Tech Landscape #429
Real-time voice models, dreaming agents, and better AI search results
Hello!
This week’s edition is short and, if I’m being honest, somewhat bitty. You should know that there will be no regular edition next week as I’m off for a short break. However, I might send out a mid-week edition that covers the news from Google I/O 2026, which is happening next Tuesday, 19th May.
I think there are going to be a number of exciting announcements at I/O, and I’d guess a lot of those will relate to agentic updates. Here are my predictions:
the next version(s) of Gemini, optimised for agents
agentic upgrades to the Gemini app, including desktop computer use (and, I hope, connectors to third-party tools)
agentic creation in Flow, using the next version of Veo (with Seedance 2.0-level capability)
I’m sure we’ll get a few surprises too.
Right, let’s get on with it. Hope you’re well!
Synthetic Audio-Visual
OpenAI launched three new real-time voice models: GPT‑Realtime‑2 with reasoning that “can handle harder requests and carry the conversation forward naturally”, GPT‑Realtime‑Translate for live multilingual support, and GPT‑Realtime‑Whisper for low-latency transcription. openai.com
These are available through the API so you’ll likely use them in third-party apps with live voice modes.
Creative Tools
ElevenCreative added Studio Agent, an assistant for creative support directly in the video timeline editor. elevenlabs.io
LTX Studio added Flows, a node-based canvas for repeatable control over visual creative workflows. x.com/LTXStudio
Dreamina launched a mobile app with the Cast feature that lets users add themselves into videos generated with Seedance 2.0. instagram.com/dreamina_ai
Filling the space vacated by the Sora app.Luma’s Uni 1.1 image model is now available via API. lumalabs.ai
This means you’ll start seeing it turn up in third-party tools, including Magnific (formerly Freepik). I talked about Uni-1 when it launched a few weeks ago [TL 423] — it offers a decent alternative to Nano Banana and ChatGPT Image.AI music startup Mozart launched Studio 1.0, a creative workspace aimed at providing pro-level tools. instagram.com/mozartaiofficial
Higgsfield Marketing Automation
Higgsfield launched three new features for the rapid and automated creation of marketing videos:
Hooks are 25+ presets for generating the opening hook of a UGC video from an avatar and product. instagram.com/higgsfield.ai
Ad Reference copies the pattern of any short video with a changed avatar and product. instagram.com/higgsfield.ai
Virality Predictor analyses videos to score their viral potential via brain activity. instagram.com/higgsfield.ai
I presume this is based on TRIBE v2, an open model from Meta that’s “trained to understand how the human brain processes complex stimuli”.
These are all available through the Higgsfield platform, but also through third-party tools such as Claude and OpenClaw which add extra reasoning and automation capabilities.
On one hand, these feel like an easy way to create spam/slop videos without human craft and brand distinctiveness. On the other, maybe that’s all that a lot of small businesses need. Marketing agencies and production companies should be paying very close attention to what’s becoming possible now.
Assistants & Search
Google boosted AI Mode and AI Overviews with five new features including follow-up suggestions, human advice from public online sources, and highlighting results from a user’s paid news subscriptions. blog.google
Agents for Work
ElevenLabs’ ElevenAgents supports new modalities to comprehend and process images, files, audio notes, contacts, and locations across WhatsApp and Web widgets. elevenlabs.io
Manus introduced Recommended Connectors to suggest relevant third-party app integrations based on the user’s task. manus.im
Perplexity’s Personal Computer is available to all Mac users in a new app. perplexity.ai
Unity developers can start agentic coding as Unity AI is now in open Beta, available as an assistant in the main software development tool or via a connector to third-party tools. instagram.com/unitytechnologies
Agents can become less effective over time in lengthy and more complex tasks, losing context or knowledge; two interesting new features this week aim to fix that by analysing past work to improve future performance.
Manus introduced self-updating Projects to automatically monitor data and refresh tasks to ensure information remains current. manus.im
Claude added ‘dreaming’ in Managed Agents, a background process that analyses and reviews past sessions to help agents self-improve over time. claude.com
Social & Messaging
Instagram is testing an AI Creator label for accounts which post a lot of synthetic content. instagram.com/creators
Meta expanded its AI-powered age assurance measures, using visual analysis and contextual profile clues to identify and remove underage users or place suspected teens into protected accounts. about.fb.com
I think this was the first issue where every single story involved AI.