Vlad 🇧🇬🇬🇧🇺🇦

Most teams don’t have a documentation problem. They have an access problem. The answers are in the system — but only developers can reliably get to them. Everyone else has to ask. We built Dawn to remove that dependency. It connects to your codebase and surrounding systems, and lets anyone ask questions and get answers grounded in how things actually work. From Slack, Teams, Telegram, etc. Developers have Cursor, Codex and ClaudeCode. Everyone else now has Dawn. dawnhq.ai

Shipped JavaScript execution engine in Dawn this week. The short version: agents can now run short JavaScript scripts inside a conversation when a task is better suited for an algorithmic approach. Models are uneven at the kind of mechanical work where the answer has to be exact. Ask one to count the rows in a CSV missing an email, and you'll usually get an answer that's close, occasionally right, and once in a while confidently wrong. Scripts handle the mechanical part. The agent decides when one will help and explains the result back in plain language. We made a deliberate call to keep the runtime small. No VM, no container, no remote sandbox. A typical script returns in milliseconds, so it feels like part of the conversation. The tradeoff is that scripts are scoped to data and file work. They don't install packages, run builds, or browse the network. For those, Dawn has other tools. Read the announcement at dawnhq.ai/blog/javascrip… #AI #AgenticAI #FutureOfWork #FutureOfCode #AIFirstCompany #AskDawn

@KaiXCreator Yup, on Codex when I have an influx of ideas I basically run several spec sessions while other agents are building the previous ideas.

Does anyone actually run 10 agents in parallel?

@gdb Not recently I’m afraid, it’s gotten really slow lately and it’s eating through tokens like they’re free :)

GPT-5.5 is a uniquely good coding model

@code_coded Yes, super slow and slightly dumber. Hopefully this means 5.6 is coming, lol

Anyone else feeling that Codex 5.5 has become frustratingly, near unusably slow for even simple tasks? Last 2 days it's been just horrendous. Having to jump back onto backup Claude account...maybe I cancelled Claude too earnestly =/

@OlleMattsson @thsottiaux I have the same experience, weekly limits on fast get drained in 2 days when before I could get a full week’s worth of work.

@thsottiaux The weekly limit still seems to go down way to hard on /fast mode. I wish i had the number to back this up.

Team is investigating elevated latencies across the board for API and ChatGPT. Updates will be posted on status.openai.com

@thsottiaux Well, Codex has been super slow for the last 2 days at least.

This is completions API only and does not affect Codex.

@thsottiaux @Dimillian It’s been super slow since at least yesterday. Just had a bugfix take 2 hours, this is not normal

@jturntdev I use both. Codex is nicer for long running tasks, worktree management, parallel work. I do prefer the CLI in VS Code for bugfixes, investigation and quick changes though.

Those who switched from the Codex CLI to Codex app. What actually made you switch? What does it unlock that the CLI doesn’t?

@TravelerOfCode Yes, that’s what pushed us to improve it as workflows started more complex. Serverside is the source of truth, we then let users reconcile visually conflicting edits.

@vladpetroff visual canvas for workflows is the kind of thing you dont realize you needed until you have it. text-only graph definitions get unreadable past 6-7 nodes. how are you handling the merge UX when two people edit the same graph?

Workflows have been the part of Dawn I've been waiting to get right for months, and the visual canvas we shipped this week is the version I actually want to use myself. The honest backstory: when we first launched event-driven workflows, you configured them as a single trigger plus a prompt. It worked for the basic PR review case, but anything with branching, retries, or "if this happens then ask the agent about that" needed a builder. So we built one. What surprised me during testing is how much faster it is to just describe what you want and let Dawn draft the workflow on the canvas. You say "review every PR, flag changes to billing code, notify me if a migration is touched", and the agent lays out the trigger, the steps, the branches, the output. Then you tune it. That's the pattern we keep coming back to in this round of releases: agents that help shape the workspace they work in, with the team staying in control of what actually lands. Read the announcement at dawnhq.ai/blog/visual-wo… #AI #AgenticAI #FutureOfWork #FutureOfCode #AIFirstCompany #AskDawn

@mitchellh It takes great discipline to give up something you love to do so much. Respect!

