A hypothesis: agents don't need a sandbox. They need a platform that thinks, acts, remembers, and improves as one system.
There's a debate running this week about sandboxes. One camp says sandboxes are dead and the future is a private cloud for every agent. Another says a sandbox is a compute primitive, like Postgres or Redis, and primitives don't win markets, platforms do. A third says sandboxes are a winner-takes-most market. I think all three are arguing about one layer of the problem.
My hypothesis is that agents are the first workload that exercises every layer of a platform inside a single loop. They think, they act on real systems, they remember, and they improve.
A web server touched compute and maybe a database. An agent touches inference, execution, tools, storage, retrieval, telemetry, and evaluation, in one task, often in less than a minute. So what an agent needs isn't a better sandbox. It's a platform where those four things run as one system. If that's right, the unit of competition isn't the sandbox and it isn't the model. It's the integrated cloud for agents.
Whoever builds one that's compelling across agents, data, and model access at the same time wins.
Let’s start with how an agent actually runs. One “prompt” fans out into a dozen model calls, a web search, a database query, a code execution, a tool call with a credential, a checkpoint, and a trace. Today a typical team buys those from five or six vendors and glues them together. Six vendors means fifteen seams. Every seam is an integration someone maintains, a credential someone stores, a network hop that adds latency, an egress fee, and a place where the trace dies.
Point solutions are excellent but partial: most sandbox companies don't run inference, and model companies don't manage agent memory or observability. The loop is what breaks on a stitched stack.
Models have to be routable, and routing only works when inference sits next to the agent. Every few weeks a new model tops a chart. Open-weight models went from <20% of token volume on our Inference Engine in April to >85% by August.
My read: the model is the most swappable layer in the stack, so you want a router in front of it, not a hard-coded choice. We built an Inference Router for that. On Kimi K3 we see an aggregate cache-hit rate above 90% across our own workloads. The router has to know what the inference layer has cached, and the cache has to sit next to the sandbox that's going to need it. That's a function call on one platform.
The agent and harness layer is where the business logic lives, so the runtime has to be built for it. I believe the harness, its prompts, its tool permissions, its memory, and its evals will hold more of a company's business logic than the model ever will. The model is rented. The agent is yours. So we built Managed Agents from the ground up as purpose-built primitives, not VMs with an agent sticker on them. Every session gets its own Firecracker microVM, resumes from pause in a few hundred milliseconds, and is billed for active CPU at $0.044 per vCPU-hour, so the wait costs zero. Action Gateway gives agents governed access to 20K+ tools across 600+ providers through one MCP endpoint, matched intent to tool with 99%+ accuracy in our testing, and brokers credentials at call time. The agent never sees the key. You can only make that promise when the runtime, the gateway, and the secret store are the same system.
Data and recursive loops are what make an application more intelligent over time. This is the part I believe in most and the part with the least proof so far, because the products are weeks old. The hypothesis: two teams start with the same model and the same harness, and within a quarter they diverge, and the one with the loop wins. On DigitalOcean the loop runs on two services. Agent Data Services is one governed catalog where video, documents, logs, and agent sessions land agent-ready and searchable by meaning. Agent Learning Services closes the loop: Insights records every step, Signals flags the sessions where the agent stalled, looped, or lost the user, and Evaluations and Simulations test the fix before a customer sees it. In an internal demo, an optimized support agent went from failing to passing it with 90% fewer input tokens.
Intelligence stops being something you rent via models and becomes something your application accumulates, tuned to the exact problem it's solving.
And the whole thing has to go down to the floor. DigitalOcean builds the routing model and the microVM runtime operates the GPUs they run on, across 20 data centers with NVIDIA HGX B300 and AMD MI350X fleets.
Much of the agent infrastructure market rents GPUs from someone who rents a data center from someone else, with margins stacked on margins. My view is that active CPU billing, a warm KV cache next to the sandbox, and a router that can see the inference layer are only possible when you control the data center, the hardware, and the software as one stack.
