
The AI-Native Cloud. ☁️ Status: @DOstatus Support: https://nitter.cf/t.co/5gkvyionbi
CO
Joined January 2012
- Tweets82.8K
- Following287
- Followers227K
- Likes23.3K
Pinned Tweet
Your agent wants permission to run a migration and you’re not at your desk.
Cuttlefish, our iPhone app for Managed Agents, puts the approve button in your pocket with your agent's requests on screen. Send a correction mid-run, or stop it if it goes sideways. do.co/3UkcsA1
Jev sent half of 661 real agent tool calls to a human for review. A 26B open-weight model sent 8%. Same calls, five models compared. do.co/4C89Tll
3
2,042
A typical agentic session runs 43 turns with 142K tokens in and 444 tokens out. Serving that well comes down to the cache.
@yifandotqiao of @Inferact (the team behind vLLM) on taking one card from 84K to 480K tokens a minute on @Kimi_Moonshot K3. do.co/4zJtvdv
$18M per megawatt across the entire fleet, not one customer or one data center. @paddix on how DigitalOcean’s focus on software products impacts the economics of AI infrastructure.
Your backlog has a dozen small bugs nobody wants to pick up.
Tag them agent and let Harness Runtime write the fix, run your tests, and open the PR. You just review it.
Full tutorial from @r_o_connor: do.co/4sXNXFs
Open models are six months behind the frontier. @latkins of @arcee_ai says that matters less than you think: if a task has never been done, use a closed model. If you run it daily or weekly, open will do.
Closing panel at Open Intelligence Summit. do.co/48jCR4h
DigitalOcean retweeted
Who is going to @TomorrowXSummit next week in Austin!? @kevinvangundy & I are organizing a dinner at one of my fave new ATX spots. @digitalocean has also has some extra passes to giveaway.
Interested in joining? let me know!👇🏻
A coding agent ran 40 times and opened 40 working PRs.
All 40 hit the same bug. The sandbox had no git identity, so every first commit failed and the agent invented one.
DigitalOcean Signals caught it, and a Simulation proved the fix before production changed. do.co/4AkCemM
“Agents need memory, agents need data services, agents need complex orchestration, agents need secure operating environments called sandboxes.”
@paddix on the software it takes to run agents in production. On @theinformation’s TITV.
Everyone has figured out how to generate code. Almost no one has figured out how to QA it at scale or monitor it in production.
@rbren_dev of @OpenHandsDev on agents that start work on their own, and the cloud they run in. do.co/3WqfseS
DigitalOcean retweeted
QM, @ycombinator's open-source agent harness, now deploys in one click on @digitalocean: compute, sandboxes, inference, database and storage.
One cloud. One bill. Your choice of models.
“DigitalOcean makes it easy to get started.” @ycombinator Labs’ Joshua France on the first one-click deployment for QM, YC’s open-source agent harness.
Check out QM on DigitalOcean today: do.co/4k99Sr2. To learn more: do.co/3VdWasG.
“DigitalOcean makes it easy to get started.” @ycombinator Labs’ Joshua France on the first one-click deployment for QM, YC’s open-source agent harness.
Check out QM on DigitalOcean today: do.co/4k99Sr2. To learn more: do.co/3VdWasG.
The agent sends. The key stays out.
@neimuc27 let an AI support agent reply on its own without ever holding the credential. 57 of 335 customers got answers with no human click, and nothing sensitive went out.
digitalocean.com/community/t…
Last time, my AI support agent sent zero replies. On purpose.
This time it sent 57 of 335 on its own. Nothing sensitive went out.
It never held the credential. But two things still broke that I didn't see coming: a policy that never fired, and an attack the agent caught and still let through.
nitter.cf/neimuc27/status/211517…
DigitalOcean retweeted
Check out Claude Fable 5.2 on @digitalocean serverless. We have faster Day 0 access on our serverless platform than I have on my Claude desktop. Go figure !!
Claude Fable 5.2 from @AnthropicAI is live on DigitalOcean Serverless Inference.
Anthropic calls it its most capable GA model for technical work and reports output up to 30% faster than Fable 5.1. 8/40 per 1M input/output tokens. One account, one bill. do.co/4AvIpW3
DigitalOcean retweeted
.@digitalocean Cuttlefish looks absolutely amazing on iPhone Duo! You can now work side by side with your Artifacts and prompt your agent. Install it new update out!! do.co/cuttlefish
Claude Fable 5.2 from @AnthropicAI is live on DigitalOcean Serverless Inference.
Anthropic calls it its most capable GA model for technical work and reports output up to 30% faster than Fable 5.1. 8/40 per 1M input/output tokens. One account, one bill. do.co/4AvIpW3
DigitalOcean retweeted
A shift most of the industry hasn't priced in: agents are becoming the buyer of cloud. About half of the unique visitors to digitalocean.com are now agents, not people.
And they don't just run on the cloud. They select it, sign up, and deploy. More than 37,000 unique customers have already deployed on DigitalOcean through an agent this year.
For fifteen years, cloud was designed for a human reading a console and a docs page. An agent can't wait for someone to interpret a dashboard, and it can't file a ticket. It chooses the platform it can read, reason about, and use end to end. Clear APIs, auth, and billing are now the front door.
I think this will prove as consequential as self-serve cloud. Self-serve took the sales rep out of the buying loop. Agents take the human itself out of it. We are building for agents as builders and as buyers, and we are increasingly winning both.
Fast inference goes way beyond KV cache tuning.
@DebarshiRaha breaks down how we co-designed the full stack with @vllm_project: precision selection, kernel fusion on DeepSeek’s attention path, programmatic dependent launch, and speculative decoding. 🔧
Fast #inference goes way beyond just KV cache tuning.
Here’s a deep dive into the full-stack co-design by @digitalocean and @vllm_project:
• Hardware features & precision selection
• Serving engine architecture
• Kernel fusion (deep dive into the DeepSeek attention path)
• Programmatic Dependent Launch (PDL)
• Speculative decoding & MTP
Hope you learn some of the concepts here! Full video below 👇
We timed four models on DigitalOcean Serverless Inference for 48 hours, one request at a time.
On ~1,000-token prompts they all sit within a third of a second of each other at the median.
At the slowest 1 in 100 they split. Three stay under 3.5s. @deepseek_ai V3.2 takes 13.65s.
A 10s first-token timeout covers the first three. DeepSeek needs 30s. do.co/4zw7UFW