Here's the ATC audio clip of my final landing ever. From humble beginnings in a DA40 to ~600 hours in a Vision Jet at 31,000 feet, I love to fly. Due to my growing personal and professional obligations, my final and best ADM decision as a PIC was to stop. This was my toughest landing. The skies were clear. The wind was straight down the runway. It's my home field I've landed at hundreds of times. But I knew it was my last. You can't hear it in the clip but my voice felt shaky. I'm doing my best to focus on a safe landing but I knew that was the last "cleared to land" readback I'd ever give. The landing was smooth. I parked the jet, locked her up, and gave her one last pat on the nose. We had good times. I'm no longer a pilot. Locking in for what's next. 🫡

Is Codex super slow today or is this just me?

Today we shipped Voice Intelligence in Dawn. Record a meeting, attach an audio or video file, or dictate a prompt, and Dawn gives you back a transcript with speakers separated, ready to do something useful with. This one's been on my list for a while. Almost every team I talk to records the important conversations (sales calls, design reviews, planning sessions, customer interviews), and almost none of them get any real value back from those recordings. The transcription tools they use are decent, but the transcript usually ends up in a Google Doc nobody opens. What changes when transcripts become part of your agent's working context is that you actually do things with them. Action items get written up. A decision becomes a ticket with the rationale attached. The customer call turns into a follow-up email in your voice. All of that happens in the same Dawn you're already using for everything else, in the same workspace, with the same agent. No copy-paste between tools. That's the bet, anyway. Curious to see how teams use it. Read the announcement at dawnhq.ai/blog/voice-int… #AI #AgenticAI #FutureOfWork #FutureOfCode #AIFirstCompany #AskDawn

For a long time, the limit on how useful Dawn could be was the box it lived in. Slack, Teams, the CLI, Dawn Web. They all work, and they all need someone to bring the context into the conversation. When the thing you wanted help with was a page in front of you, that meant describing it. Or screenshotting it. Or pasting a URL and hoping the agent could fetch it. Today we shipped the Dawn browser extension. It puts Dawn beside the page you're actually looking at. The part I'm proudest of: you can pin tabs the agent keeps working with after the popup is closed, even from a thread in Dawn Web on another machine. That changes what the browser is. It becomes a place Dawn can act on whenever it needs to. Worth installing if you spend most of your day in tabs. Read the announcement here: dawnhq.ai/blog/dawn-brow… #AI #AgenticAI #FutureOfWork #FutureOfCode #AIFirstCompany #AskDawn

@sama So you added screenshot attachments to Codex. I hope you are working on something really interesting that just didn’t cut it for this Thursday because screenshots are hardly groundbreaking…

new codex ships today!

The GOAT prompt of all prompts, amirite #Codex

Can’t reference files with @ in ChatGPT Codex @thsottiaux it shows me chatgpt plugins and not Codex files :( minor hiccup though, otherwise it works great 👍

@steipete This is super cool and landing just as I'm building a semantic code search engine to optimize token usage and latency for my cloud agents

🩹 clawpatch 0.1.0 is live: Clawpatch maps codebases into semantic feature slices, reviews them for bugs and quality issues, and records explicit fix attempts with validation. You'll be surprised how much this will find. npm install -g clawpatch clawpatch.ai

@steipete Absolutely, I run it multiple times until it either comes up with nothing or starts pulling non-existent issues out of thin air. But it takes a lot of babysitting

All I want is codex automatically entering /review mode after it's done and just looping until it stops finding booboos. (Yah I'm gonna build that)

@thsottiaux @steipete I thought ChatGPT was supposed to be the superapp for AI? What happened? Why u no love ChatGPT no more :(

Now that the Codex app is close to being the super app. What should the super duper app do?