The obvious objection is that "integrated" is a polite word for lock-in. I'd argue the opposite, and I think it's the real test of whether a platform is integrated or just bundled. Integration should be about removing seams, not closing doors. Bring Claude Code, Codex CLI, OpenCode, Hermes, LangGraph, or your own OCI image. Hit one endpoint compatible with OpenAI, Anthropic, and 60+ Open models. Bring your own key for a frontier model, or run open weights we host. Our routing model and the proxy under it are open source.
The point is that the seams are gone, not that the exits are.
One early signal that the hypothesis might be right: roughly 70% of our large AI customers have attached a one or more platform primitives. Agents come in through inference or a sandbox and pull the rest of the cloud behind them. That's what I'd expect if agents really do need every layer at once.
So, back to the debate. The platform that wins won't be the one with the best sandbox, and I don't think it'll be the one with the best model.
My bet is on the platform where an agent thinks, acts, remembers, and improves as one system.
DM me if you are interested in trying our Agent primitives.
#digitalocean #agents #Agenticloop #learning #Context #sandbox #Runtime #DOCN
There's a debate running this week about sandboxes. One camp says sandboxes are dead and the future is a private cloud for every agent. Another says a sandbox is a compute primitive, like Postgres or Redis, and primitives don't win markets, platforms do. A third says sandboxes are a winner-takes-most market. I think all three are arguing about one layer of the problem.
My hypothesis is that agents are the first workload that exercises every layer of a platform inside a single loop. They think, they act on real systems, they remember, and they improve.
A web server touched compute and maybe a database. An agent touches inference, execution, tools, storage, retrieval, telemetry, and evaluation, in one task, often in less than a minute. So what an agent needs isn't a better sandbox. It's a platform where those four things run as one system. If that's right, the unit of competition isn't the sandbox and it isn't the model. It's the integrated cloud for agents.
Whoever builds one that's compelling across agents, data, and model access at the same time wins.
Let’s start with how an agent actually runs. One “prompt” fans out into a dozen model calls, a web search, a database query, a code execution, a tool call with a credential, a checkpoint, and a trace. Today a typical team buys those from five or six vendors and glues them together. Six vendors means fifteen seams. Every seam is an integration someone maintains, a credential someone stores, a network hop that adds latency, an egress fee, and a place where the trace dies.
Point solutions are excellent but partial: most sandbox companies don't run inference, and model companies don't manage agent memory or observability. The loop is what breaks on a stitched stack.
Models have to be routable, and routing only works when inference sits next to the agent. Every few weeks a new model tops a chart. Open-weight models went from <20% of token volume on our Inference Engine in April to >85% by August.
My read: the model is the most swappable layer in the stack, so you want a router in front of it, not a hard-coded choice. We built an Inference Router for that. On Kimi K3 we see an aggregate cache-hit rate above 90% across our own workloads. The router has to know what the inference layer has cached, and the cache has to sit next to the sandbox that's going to need it. That's a function call on one platform.
The agent and harness layer is where the business logic lives, so the runtime has to be built for it. I believe the harness, its prompts, its tool permissions, its memory, and its evals will hold more of a company's business logic than the model ever will. The model is rented. The agent is yours. So we built Managed Agents from the ground up as purpose-built primitives, not VMs with an agent sticker on them. Every session gets its own Firecracker microVM, resumes from pause in a few hundred milliseconds, and is billed for active CPU at $0.044 per vCPU-hour, so the wait costs zero. Action Gateway gives agents governed access to 20K+ tools across 600+ providers through one MCP endpoint, matched intent to tool with 99%+ accuracy in our testing, and brokers credentials at call time. The agent never sees the key. You can only make that promise when the runtime, the gateway, and the secret store are the same system.
Data and recursive loops are what make an application more intelligent over time. This is the part I believe in most and the part with the least proof so far, because the products are weeks old. The hypothesis: two teams start with the same model and the same harness, and within a quarter they diverge, and the one with the loop wins. On DigitalOcean the loop runs on two services. Agent Data Services is one governed catalog where video, documents, logs, and agent sessions land agent-ready and searchable by meaning. Agent Learning Services closes the loop: Insights records every step, Signals flags the sessions where the agent stalled, looped, or lost the user, and Evaluations and Simulations test the fix before a customer sees it. In an internal demo, an optimized support agent went from failing to passing it with 90% fewer input tokens.
