Co-founder of Standard Agents (🤫). Compulsive open source builder. Creator of dmux, ArrowJS, FormKit, AutoAnimate, Tempo, zodown...
'merica
- Tweets 6,530
- Following 772
- Followers 13,894
- Likes 4,634
It still takes a lot of engineering (zero coding) to build something truly fresh and valuable in the world.
1
1
23
896
You’re breaking my heart Apple…
“You either die a hero, or you live long enough to see yourself become the villain.”
3
8
1,727
Do I know anyone at npm that can look into this? It’s been hours and is breaking our deploys…
3
9
2,486
I am a plebe among these greats. I seriously don't know why you wouldn't go to this conference...
- Peter (@steipete)
- Shawn (@swyx)
- Evan (@evanyou)
- Fred (@FredKSchott)
- Michelle (@michellechen)
- Sunil (@threepointone)
- Dillon (@dillon_mulroy)
- Kenton (@KentonVarda)
- Kent (@kentcdodds)
- Tanner (@tannerlinsley)
...and way way way way more cloudflare.com/connect/speak…
1
2
17
3,167
The freaking irony...SO much effort on "sandboxing" agents just for everyone to willingly give root access to their whole life.
78
49
1,931
102,308
Cloudflare just shipped a Jev alternative: Clef! So I swapped it out in our Tesla FSD simulator.
The good:
- It works! The car drives and reaches the destination!
- They open sourced it!
- It has vision! I’ll post more on awesome things this unlocks soon!
The bad:
- driving is notably more jerky as 1/20 choices it makes seem almost random.
- It ended up costing more than double Jev (still very cheap) for the same drive. Need more reps to prove if this is consistently true.
The (surprisingly) ugly:
- RTT was 4x slower than Jev, even when called within Cloudflare worker. I'm guessing this will get worked out over time cause @michellechen kicks ass!
introducing 𝚌𝚕𝚎𝚏: our first models trained by @cloudflare's workers ai team.
today, we're releasing two fast and accurate decision models that top the benchmarks for quality and latency. use them hosted on workers ai or grab the weights from @huggingface, because we open-sourced it too.
blog.cloudflare.com/clef-dec…
26
9
243
27,913
My first community note!
I maintain it is creepy, Minecraft ending poem, pre-programmed, or not.
Fun, but creepy.
This is so creepy.
Community note
This is an intentional Easter egg. The America.gov chatbot contains a prewritten bureaucratic remix of Minecraft's End Poem, triggered by typing "play Minecraft".
runtimewire.com/article/scoop-…
techcrunch.com/2026/09/29/ame…
en.wikisource.org/wiki/End_Poem
4
1,315
This is so creepy.
1.) Go to America.gov
2.) type “hello”
3.) type “play Minecraft”
4.) watch the agent’s reality implode
5.) ????
6.) PROFIT!
Community note
This is an intentional Easter egg. The America.gov chatbot contains a prewritten bureaucratic remix of Minecraft's End Poem, triggered by typing "play Minecraft".
runtimewire.com/article/scoop-…
techcrunch.com/2026/09/29/ame…
en.wikisource.org/wiki/End_Poem
132
72
2,111
775,819
I had early access to OpenAI's new "dots" and I have thoughts:
1. It's more work-focused than Grok Bot or muse. Think a bit less personal assistant and a bit more Codex orchestrator and Slack collaborator.
2. It's great in Slack. This is where it really shines. Your team can message your bot directly and your bot can take actions, like spin up new panes.
3. You can have a "call" your dot and this is incredible! So much of what makes personal agents magic is the way the experience is packaged. Grok bot didn't have any specific "new" technology, it was just a brilliant way to deliver it. Phone calls feel the same to me! Just naming it a "call" is genius. In the preview it was a bit slow, this may improve in production, but I loved be able to chat with it while I drive to work and my hands aren't available.
4. It should have been a separate app. ChatGPT is great. Codex is great. They are very different products, imo, and dots are way more Codex than ChatGPT.
5. As a personal assistant, its not as good as Grok Bot or Muse...yet. From day one Grok Bot had an outrageous plugin ecosystem that just works. "dots" are a lot less clear. They will occasionally give you little cards to auth your email or something like that, but often you have to go install a plugin in settings, and then your dot magically has access. I also don't love this lack of separation. If you auth an account in ChatGPT your bot has access to it, even if you didn't explicitly give it access.
6. Payments don't really work yet. The like @sama gave on stage about getting you tickets is...sorta true...its as true as it was with ChatGPT, but there are no real payment integrations yet.
27
8
161
33,264
Prediction: we'll see more and more AI products that do not let you pick the model.
1. It's a bad user experience for everyone except power users.
2. People care about efficacy, so they always pick the most expensive (especially in a subscription), and then complain about limits.
3. Model routers don't work, but agents can mask where the inference is occurring for various tasks while still offering great efficacy.
4. Individual agents are instances. Businesses can sell access to `n` agents without inference limits.
This is how your subscriptions allocations will work in the future.
3
13
2,218
As painful as it will be (and it will be painful) reducing subscription subsidies is actually a net positive in the long run.
There are a thousand startups whose products just don’t make sense while these subsidies last.
Hi,
Tomorrow we are re-opening the Pro $200 subscriptions to new subscribers, but together with it we are also changing how we calculate the usage for it. In effect, if you do the math, it will net out at half the dollar in API spend compared to the old Pro $200 plan.
Now that it's said, let me explain why this is happening and why you will still get more work done than if you were on the Pro $200 subscription one month ago.
(a) We didn't want to compromise in other ways and are committing to not reintroducing the 5h limit, so that you can fully use the weekly usage when you want.
(b) On the subscription, we guarantee that over time you always get more work done and with an increasing level of quality. This means that you will continue to get more value per dollar spent as a result of models getting more efficient and us passing down the improvements in the form of API price reductions.
(c) We don't want to put an incentive on ourselves to artificially inflate the API list prices to make it look like you are getting a lot (and workaround it through discounts, etc). Instead we want to continue to both rapidly reduce prices and increase capabilities of models on the API. This week we introduced GPT-6 Sol and GPT-6 Luna at 50% of their previous price. Over time, we see prices go low enough that it makes sense for most to buy usage as needed without there being a significant gap between what you get in a subscription and what you get in the API for a dollar spent.
(d) Tomorrow, we are adding more things to the subscription that won't draw on the usage, I won't reveal what that is yet.
I wanted to be transparent before all the big announcements tomorrow. Lots of new exciting things are coming to the subscriptions that will make it super compelling, but I wanted to make sure to share this change ahead of time so you can all understand it before we shower you with good news.
Codexingly,
Tibo
21
1
45
4,991