Tech Landscape #428
Happy Horse enters the race, vibe creating and vibe working, and the value of being authentically human.
Hello!
NB This edition is a day later than usual because of the public holiday here in the UK.
In the past few weeks some clear trends have been emerging around AI assistants moving into creative design and workplace tools. I work in a marketing agency and creative design is a large part of our work, so we need to try to predict the impact of this — but prediction is hard. I can tell you the direction the technology is moving in, and I can tell you what has happened in similar circumstances, but I can’t give any guarantee of what will happen. As the standard disclaimer goes: “past performance is not an indicator of future results”.
I firmly believe there will always be places where human creativity is needed, but maybe there will be fewer places, or maybe there will be the same or more places but they’ll look very different.
To give a tangible example: for many years the typical marketing agency creative pair has been art director + copywriter. Now, with AI tools, each can do most of the other’s work. Whether or not that’s a threat depends largely on if the pairing only exists for the division of labour by speciality, or if it exists because the sum of the two parts is greater than the whole. I suspect we’ll see more of a hybrid approach, with some “unicorn” creatives doing everything themselves, and some where the value is in the tandem.
But I don’t know for sure. If I did I’d be a millionaire consultant getting someone to write this for me as I sit on my yacht; but I’m not, I’m just a bloke writing this as I sit on my sofa.
Anyway, let’s get on with it. Hope you’re well!
Synthetic Audio-Visual
New video model: HappyHorse 1.0
It has the capabilities of other best-in-class models: it can generate multi-shot videos with native audio at up to 15 seconds length in up to 1080p resolution, from text prompts and up to nine image references, and has video editing functions including motion transfer. It comes from Alibaba’s ATH team and is available on its own website or through various creative tools.
I ran a few tests ⬇️ and you can see that it’s a very good model with excellent detail but motion and coherence can be just slightly off, which stops it from being a great model. I’m a bit surprised it rated over Seedance 2.0 on the leaderboards.
ElevenLabs launched ElevenMusic, a platform for human and AI music where users can create or remix songs and earn revenue from listener engagement. elevenlabs.io
Remixable Spotify, with no big name artists. I’m very curious to see how this plays out.
xAI’s Grok introduced Custom Voices for users to create synthetic voices from “a few seconds” of recorded audio. x.ai/news
Creative Tools
Until now, AI in creative tools has largely been limited to the output stage, generating and editing synthetic content; but it’s increasingly making its way into assisting the creative process itself. After vibe coding: vibe creating?
Adobe launched Firefly AI Assistant in public beta. It’s a conversational agentic system that orchestrates multi-step workflows across Creative Cloud apps from prompted tasks. blog.adobe.com
So you set it a task, and it plans how to accomplish it with the tools available; for example, given a batch of images it might call a Photoshop tool to retouch them, Firefly to change their backgrounds to something unified, and Express to resize them for social. It’s fairly limited at the moment, but more features will be added. I think this is the next stage in creation for most people; more powerful than prompting, less complex (and controllable) than nodes or UIs.
Grok Imagine added Agent Mode, an assistant for creative workflows. x.com/imagine
Pika launched Agents, a creative assistant to the generation of personalised content. instagram.com/pika_labs
The inverse of putting an assistant into creative tools is putting creative tools into an assistant.
Claude added connectors to creative tools including those from Adobe, Autodesk, and Blender, aiding in everything from on-demand learning to batch image edits. anthropic.com
This comes on the heels of Claude Design [TL 426].Both Higgsfield and Pika launched MCP servers, making them available to use in any (compatible) assistant.
For those who want more fine control over their output than a conversational assistant can provide, the infinite canvas is the popular interface.
LTX Studio added Canvas, a collaborative infinite workspace that allows teams to generate and iterate AI visuals in real time before moving into production. ltx.studio
Higgsfield added Canvas, a node-based workspace for assembling complex creative workflows. threads.com/@higgsfield.ai
The Higgsfield team release new features at an incredible rate. I suppose they’re using AI coding; if so, I wonder how stable their codebase is?Kaiber launched a new suite of tools, Canvas, Cuts, and Editor, with a focus on music video creation. instagram.com/kaiberai_
After relaunching itself as Superstudio (which I found very confusing to use) it’s now re-relaunching itself as a music video app, which seems… limited.
New AI updates to Photoshop include 3D rotation of 2D objects, the Firefly 5 model in Generative Fill, automated layer renaming, and more. threads.com/@photoshop
3D & World Models
SpAItial announced Echo-2, a world model that uses ‘physical grounding’ for persistent spaces that feel consistent with the real world. spaitial.ai
Here ⬇️ is a quick test I made; it works well, but I can’t see a huge improvement over other world models that do the same job. Maybe I need to test more.