Intelligence stops being something you rent via models and becomes something your application accumulates, tuned to the exact problem it's solving.
And the whole thing has to go down to the floor. DigitalOcean builds the routing model and the microVM runtime operates the GPUs they run on, across 20 data centers with NVIDIA HGX B300 and AMD MI350X fleets.
Much of the agent infrastructure market rents GPUs from someone who rents a data center from someone else, with margins stacked on margins. My view is that active CPU billing, a warm KV cache next to the sandbox, and a router that can see the inference layer are only possible when you control the data center, the hardware, and the software as one stack.
The obvious objection is that "integrated" is a polite word for lock-in. I'd argue the opposite, and I think it's the real test of whether a platform is integrated or just bundled. Integration should be about removing seams, not closing doors. Bring Claude Code, Codex CLI, OpenCode, Hermes, LangGraph, or your own OCI image. Hit one endpoint compatible with OpenAI, Anthropic, and 60+ Open models. Bring your own key for a frontier model, or run open weights we host. Our routing model and the proxy under it are open source.
The point is that the seams are gone, not that the exits are.
One early signal that the hypothesis might be right: roughly 70% of our large AI customers have attached a one or more platform primitives. Agents come in through inference or a sandbox and pull the rest of the cloud behind them. That's what I'd expect if agents really do need every layer at once.
So, back to the debate. The platform that wins won't be the one with the best sandbox, and I don't think it'll be the one with the best model.
My bet is on the platform where an agent thinks, acts, remembers, and improves as one system.
DM me if you are interested in trying our Agent primitives.
#digitalocean #agents #Agenticloop #learning #Context #sandbox #Runtime #DOCN
24
11
109
11,861
Vinay Kumar retweeted
I don't think people need to imagine it anymore; this is collaboration done right for agents. Go @Box and @chriskim_dev !!
yeah AI agents are no longer just chats anymore.
Imagine a sandbox agent opening cloud files with an ordinary file path.
That's Box Mount.
Box Mount gives sandbox agents ordinary file paths to cloud files in @Box without building a download and upload workflow.
In this demo, an reads deal room documents locally and writes a diligence memo back to the cloud.
All through Box Mount.
2
3
15
1,079
Vinay Kumar retweeted
This was such a fun proof of concept to see in person! Go @Box (@levie ) and @digitalocean (@paddix )
I just automated deal room acquisition screening with @Box & @digitalocean Managed Agents.
Drop a new file in the deal room → Triggers the sandbox agent to wake up and rebuild the review memo with the latest info.
No more stale deal room memos.
Full setup deep dive video coming soon.
0
4
21
717
Vinay Kumar retweeted
"If you commit to one particular lab, there's a good chance you're going to be left behind."
@OpenHandsDev CEO @rbren_dev on choice and owning your intelligence. Hear more from him at the Open Intelligence Summit, Oct 13 in SF: do.co/4ArVTlL
@OpenHandsDev CEO @rbren_dev on choice and owning your intelligence. Hear more from him at the Open Intelligence Summit, Oct 13 in SF: do.co/4ArVTlL
1
3
20
208,851
Years ago DigitalOcean's Droplet made the cloud something one developer could afford and get started quickly.
Today we're doing it again for agents.
Agent Droplets: one subscription for everything an agent needs. Sandbox, inference, memory, 16,000+ tools. Unlimited agents. No seat charges.
Pro: $50/mo
Team: $200/mo
Free trial: $5 credit, no card.
You pick a size and start.
Why we built it: digitalocean.com/blog/why-we-bu…
Get started: cloud.digitalocean.com/registrations/…
Today we're doing it again for agents.
Agent Droplets: one subscription for everything an agent needs. Sandbox, inference, memory, 16,000+ tools. Unlimited agents. No seat charges.
Pro: $50/mo
Team: $200/mo
Free trial: $5 credit, no card.
