✨What's New

Until now, you couldn’t see which model powered your agent, or choose another. Now Claude Sonnet 5.5 is the default, and in an agent’s settings you can pick Claude Opus 5.5, the most capable, or try an experimental model. You can see which model your agent runs on, and new jobs name the model they used. Once an agent is at work, its settings are simpler too.

• Claude Sonnet 5.5 is the new default

  Every agent now runs on Claude Sonnet 5.5 by default, up from Claude Sonnet 5, in chats and jobs alike. Your existing agents move over on their own, with nothing for you to change.

• Choose the model for each agent

  In an agent’s settings, pick the model it runs on: Claude Sonnet 5.5 is balanced and recommended, and Claude Opus 5.5 is the most capable. Each model shows a simple $ to $$$ cost guide.

• Try other models, clearly marked Experimental

  Models other than Claude sit apart under Experimental models and carry an Experimental tag wherever you pick them, so you know when you’re trying something less proven.

• See which model is doing the work

  An agent at work shows its model in the top bar; tap it for details and a shortcut to change it. Once a new job finishes, your Jobs list names the model it ran on.

• Simpler settings, and a setup that stays put

  Once an agent is at work, its settings come down to Model and Routines. The setup its maker built stays put, and the chat can no longer rewrite it.

A working agent used to give you a scrolling machine log and little sense of where things stood — and the moment you left the chat, you lost sight of it entirely. Now the work reads like a calm journal: the current step shimmers as it happens, your document glows for exactly as long as a job is running, and each job’s tasks are laid out as a List or a Board. Leave the chat and the job rides along in a small status pill you can tap to check on. And every file it makes is tagged with the job that made it.

• See every step as it happens

  While your agent works, the chat shows a live, unhurried timeline of what it is doing, with the current step gently shimmering. No more reading a wall of machine output to work out where things stand.

• Your work glows while a job runs

  The panel around your document blooms a soft light for exactly as long as a job is running, and goes quiet the moment it finishes. You can tell something is happening without watching the chat.

• Tasks as a List or a Board

  Each job’s steps appear as a clean list — newest on top, the live one marked, the done ones checked off — and a finished job folds to a single line you can reopen, with a jump straight back to that moment in the chat. Prefer columns? A Board is one tap away, and your choice is remembered.

• A status that follows you

  Leave the chat and a running job rides along in a small pill in the header bar: the current step, the time elapsed, and a tap to open the full plan. Start a second job and it splits off into its own dot — tap it to switch which job you are watching.

• Files know which job made them

  Every file your agent produces now carries a tag for the job that made it, pictures show a real thumbnail, and you can filter down to a single job’s output. Finding the thing you just made no longer means guessing.

• Honest titles, verdicts, and timings

  Every job shows its real name, whether it delivered, and how long it took — one number, the same everywhere you see it. Where a per-step time could not be measured truthfully, it is left off rather than faked.

• Agents now run on Claude Sonnet 5

  Every agent moves up from Claude Sonnet 4.6 to Claude Sonnet 5, in chats and jobs alike. There’s nothing to switch on: the agents you’ve already hired move up too.

Give an agent a thin brief and it used to do one of two unhelpful things: guess, or write a paragraph of questions back at you and leave you to compose the reply by hand. Now it asks properly — a real question with real options, one at a time, with room to think. It waits for your answer before it starts, so you no longer find out it guessed wrong once the work is already done. And when it needs something only you have, it can now ask you for the file.

• It asks instead of guessing

  Hand over a thin brief and your agent stops and asks for what is missing, with real options to pick from. It cannot ask and then start anyway — the question genuinely ends its turn, and your answers become the brief it works from. The cheapest time to fix a misunderstanding is before the work, not after it.

• One question at a time, with room to answer

  Each question takes the main screen instead of a cramped corner of the chat, with options big enough to read and compare. Tap one and it moves you straight on — no extra confirm click. You can go Back at any point to change an earlier answer without losing the later ones.

• You are never trapped on a question

  Every set of options carries a “You decide” — added automatically, to every question. If you have no opinion about the format or the tone, say so and let your agent choose. Being asked is only better than being guessed at if you can also decline to answer.

• It never interrupts what you are reading

  Open a document while a question is waiting and the question steps aside into the chat rail rather than throwing your document off the screen. On a phone it becomes a full-screen sheet. Your answers and your place survive every one of those moves.

• Hand it a file when it needs one

  Ask for a poster made from your logo and your agent can ask you for the logo: drag it in, click to browse, or paste. Any file type, up to 100 MB, with real progress. It lands in your project files, so it stays put after the job and you can point at it again later. Nothing to hand over? “Continue without a file” is always there — one missing file never holds your other answers hostage.

• “Just chat instead” is honest now

  Setting a question aside leaves a quiet, dimmed record — “Set aside — you’re chatting about this instead” — with nothing left to click. It used to leave the full, live panel sitting in your chat, and if you idly completed it, the answer would still send.

• Put it down, pick it back up

  Close a question with ✕ or Esc and it drops back into the chat still holding everything you had already told it; ⤢ puts it back on the main screen at the same step. Meanwhile the message box stays a message box — type, send, and attach as normal, and the question simply waits.

• FixAttaching an image to a message

  Attaching a picture to a chat message was silently broken, and it did not just drop the attachment — it failed the whole message. Fixed.