World Labs’ Expand tool is now available to all users to generate vast spaces by extending existing areas. x.com/theworldlabs
These place-generation world models are still very experimental and don’t have much practical application right now. But at the same time we’re watching a new type of 3D creation experience being developed in front of our eyes:
Moonlake AI introduced 3D Agent, a computer use agent which can control Blender to generate scenes and rigged assets from a prompted task. moonlakeai.com
ByteDance’s Seed-3D 2.0 is an upgraded image-/text-to-3D object generation model with improved geometry, textures, and faster processing speeds. seed.bytedance.com
Roblox showed off Reality Hybrid Architecture, a system that uses AI to render high-quality graphics based on the basic shapes that they’re composed of. about.roblox.com
This is kind of bonkers; users create games with the Roblox engine and the neural rendering system adds rich and detailed textures to make them look highly polished. Experimental, but watch this space.
Social
Instagram’s crackdown on unoriginal content came to Feed, in both posts and carousels. instagram.com/creators
Snapchat introduced AI Sponsored Snaps, a conversational ad format that allows brands to deploy AI agents directly into the Chat tab to interact with users. newsroom.snap.com
Divine is a new short-form social app based on Vine, seeded with half a million archived videos from that platform. about.divine.video
It is, to all intents and purposes, Vine reborn. But the world has moved on and I don’t know if nostalgia is enough to make a successful app. Maybe the desire to create quick and easy content that doesn’t need a high level of polish is enough of an incentive? IDK. There is one thing that’s interesting about it, though: a toggle to only show ‘human made’ content.
Divine is based on Nostr, yet another ‘open social’ protocol like Mastodon’s ActivityPub and Bluesky’s atproto — and of course the three aren’t interoperable. But Divine has developed ProofMode authentication, which has levels of verification about how content was made. At its highest level it can attest that a video was made with Divine’s built-in camera and users can opt in to see only this content, knowing for sure it was shot with the camera by a human, and not recycled or synthetic.
Also this week Spotify announced Verified, a badge to indicate that an artist is authentic and trusted, based on either human verification or certain engagement signals. It’s an attempt to reduce the impact of sloppy or spammy accounts.
I think these are largely a good thing; I’m clearly supportive of intentional AI-made content, but I also don’t want my feeds to be inundated with low-quality slop (either human or synthetic).
Assistants & Search
Gemini brought the Memories feature to users in the UK, remembering key details and preferences from past chats to personalise results in new ones. blog.google
Gemini can now generate files directly from prompts, letting users create and export files including Docs, Sheets, and PDFs. blog.google
Google is testing all sorts of Gemini-powered features across its products; some will work, some won’t.
Google Photos’ new wardrobe feature uses Gemini to pull out every item of clothing you’ve worn in your photos recently and offers to combine them in a moodboard or virtual try-on so you can experiment with outfits. blog.google
Google TV added Gemini-powered features including personalised dynamic screensavers, photo remixing, and video generation. blog.google
I can’t imagine myself using any of these, but maybe if you have young children?
Amazon customers can interrupt audio product overviews with questions. Join the Chat builds on the Hear the Highlights feature available in the US. aboutamazon.com
Agents & Work
I’m probably not going to keep talking about this as it’s slightly outside of my area of interest, but it’s worth noting this “vibe working” trend of agentic tools pushing into business processes.
Claude Cowork can now build live artifacts such as dashboards and trackers that update with data from apps and files. threads.com/@claudeai
Amazon launched the Quick desktop app, a generative AI assistant for macOS and Windows that provides coding, debugging, and productivity support directly from the desktop. aboutamazon.com
Mistral launched agentic features — a multi-step Work Mode for complex cross-app tasks in Le Chat and remote coding agents in Vibe — powered by its new Medium 3.5 foundation model. mistral.ai
Manus launched Cloud Computer, an always-on virtual environment for AI agents to run complex tasks (like coding and browsing) safely. manus.im
This is basically like a safe space to run OpenClaw.
Bonus Links
Why AI Agents are either the best or worst thing we’ve ever built. Dr Hannah Fry on the sudden emergence of OpenClaw and its potential ramifications. This is excellent.
Some interesting data from recent reports:
More people turn to AI assistants than company or brand websites for product information and research. Adobe AI and Digital Trends
Retail, travel, and financial services are the sectors seeing highest growth in visit share from AI tools. Adobe AI Traffic Report April 2026
US teens say TikTok is the platform that negatively impacts their sleep and productivity the most. Teens’ Experiences on TikTok, Instagram and Snapchat
Most people use Meta’s social platforms in the same way regardless of their age group. Generation Zeitgeist 2026
The mental models we've built are increasingly disconnected from reality. We wanted to test a hypothesis: is the generational playbook still fit for purpose? The answer is a resounding no. We found a profound convergence in attitudes and behaviors across all age groups, one that demands a fundamental rethink of how brands approach audience strategy.