You pick a size and start.
Why we built it: digitalocean.com/blog/why-we-bu…
Get started: cloud.digitalocean.com/registrations/…
0
2
18
499
Proud to be part of the @nvidia Open Agent Safety Platform launch.Jensen's framing: "Safety and security require full-stack engineering."
That's how we built Managed Agents. The agent never holds a live credential. Policy designed to fit OpenShell.
More here - digitalocean.com/blog/how-digit…
That's how we built Managed Agents. The agent never holds a live credential. Policy designed to fit OpenShell.
More here - digitalocean.com/blog/how-digit…
3
1
4
281
Vinay Kumar retweeted
It's time to move your AI agents to the cloud.
Qencode built one for customer support triage. A request comes in, the agent wakes up, reads it, and files the ticket properly. @r_o_connor shows you how to build one just like it.
Bring the harness, model, and tools you already talOcean Managed Agents is now in public preview. do.co/4xN8zRM
Qencode built one for customer support triage. A request comes in, the agent wakes up, reads it, and files the ticket properly. @r_o_connor shows you how to build one just like it.
Bring the harness, model, and tools you already talOcean Managed Agents is now in public preview. do.co/4xN8zRM
3
4
22
5,299
Jev is having a moment. Every platform rushed to host it the same way: another model endpoint in the catalog, and good luck wiring it into your agent loop yourself.
We took a different route at DigitalOcean. In addition to Jev being available in our Inference engine, Jev is also Gateway as a first-class classifier tool, right next to browser automation, and 16,000+ other tools behind one governed MCP endpoint.
That distinction sounds small. It isn't.
Treat Jev as a model and your agent's LLM has to know about a second inference provider, manage another API key, and hand-roll the request plumbing. Treat it as a tool and any harness you already run (Codex, Claude Code, LangGraph) just discovers it and calls it, same as any other tool in the belt.
Ask typed questions: a yes/no, a choice from your own list, a score on your own scale. Get back decisions with calibrated probabilities, schema conformance guaranteed. Answers land as data your code uses directly, so agents route, filter, and escalate without parsing prose.
The gateway part matters too. Jev calls ride the same governance as everything else: scoped permissions, credentials that never enter the model context, low-confidence answers flagged for a human, every decision in the audit trail.
We ran a Codex agent through it this week. One YAML, one prompt, and it turned a day of X noise into a shortlist worth reading, with the uncertainty left visible.
If you want raw inference, Jev is on Serverless Inference too, at $42 per billion input tokens with output free. But agents shouldn't have to treat a classifier like a chatbot.
Models reason. Jev decides. Ship both. On the same platform.
We took a different route at DigitalOcean. In addition to Jev being available in our Inference engine, Jev is also Gateway as a first-class classifier tool, right next to browser automation, and 16,000+ other tools behind one governed MCP endpoint.
That distinction sounds small. It isn't.
Treat Jev as a model and your agent's LLM has to know about a second inference provider, manage another API key, and hand-roll the request plumbing. Treat it as a tool and any harness you already run (Codex, Claude Code, LangGraph) just discovers it and calls it, same as any other tool in the belt.
Ask typed questions: a yes/no, a choice from your own list, a score on your own scale. Get back decisions with calibrated probabilities, schema conformance guaranteed. Answers land as data your code uses directly, so agents route, filter, and escalate without parsing prose.
The gateway part matters too. Jev calls ride the same governance as everything else: scoped permissions, credentials that never enter the model context, low-confidence answers flagged for a human, every decision in the audit trail.
We ran a Codex agent through it this week. One YAML, one prompt, and it turned a day of X noise into a shortlist worth reading, with the uncertainty left visible.
If you want raw inference, Jev is on Serverless Inference too, at $42 per billion input tokens with output free. But agents shouldn't have to treat a classifier like a chatbot.
Models reason. Jev decides. Ship both. On the same platform.
12
13
81
15,181
Vinay Kumar retweeted
Our Staff Developer Advocate @amit took DigitalOcean Managed Agents for a spin.