New users used to land in a powerful product with very little to orient them, and the moment right after hiring an agent spoke in technical terms and often showed a blank page. This release fixes that first hour. You now get a quick guided tour of the home screen, a friendly introduction when you put an agent to work, and a first run that actively helps you start instead of waiting for you to figure it out.

• A guided welcome tour

  Learn the home screen in four quick steps the first time you arrive: where to explore ready-made agents, how to create your own, where your hired agents live, and where to find your plan and credits. It appears once, automatically, and never interrupts established users — and a compass button in the sidebar replays it whenever you want.

• Hiring an agent feels like an introduction

  Putting an agent to work used to read like server output. Now it introduces the agent by name, speaks plainly through three human stages, and gives you something worth reading while you wait. When it is ready you choose where to go next, rather than being redirected on your own.

• A first run that helps you start

  A brand-new agent no longer shows you a blank page. It offers a few real example briefs tailored to what it does — tap one and it drops the full brief into your message box, ready for you to read, tweak, and send when you like. Nothing is ever sent automatically. The hardest part of a blank page is the first sentence; this writes it for you, and hands you the pen.

• A more human first reply

  The very first response used to open with a cryptic “Hmm…”. Now it shows who is working — your agent, by name — reading your brief. If it is taking a moment longer than usual, it says so, so you are never left wondering whether anything is happening.

• A steadier first run

  The first-run introduction now appears in the same place your finished work shows up, rather than taking over the whole screen. Your message box stays exactly where it always is and never moves, so nothing shifts around the moment a reply starts arriving.

• Agents that introduce themselves

  An agent can now carry a short profile of what it is good at, plus a few first briefs worth trying — which is what powers the tap-to-try briefs above. Every agent can speak for itself, so you know what to expect and where to begin the moment you meet it.

• Agents and Studios, in plain words

  You hire agents, who work in Studios, and now the agents page, sidebar, steps for creating your own, and each agent’s settings use those names. A Studio’s technical details wait behind “Under the hood.”

The gallery used to sort agents into loose categories drawn from tags, which said little about who each one was for. Now it’s organized by role, such as designers and marketers, and people who hired an agent can rate and review it. Each Studio has its own color and mark, and creating an agent takes three short steps.

• Browse by who an agent is for

  The gallery now has a row for each role, such as designers, marketers, and video producers. Filter by role, or by the kind of work you need done.

• Ratings and reviews from people who hired

  Cards show star ratings, and each agent’s page has a wall of reviews anyone can read. Only people who hired an agent can review it, right from its page.

• Each Studio has its own look

  General, Animation, and Video each have their own color and mark, and your agent’s page shows its Studio in that color.

• Create an agent in three steps

  Creating your own agent now takes three short steps: choose its Studio, describe it, and review. Hiring one goes straight to the review.

• FixBroken links lead you home

  A mistyped link or an old bookmark now shows a clear “We couldn’t find that page” screen with a button back home, instead of a blank page.

Every agent used to work with the same tools, and only while you were there to ask. Now you choose an agent’s Studio when you set it up, and Animation and Video join General. Routines let an agent run a request on a schedule, even while you’re away.

• Choose where each agent works

  When you set up an agent, choose its Studio, such as General or Animation. A badge on its page shows the Studio; tap it to see what’s included.

• An Animation Studio for animated explainers

  Agents in the Animation Studio come with tools for animated explainers and motion graphics, including neatly written math.

• A Video Studio for putting videos together

  The new Video Studio adds tools for joining and trimming clips, preparing images, and adding captions, so an agent can assemble a finished video.

• Routines: work on a schedule

  In an agent’s settings, set a request to run on a schedule, such as every day at 9 AM in your time zone. Each run’s reply arrives in a new chat.

Replies could pause for a slow check before starting, a finished job could take up to 22 seconds to report back, and long videos could time out. Now every agent runs on Claude Sonnet 4.6, replies and results come sooner, and long videos run to the end. When something goes wrong, the error stays on screen with a way to fix it.

• Every agent runs on Claude Sonnet 4.6

  Chats and jobs alike now run on Claude Sonnet 4.6, up from earlier Sonnet models, with nothing for you to change.

• Faster replies, faster results

  A check that added up to three seconds before each reply is gone, and a finished job now reports back in under a second, not up to 22.

• Errors stay on screen, with a fix

  If a message can’t go through, a clear alert stays in the chat with a button for the next step, such as trying again or signing in.

• Long videos run to the end

  Videos that take longer than about five minutes to make used to be cut off partway. Now they keep going until the video is ready.

• FixTasks stuck after a job ends

  When a job ends early, its unfinished tasks are now marked failed, instead of staying in To Do or Active forever.

Until now, your work lived in projects with no character of their own. Now everything you make is an agent with its own page, set up in one guided flow, and it chats with the character you give it. You can look over a ready-made agent’s work before you hire it, and the images and videos it makes open right in Slate.

• Everything you make is an agent

  Each one has its own page and shows its emoji in the sidebar, and you set it up the same way whether you start from the sidebar, the home page, or a ready-made agent.

• Set it up at your own pace

  Give it a name, an emoji, and a character on a roomy setup page that saves as you go. Keep it as a draft until you’re ready to put it to work.

• An agent that sounds like itself

  It chats with the character you gave it, knows which skills it has, and can look back at its earlier jobs when you ask.

• Look before you hire

  Each ready-made agent has a page with samples of its work, what it can do, and when to use it, and its card shows its real work instead of stock photos.

• Zoom into images, play videos in place

  Click an image your agent made to zoom in, and play the video files it makes right in the viewer, instead of downloading them.