Building your AI computer in the cloud @usetoyo. Prev founded @pusher, @thoughtbot
Joined November 2006
- Tweets 17,454
- Following 284
- Followers 5,164
- Likes 11,406
662 Photos and videos
Might try using Pi with this endpoint for planning anything complex
Jun 13
Introducing the Fusion API, the smartest compound model in the market.
Fusion achieves Fable-level intelligence at half the price.
How it works 👇
1
404
Jun 14
My recent thinking has been that LLM compatibility and recovery stuff is best pushed out to the LLM gateway and LLM provider APIs.
I.e. we’re redesigning @usetoyo around Responses API. We’re stripping out abstraction layers from our code. For non OpenAI models we use @OpenRouter compatibility. Responses API also finishes requests server side even if client cancels the request.
Jun 14
weird feeling: started the morning thinking I'll write a small blog post on something I've been experimenting with (inference buffers with durable objects, so you can have truly resumable llm calls even if the model provider doesn't support it, see previous experiment here github.com/cloudflare/agents…) and it's leading me down a rabbit hole to add some brand new features on top of workers-ai-provider based on the new ai platform stuff (blog.cloudflare.com/ai-platf…) and while you'd think, yeah this is all achievable with coding agents, my recent experience knows this is actually exhausting and will suck up time. like, I'll have to do it anyway, but now I'm wondering whether I want to spend my sunday on this. decisions, decisions...
(ps: super glad cloudflare just lets me tweet out whatever without worrying about it being a secret etc. I think there's value in working in public.)
let's aim for a blog post by EOD. maybe. I'm also watching Sneakers (1992) in the background.
2
2
1,531
Jun 14
CF gateway with stateful Responses compatible api would be S tier (even Openrotuer doesn’t do stateful).
1
103
Damien C. Tanner retweeted
Jun 13
Looking for a good pod for the weekend?
Give napkin math a go!
Jun 11
Listening to users almost killed my company.
So this year...
We made the hard call to reinvent everything.
Here's how it saved us & sent our growth through the roof👇
2
1
18
2,908
Jun 13
now I have to actually build that treehouse I’ve been taking about
Jun 13
this seriously messes up my weekend plans i may have to actually see people
167
Jun 13
Feeling very precious now about the lovely code I managed to crank out the last few days
Jun 13
The US government, citing national security authorities, has issued an export control directive to suspend all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees.
The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance.
Access to all other Claude models is not affected.
We apologize for this disruption to our customers. We believe this is a misunderstanding and are working to restore access as soon as possible.
Read our full statement: anthropic.com/news/fable-myt…
1
176
Jun 10
You're gonna be meeting lots of tired looking engineers this month as we all prompt Fable like crazy until the subscription time ends!
1
108
Damien C. Tanner retweeted
Jun 10
Feeling pretty good about things
222
98
4,266
397,805
Damien C. Tanner retweeted
Jun 4
I get the instinct to pattern-match @Cloudflare into ‘big tech.’ But this thing isn’t like the others.
3
4
80
31,497
Damien C. Tanner retweeted
The best tools to build the agentic future: all native on @Cloudflare. Welcome to the team, VoidZero!!
Jun 4
VoidZero, the team behind Vite, Vitest, Rolldown, Oxc, and Vite , is joining Cloudflare. Vite stays open source, vendor-agnostic, and built for everyone. cfl.re/3Q1XYSX
13
29
519
46,885
Damien C. Tanner retweeted
Jun 2
An internal Slack post I shared with the Amp team a few weeks ago:
Building software is learning.
registerspill.thorstenball.c…
11
32
291
42,918
May 24
Latest model releases have been a real let down. Haven’t felt any real step change improvements. It feels like the labs are having a hard time wrestling the best of intelligence. Each release is not really better but just ‘different’ in some helpful but also counterproductive ways.
May 23
this has been my experience as well. there definitely were improvements, specifically wrt shell based computer use, but also regressions, especially in the last 3 version bumps of flicker and gerpertee corp.
the last step change was 3.x to 4.x in flicker land, probably mostly due to them getting all coding sessions from april to october 2025 via CC. similar timeline with GPT and Codex.
at least in my line of work, no big jumps after that. the benchmark increases mean literally nothing in the real world.
i suppose we have a data problem now. only so much you can RL into those damn things. and with ralph loops/swarms/agents reviewing agents/whatever, you get less and less human signal to improve RL, would be my uneducated guess.
also very hard to capture design/system thinking in RL would be my guess.
all that said: if we are at the top of the S curve now, then i'll take what we got. plenty useful, even if it won't replace me fully nor partially anytime soon.
3
426
May 23
Recently been feeling a need for a ‘Human Reviewed’ badge for infra providers and critical libs.
If I’m using your thing for critical infra. I want it to be extremely solid and safe from the temptation to slop.
May 23
This is peak 2026 - 180 retweets, 1.3k likes, 1.2 bookmarks, not a single issue on the repo.
So I looked at what's actually in here. I was not impressed.
Since none of you will go to github anymore, here's a screenshot of my questions.
github.com/workos/auth.md/is…
1
259
Apr 24
Matt's talk was the highest value things at AI Eng Europe Conference. Changed my Codex flow and improved code quality a lot after implementing some of his suggestions.
Apr 23
A talk I gave a few weeks ago.
Software fundamentals matter more than ever. Here's why:
youtube.com/watch?v=v4F1gFy-…
1
1
4
1,992
Apr 15
1
12
783
Apr 15
How does Peter deal with thousands of open PRs?
He set up an agent to read Discord, correlate complaints with open submissions, and hand him the five PRs to focus on each day.
Agents triaging and helping humans focus and apply judgment is how many AI-native teams are starting to operate.
1
1
215
Apr 15
Full interview → toyo.ai/blog/peter-steinberg…
Full video → youtube.com/watch?v=0MAvyd-J…
Thanks to @tessl_io for hosting and @mastra for recording.
1
226
Apr 15
Great to talk at @Cloudflare Connect today and share a bit about how @usetoyo works under the hood - hint, it's Durable Objects all the way down.
1
2
11
2,517
Mar 31
The core challenge I think we’re all having with coding agents in prod systems is they just want to write code. Lots of it. For every edge case and fallback.
LLMs are trained to give you an output no matter what. Rarely does the LLM say ‘nah you don’t need that feature’.
Once the code is written, it often feels more work to then remove it. So it’s tempting to just accept.
One tool we’re using to counter this is a simple spec based dev process. It lets us review and cut down a coding agent’s scope of work, before it goes wild generating the code.
1
1
6
820