He made up a coworking company called Atrium, then set it up the way a real team works: a backlog in @linear, a handbook in @notion, a channel in @SlackHQ, a repo on @github. Then he handed an agent alone.
It came back with a pull request, and a problem nobody had asked it to look for.
He made up a coworking company called Atrium, then set it up the way a real team works: a backlog in @linear, a handbook in @notion, a channel in @SlackHQ, a repo on @github. Then he handed an agent alone.
It came back with a pull request, and a problem nobody had asked it to look for.
DigitalOcean Managed Agents is now in public preview.
Run Claude Code, Codex, or your own LangGraph agent in a runtime environment that pauses when idle. Put its tools behind one governed endpoint, and pick from 75+ open and proprietary models. One cloud, one bill.
Prompts to n the blog: do.co/4ysh3it
3
2
22
271,613
Vinay Kumar retweeted
👉 Will AI make hosting invisible?
I had the pleasure of talking with @suhaib__zaheer , SVP & GM of Managed Hosting at @digitalocean and GM of @Cloudways, about future of the industry.
It is especially interesting because Cloudways became famous for managed WordPress hostthat positioning are being challenged at the same time:
- AI builders are creating a new generation of projects outside WordPress
- AI is starting to take over much of the work that made hosting “managed” in the first place.
That raises a bigger question for Cloudways and managed hosting in general. If AI can monitor, diagnose, fix problems and operate infrastructure in the background, what are customers actually paying a premium for? and that's what I asked Suhaib directly :)
In this episode:
🔹What happens when AI does the managing
🔹Why hosters may be chasing the wrong AI opportunity
🔹Why Cloudways needs to think beyond WordPress
🔹Hosting without tickets, manual fixes or dashboards
🔹How AI agents may replace the traditional hosting interface
🔹... and MUCH more!
Watch the full episode on @WebhostingToday and YouTube, or listen on Spotify and Apple Podcasts.
#webhosting #cloudhosting #wordpress #ai #hostingindustry
I had the pleasure of talking with @suhaib__zaheer , SVP & GM of Managed Hosting at @digitalocean and GM of @Cloudways, about future of the industry.
It is especially interesting because Cloudways became famous for managed WordPress hostthat positioning are being challenged at the same time:
- AI builders are creating a new generation of projects outside WordPress
- AI is starting to take over much of the work that made hosting “managed” in the first place.
That raises a bigger question for Cloudways and managed hosting in general. If AI can monitor, diagnose, fix problems and operate infrastructure in the background, what are customers actually paying a premium for? and that's what I asked Suhaib directly :)
In this episode:
🔹What happens when AI does the managing
🔹Why hosters may be chasing the wrong AI opportunity
🔹Why Cloudways needs to think beyond WordPress
🔹Hosting without tickets, manual fixes or dashboards
🔹How AI agents may replace the traditional hosting interface
🔹... and MUCH more!
Watch the full episode on @WebhostingToday and YouTube, or listen on Spotify and Apple Podcasts.
#webhosting #cloudhosting #wordpress #ai #hostingindustry
1
2
2
785
Today we shipped DigitalOcean Managed Agents into public preview.
Every era of the cloud has an atomic unit. For twenty years it was the provisioned VM. You came for compute and paid by the hour, working or waiting. Agents don't fit that box. They think in tokens, act in bursts mory (state) that outlives the session. Compute, inference, and data working as one.
The compute-first cloud meters the VM. The agent-first cloud meters the work.
I've talked to a lot of founders building agents, and I hear the same stack every time: a harness on a laptop, sandboxes from one startup, inference from two more, storage at a hyperscaler, search from a sixth vendor. Six vendors means fifteen seams to secure and operate. Nobody can say what a single run cost.
So we built Managed Agents deeply integrated with our Inference Engine.
Harness Runtime: every session in its own Firecracker microVM. Pause, resume, checkpoint, fork. Bring Claude Code, Codex CLI, OpenCode, Hermes, or your own image.
Action Gateway: one endpoint to 16,000+ governed tools. The agent never sees the key.
