What is being built with AI, how it is put to work, and what happens when it goes wrong.
Ramp Labs, at the finance company Ramp, on how the week's models do on accounting work.
Ramp Labs
Two new challengers to the Pareto frontier on accounting work.
First Sonnet 5.5, then GPT-6.1 Sol: each one scores more than any model at its price or lower. GPT-6.1 Sol then pushes both GPT-6 Sol and Sonnet 5.5 off the frontier.
Sergio Mattei, who works on the Codex app at OpenAI, on the details of Space, a thread of nine posts filed here as one in its own order; Tibo pointed readers to it as the one thing to read about Space. The film is the thread's fifth post, visualizations embedded in a page.
Sergio Mattei
We can finally show what we've been cooking. 🧑🍳
Here's some details about it that I really like:
First, the home view is really beautiful--the team cooked with the design. Rich previews the documents you've been working on:
Ask complex questions and generate text inline without breaking flow.
A photo printer generates images for you. Add creative artwork to your Pages.
Turn images and tables into beautiful, widescreen displays.
Embed rich, multiplayer visualizations and apps within your pages.
This one's mind blowing btw--
Prompt blocks. Embed one-click chat prompts for readers.
And to celebrate the launch of Space, we're dropping one more icon choice for your Dock.
Introducing the Spaceblossom.
Appearing shortly in a Settings section near you.
Now, of course, there's so many more amazing things that I haven't mentioned:
⋅Collaborative editing
⋅Bring in your dots to keep pages up to date!
⋅@ChatGPT in the comments
...and much much more
we hope you like it!
https://x.com/ajambrosino/status/2104991411930357862
Alex Finn, another early tester, on who dots are for and who they are not for.
Alex Finn
OpenAI just released Dots, their competitor to Grok Bot and Muse
I've had early access for a bit, and I really like it
Its biggest differentiators are that it has access to ALL your ChatGPT memories, as well as all the amazing tools in the ChatGPT harness including:
• Voice mode
• ChatGPT plugins
• The new ChatGPT Spaces
• much more
Is it wildly differentiated from the other agents? No.
But if you've been using ChatGPT for a while, the fact that it just knows you and all your past chats really well is massive
And if you are a desktop app power user like me, being able to have your agent spin up codex threads on the go is wildly helpful
If you are more in the 'normie' crowd who primarily just has your agent make restaurant reservations, probably not for you
But if you use AI daily for more advanced stuff like building and doing all your knowledge work, it's excellent
Chubby, editor-in-chief of the Superintel newsletter, on where dots are not. OpenAI's help page says dots roll out to Pro users in every market except the European Economic Area, Switzerland and the UK; Business Premium gets them in every supported region.
Chubby
Not again. They cant be freaking serious. What the heck is wrong with europe.
Luiza Jarovsky, who writes about AI and privacy, on how dots and Meta's Muse are designed.
Luiza Jarovsky
🚨 WHY does every AI agent now have to be fluffy and toy-like, appealing to children?
Last week it was Meta's Muse; today it's OpenAI's Dots.
These are risky, privacy-invasive products meant for adults only.
It would be great if their design and marketing respected that.
Ray Fernando, who spent twelve years at Apple, with an early-access review that sets out what did not work beside what did, and says it was not paid for.
Ray Fernando
I’ve had early access to OpenAI dots and I’m not here to join in on the glaze fest. Is this a Grok Bot or Hermes killer right now? NO. But…Spaces and allowing agents to thrive in apps feel like the right direction for these agents.
•• The Good
What worked well during my usage was how proactive the agent is right from the start with my existing plugins/connectors. The team is shipping insanely fast and have been quickly making changes to fine tune this experience.
•• Caveats
With a slow internet connection in the middle of the South Pacific Ocean I didn’t have a good enough connection for voice and had to heavily rely on cloud. There were other quirks related to this that might be noise but worth mentioning.
•• Bad at Writing
I tried connecting my dot to GitHub and it performed a research task to make a Spotify wrapped + story about my weeks at sea where I made over 294 PRs and shipped over 30 releases. The design felt very GPT-4 like with lots of text and the story was even worse.
