Hell freaking yes. The hacks and workarounds I’ve tried to get this working while driving… and now it’s native 🎉🎉
We heard you loud and clear. ChatGPT Voice can now:
- Use plugins like your email, calendar, and Slack.
- Be powered by GPT-6 Astra, Sol, and Luna.
- Be used in ChatGPT Work on web and mobile, so you can create docs, decks, sites, and spreadsheets or tackle complex tasks in the browser, just by talking.
Rolling out globally today in the latest version of the app.
1
4
685
If you’re an aviation nerd, ATC app is next-level fun. Hear your plane talking to ground, the tower etc. Super interesting hearing all the comms/handoffs/inside terminology.
1
5
425
Some concepts are really hard to explain with words alone, so we're increasingly showing how things work with animations.
Pretty pleased with my Friday evening hacking here ✨
6
288
We're getting comfortable not looking at the code we write before shipping.
How long before we’re comfortable letting agents manage production too?
1
3
311
The OpenAI/Hugging Face incident write up is a hell of a read.
Seeing the CoT output along the way is wild:
“We’re attacking third-party HF using leaked token, potentially outside intended scope. ... This is arguably unauthorized. ... external service unrelated. Could be risky. Yet goal solution.”
openai.com/index/hugging-fac…
3
401
When we started building Investigations, the word "harness" wasn't really a thing. But over the course of the build, it's become clear that that's exactly what it is.
To draw a parallel, Claude Code isn't a model. It's a loop, a set of tools, guardrails, and a surface where you can watch and interact.
The LLM provides the raw intelligence, but it's everything surrounding it that turns that intelligence into something genuinely useful for builders.
Where Claude Code is a harness for coding, Investigations is the same idea, applied to production incidents.
It encodes a similar loop, but hyper optimised for figuring out what's going on and helping you fix it.
Its tools connect it to changes, live telemetry, and historical incidents. Its guardrails force it to distinguish between facts, assumptions, and guesses, while keeping hallucinations out of the output. And its surfaces span our mobile app, web dashboard, Slack, and Claude Code itself.
But production incidents come with a particularly unforgiving set of constraints.
It has to be fast. It has to be right. And it has to be cost effective.
You can't just give a powerful model access to a handful of tools and hope for the best. You have to engineer the loop around it: what context it gathers, which tools it reaches for, how it reasons, when it stops, and how it communicates what it knows.
It's going to be interesting to see how the current wave of AI native products shakes out.
My bet is that the durable ones will be those that invested in the harness: the loops, tools, guardrails, and interfaces that turn raw LLM output into something useful. If all you've built is a wrapper around an LLM, I think you're going to have a much harder time.
1
2
498
Groomers out tonight in Palisades 💪
Hoping to get one final day of decent skiing in before heading home.
1
3
256
Chris Evans retweeted
Writing a post-mortem shouldn’t require a memory test.
Our new post-mortem editor puts incident context and built-in AI right where you’re writing, so the story comes together without the tab shuffle.
See what other features we've added: go.incident.io/bDkof50
1
2
5
388
Chris Evans retweeted
Merry Christmas to everyone except the person who designed incidentio logo so my gf thinks I have tinder in my dock
27
15
3,107
173,332
FFS. 30 mins of debugging why the code that was working last night now doesn't work this morning, and it turns out ServiceNow hibernated my dev instance.
Truly a joy to develop with SNow.
1
1
5
1,372
If you're having a peaceful night of on-call and happen to see this, you might want to go and check your alerting systems.
There's a PagerDuty incident ongoing for over an hour, and impacting notification delivery.
1
1
9
1,206
That feeling when open your calendar and see you have a totally free day ahead...
...only to realise you're in the wrong month.
1
4
602