Pricing that matches the workload: active CPU billed by the second. An hour of agent work that bills $0.126 fully allocated costs about $0.031 with us for a typical agent run (25% active). While the agent waits, you pay zero.
Agents think, act, remember, and improve. Now the whole loop runs on one platform.
Public preview is live today. Bring your harness and choose your model. We'll handle the rest.
#managedagents #agents #AInativecloud #digitalocean #Bringyouragents
Every era of the cloud has an atomic unit. For twenty years it was the provisioned VM. You came for compute and paid by the hour, working or waiting. Agents don't fit that box. They think in tokens, act in bursts mory (state) that outlives the session. Compute, inference, and data working as one.
The compute-first cloud meters the VM. The agent-first cloud meters the work.
I've talked to a lot of founders building agents, and I hear the same stack every time: a harness on a laptop, sandboxes from one startup, inference from two more, storage at a hyperscaler, search from a sixth vendor. Six vendors means fifteen seams to secure and operate. Nobody can say what a single run cost.
So we built Managed Agents deeply integrated with our Inference Engine.
Harness Runtime: every session in its own Firecracker microVM. Pause, resume, checkpoint, fork. Bring Claude Code, Codex CLI, OpenCode, Hermes, or your own image.
Action Gateway: one endpoint to 16,000+ governed tools. The agent never sees the key.
Pricing that matches the workload: active CPU billed by the second. An hour of agent work that bills $0.126 fully allocated costs about $0.031 with us for a typical agent run (25% active). While the agent waits, you pay zero.
Agents think, act, remember, and improve. Now the whole loop runs on one platform.
Public preview is live today. Bring your harness and choose your model. We'll handle the rest.
#managedagents #agents #AInativecloud #digitalocean #Bringyouragents
10
10
71
6,330
Super! Happy to help. We are just getting started. Can't wait to package Omarchy with our compute service and Inference Engine.
Our new @digitalocean servers are already getting put to the torture test building our new bespoke kernel! This one built in just 11 minutes maxing out all cores on the AMD EPYC 9555P!! 🔥
0
0
7
696
With all the tweets over the weekend, the big labs losing their moat is only half the story.
The moat doesn't disappear — it moves. If a fine-tuned open model trained on your own data is 95% cheaper, better at your task, and ready in hours to days, the model is no longer the product. Your data — and the platform that turns it into working intelligence — is the product.
Which leads to the uncomfortable version: if you don't own your weights, you don't own your product.
Build on a closed API and your COGS is someone else's pricing decision. Your roadmap is someone else's deprecation schedule. Your "proprietary AI" can be repriced, rate-limited, or retired in an announcement you don't control.
Renting intelligence isn't a strategy. Owning it is.
Kimi K3, DeepSeek, GLM, Nemotron, and Qwen erased the frontier premium for most production workloads.
Owning your intelligence takes four things, and they have to live on one platform:
1/ Own your cost structure: NVIDIA + AMD GPUs on demand with per-second billing, Spot capacity when you want the price floor — no committed-spend contracts (for most customers), no sales call. Your inference COGS becomes a dial you turn, not a bill you receive.
2/ Fine-tune + deploy: open models from Hugging Face or DeepSeek in 1 click, or bring your own weights
3/ Inference Engine: a world-class token engine across open models - Serverless, dedicated, and batch on one Inference Engine, with routing across open and closed models so you're never locked to anyone's model bet
4/ Core cloud: your model runs next to your Kubernetes, managed databases, storage, and networking — one platform, one bill, predictable pricing.
A model without an app is a demo. The app lives here too. That's DigitalOcean's Inference Engine + Agents platform. And the token math says it's working and durable.
Our curve is steeper — token volume on the Inference Engine up 30x in a matter of few months, inference on open-weight models growing rapidly.
Hyperscalers are conflicted. Neoclouds are landlords. DigitalOcean has spent years making infrastructure simple for the millions of companies that will never build an infra team. Now that's the whole game.
Own your weights. Own your product.
The moat doesn't disappear — it moves. If a fine-tuned open model trained on your own data is 95% cheaper, better at your task, and ready in hours to days, the model is no longer the product. Your data — and the platform that turns it into working intelligence — is the product.
