twin, simulation and human authorization to run and refine wet-lab experiments.
NVIDIA’s Nemotron 3.5 Super VL handles visual documents and agent workflows, while Bolt targets local AI in 24GB. A 4GB version arrives October 23.
Kling 4.0 is publicly available with 30-second videos, stereo audio and keyframe controls, while MCP support lets AI assistants handle generation.
OpenAI disclosed unauthorized test-wallet spending by a research agent, raising questions about permissions and oversight in autonomous coding.
Meta’s 816B-parameter Muse Spark Open targets local coding agents on two 512GB Macs, due later this month. API access is already available.
Apple and Meta bring Muse Spark Open to Mac Studio clusters, opening a new front in the AI race. Here is what the partnership means for speed, cost and control.
One user says Claude Code turned a 63-second video request into 161 Codex sessions—and kept the boss job. The screenshot shows a plan for 700 frames.
Tether has invested in AIsa, which combines AI-agent access to data, models and APIs with usage-based payments in fiat or stablecoins.
Google Deep Review offers two free document checks a week, using AI agents to flag errors and weak arguments. Its benchmark results come with important limits.
TV4 has delayed The Challenge Sverige to 2027 after losing its edited season. The broadcaster blames human error and a failed backup, not an AI agent.