The app layer had a strong week across creative, coding, and research agents.
Luma launches creative agents
On March 5, Luma AI launched Luma Agents, an agentic creative platform powered by a new "Unified Intelligence" model family. The first, Uni-1, is trained jointly on audio, video, image, language, and spatial reasoning to plan and render end-to-end creative work, rolling out via API to customers like Publicis, Adidas, and Mazda.
Cursor adds Automations
Also March 5, Cursor rolled out Automations — coding agents that auto-launch inside the dev environment when triggered by a new commit, a Slack message, or a timer, looping humans in only when needed. Agentic coding is moving from "ask" to "always-on."
Perplexity ships a model council
On March 6, Perplexity shipped Custom Skills (reusable workflows for Perplexity Computer), a web Voice Mode, and a Model Council that runs multiple frontier models in parallel for consensus answers — doubling down on multi-model orchestration.
Model watch
Google released Gemini 3.1 Flash-Lite (March 3), a cheap, fast tier (~$0.25/$1.50 per M) that improves agent economics; Lightricks open-sourced LTX-2.3, a 22B audio+video model; and OpenAI's Codex app reached Windows (March 4).
What it means if you're picking tools
Creative, coding, and research agents all leveled up. See image generation, coding, and research.
Sources: Luma Agents + Uni-1 (TechCrunch) · Cursor Automations (TechCrunch) · Perplexity changelog (Mar 6) · Gemini 3.1 Flash-Lite (SiliconANGLE)