Which leads to the uncomfortable version: if you don't own your weights, you don't own your product.
Build on a closed API and your COGS is someone else's pricing decision. Your roadmap is someone else's deprecation schedule. Your "proprietary AI" can be repriced, rate-limited, or retired in an announcement you don't control.
Renting intelligence isn't a strategy. Owning it is.
Kimi K3, DeepSeek, GLM, Nemotron, and Qwen erased the frontier premium for most production workloads.
Owning your intelligence takes four things, and they have to live on one platform:
1/ Own your cost structure: NVIDIA + AMD GPUs on demand with per-second billing, Spot capacity when you want the price floor — no committed-spend contracts (for most customers), no sales call. Your inference COGS becomes a dial you turn, not a bill you receive.
2/ Fine-tune + deploy: open models from Hugging Face or DeepSeek in 1 click, or bring your own weights
3/ Inference Engine: a world-class token engine across open models - Serverless, dedicated, and batch on one Inference Engine, with routing across open and closed models so you're never locked to anyone's model bet
4/ Core cloud: your model runs next to your Kubernetes, managed databases, storage, and networking — one platform, one bill, predictable pricing.
A model without an app is a demo. The app lives here too. That's DigitalOcean's Inference Engine + Agents platform. And the token math says it's working and durable.
Our curve is steeper — token volume on the Inference Engine up 30x in a matter of few months, inference on open-weight models growing rapidly.
Hyperscalers are conflicted. Neoclouds are landlords. DigitalOcean has spent years making infrastructure simple for the millions of companies that will never build an infra team. Now that's the whole game.
Own your weights. Own your product.
1
4
21
29,995
It was awesome working with you to get this done. A ton of possibilities of what next. Excited.
AWESOME: @digitalocean is joining the Omacom Foundation as a Founding Corporate Patron with a $3 million commitment to the development, maintenance, and spread of Omarchy.
This started as a DM on X to @paddix just 1 week ago! Amazing speed of execution.
omarchy.org/news/2026/09/d…
0
0
5
172
Vinay Kumar retweeted
GLM 5.3 via @digitalocean is 50% off on AI Gateway through Sept 8:
𝚖𝚘𝚍𝚎𝚕: '𝚣𝚊𝚒/𝚐𝚕𝚖-𝟻.𝟹-𝚙𝚛𝚘𝚖𝚘-𝟻𝟶'
vercel.com/changelog/glm-…
𝚖𝚘𝚍𝚎𝚕: '𝚣𝚊𝚒/𝚐𝚕𝚖-𝟻.𝟹-𝚙𝚛𝚘𝚖𝚘-𝟻𝟶'
vercel.com/changelog/glm-…
5
2
44
14,791
Yesterday we launched Managed Agents on DigitalOcean in a private preview. I want to put it in the context of a conviction I've held for a while.
The first generation of clouds was built compute-first. AWS started with EC2, and everything else — storage, networking, databases, even AI services — layered on top. That architecture won the last twenty years.
We fundamentally believe the new AI-native cloud starts token-first. Everything else, including "compute" itself, is a means to enhance the intelligence generated by tokens. Agents are the new interface to the cloud — one that brings tokens and microVMs together under a single abstraction.
DigitalOcean's Managed Agents Runtime Services (M.A.R.S.) is the first full expression of that construct, and it stands on four fundamental services we've been building deliberately:
- Serverless inference — tokens on tap. You consume tokens the way you consume bandwidth. No model hosting, no GPU capacity planning, no idle spend. The meter runs only when your agent thinks.
- Firecracker microVM compute — every agent session runs in its own hardware-isolated microVM, starting in under a second and scaling to zero. It's the right compute unit for software that works in bursts rather than runs forever.
- Durable sessions — agents that outlive your laptop. Pause and resume in as little as 200 milliseconds with files, processes, and working state intact. Continue from another device, hand off mid-task to a teammate, or fan out to parallel agents without multiplying infrastructure.