•• Conclusion
I need more time with this and need to get this hooked up on my Mac at home with Codex so my dot can fly. I think there is potential to have a dot like this have access to powerful tools, web, and apps like Space to write pages and interact with humans.
I love the ease of use and speed of Grok Bot a lot. It is so fast to spin up many bots and have them quickly work together especially now as a team.
I’ll also be doing a deep dive of Grok Bot vs dots and continue to share my honest insights as a solo engineer building software after leaving FANG.
•• Disclosure & Philosophy
I hate it when I read early access posts and they only share the good but leave out the quirks. As a fellow engineer who’s shipped to billions of people, I respect the engineers who build software and spend many hours away from their friends and family to build their life’s work. The views and opinions are from me. I have not been paid or compensated by OpenAI, in fact, I’m still a paying customer and haven’t received any compensated plans from OpenAI as of the writing of this post.
As part of the Dev Early Access I plan to continue to gift my expertise, opinions, be it good or bad, with the intention that they also value their product enough to make improvements for the broader community.
BridgeMind, a channel about building with AI, measuring what Pro 500 buys with Ultrafast on Astra that afternoon. Half an hour later the same account posted that the whole week's allowance was gone in under thirty minutes.
BridgeMind
One GPT 6 Astra prompt using ultrafast on the new $500 ChatGPT Pro Max plan.
Ran for 2 minutes. Used 2% of my weekly limit.
That is 50 prompts, or about 100 minutes of work, before the week is gone. For $500 a month.
This is a complete scam.
Tibo's second go at this post: an hour earlier it said over sixty partners, and this one says sixteen, the number OpenAI's recap gives.
Tibo
You can now use your ChatGPT subscription directly in over 16 partners products. No little rules, you can just use all your included usage right there. This includes things such as Devin, OpenCode, Notion, and many more.
Sign in with ChatGPT and go.
Quinn Slack, who runs Amp and founded Sourcegraph, on Sign in with ChatGPT in Amp.
Quinn Slack
We worked with @OpenAI so you can now (officially) Sign in with ChatGPT on @AmpCode and use your sub and credits on Amp.
You can use Amp totally for free this way (and only pay for Amp Megawatt/Gigawatt when you need orbs).
Go get orbin'!
https://ampcode.com
OpenAI's developer account on the Decisions API, with a film of it answering.
OpenAI Developers
Give your app real-time decision-making with Decisions API, powered by GPT-6 Luna.
Define questions and possible answers to classify content, route requests, or choose an agent’s next action.
Available in limited preview.
Theo Browne, who runs T3 and a YouTube channel about building software, reading the price. The reasons are Theo Browne's reading, not something OpenAI has said.
Theo Browne
I'll say the quiet part out loud: GPT-6.1 Sol was not meant to be this cheap. Clearly they cut the price a ton to fight back.
Likely also the reason they cut the "dollar amount" for the plan limits, since they cut into their margins so hard with this model
The same afternoon, from Theo Browne again, on a long-running project, a port of the TypeScript compiler to Rust. A benchmark and a real codebase do not always agree, and both are on this page.
Theo Browne
Making steady progress on ts-rust 🫡
I feel obligated to say that I threw GPT-6.1 Sol at this for a few days and it ran in circles and made no progress 🙃
Ethan Mollick, who studies AI at Wharton, after a short time with dots before the launch.
Ethan Mollick
I only used Dots briefly before launch, so can’t offer a detailed review, but I found it to be good & very much in the rapidly-expanding Clawlike category with Muse & Grokbot. Using a capable model with access to your data as an assistant & second opinion is remarkably useful.
Philip Kiely of Baseten, a launch partner of the OpenAI Marketplace, on open models inside Codex. Answering questions under the post, Kiely said it is for enterprise customers through the Marketplace and not self-serve.
Philip Kiely
Enterprise teams can now use open models like GLM-5.3 Flash and Kimi K3 natively in Codex and count spend against their OpenAI commit.