- Action Gateway — governed access to the outside world. Agents reach GitHub, Jira, Notion, Linear, and Postgres through centrally managed authentication, permissions, human approvals, and audit controls. OAuth credentials are brokered through Secrets Manager and never exposed to the model or the execution environment.
You describe the agent — its harness, tools, and guardrails. Bring Claude Code, Codex CLI, OpenCode, or agents built on LangGraph and CrewAI. No proprietary framework, no lock-in. We run the tokens and the compute underneath.
If you're wondering what to build first, here are ideas you can ship in a weekend:
- A support agent grounded in your docs and past tickets that escalates only what it can't resolve
- A code-review agent that runs on every pull request with full repo context, safely isolated in its own microVM
- A data agent that turns plain-English questions into queries against your managed database
- An ops agent that watches alerts, triages incidents, and drafts the runbook update before you're paged
- A back-office agent that reconciles invoices, chases renewals, and processes intake forms overnight
None of these require you to think about GPUs, orchestration, or scaling. That's the point. Simplicity has always been DigitalOcean's edge — we're now applying it to the agentic era.
This is where our cloud investment is going: a next-generation cloud construct built for AI natives, token-first from the ground up.
We're onboarding private preview cohorts now — request access here: try.digitalocean.com/managed-agents…. And tell me what you'd build.
The first generation of clouds was built compute-first. AWS started with EC2, and everything else — storage, networking, databases, even AI services — layered on top. That architecture won the last twenty years.
We fundamentally believe the new AI-native cloud starts token-first. Everything else, including "compute" itself, is a means to enhance the intelligence generated by tokens. Agents are the new interface to the cloud — one that brings tokens and microVMs together under a single abstraction.
DigitalOcean's Managed Agents Runtime Services (M.A.R.S.) is the first full expression of that construct, and it stands on four fundamental services we've been building deliberately:
- Serverless inference — tokens on tap. You consume tokens the way you consume bandwidth. No model hosting, no GPU capacity planning, no idle spend. The meter runs only when your agent thinks.
- Firecracker microVM compute — every agent session runs in its own hardware-isolated microVM, starting in under a second and scaling to zero. It's the right compute unit for software that works in bursts rather than runs forever.
- Durable sessions — agents that outlive your laptop. Pause and resume in as little as 200 milliseconds with files, processes, and working state intact. Continue from another device, hand off mid-task to a teammate, or fan out to parallel agents without multiplying infrastructure.
- Action Gateway — governed access to the outside world. Agents reach GitHub, Jira, Notion, Linear, and Postgres through centrally managed authentication, permissions, human approvals, and audit controls. OAuth credentials are brokered through Secrets Manager and never exposed to the model or the execution environment.
You describe the agent — its harness, tools, and guardrails. Bring Claude Code, Codex CLI, OpenCode, or agents built on LangGraph and CrewAI. No proprietary framework, no lock-in. We run the tokens and the compute underneath.
If you're wondering what to build first, here are ideas you can ship in a weekend:
- A support agent grounded in your docs and past tickets that escalates only what it can't resolve
- A code-review agent that runs on every pull request with full repo context, safely isolated in its own microVM
- A data agent that turns plain-English questions into queries against your managed database
- An ops agent that watches alerts, triages incidents, and drafts the runbook update before you're paged
- A back-office agent that reconciles invoices, chases renewals, and processes intake forms overnight
None of these require you to think about GPUs, orchestration, or scaling. That's the point. Simplicity has always been DigitalOcean's edge — we're now applying it to the agentic era.
This is where our cloud investment is going: a next-generation cloud construct built for AI natives, token-first from the ground up.
We're onboarding private preview cohorts now — request access here: try.digitalocean.com/managed-agents…. And tell me what you'd build.
0
1
16
2,420
Vinay Kumar retweeted
Congratulations @alexmashrabov for building a rocket ship 🚀
Congrats to @higgsfield_ai on a $400M Series B! 🎉
Honored to support @alexmashrabov and the team on their journey to a $700M run rate, just 16 months after launch.
One of AI's great growth stories, and you're just getting started. 🚀
0
3
21
1,609