On 1 October, DigitalOcean took a swing at that problem. It launched Agent Droplets: one monthly plan that covers the compute, memory, storage, hosted inference and tools an agent uses. It is the same idea that made the original $5 Droplet popular with small teams, applied to agents: a bill you can read.
What actually launched
Agent Droplets sit on top of DigitalOcean Managed Agents, which opened to everyone in public preview on 21–22 September. Managed Agents gives each agent its own isolated microVM, governed access to more than 16,000 tools (code interpreter, browser automation, web search and fetch among them), DigitalOcean-hosted models, and persistent storage between sessions.
The plans are simple:
- Free trial: $5 of credit for new users, no card required.
- Pro: $50 a month, with 15% off all eligible agent usage. DigitalOcean says the fee covers $58.82 of usage at list price.
- Team: $200 a month, with 20% off. The fee covers $250 of list-price usage.
There are no seat charges and no limit on the number of agents or environments, and single sign-on, MFA, role-based access and audit logs are included on every plan. When the allowance runs out you choose: stop there, or carry on at list prices from a separate prepaid Inference and Agents Balance ($5 to $500 at a time). If that balance hits zero, running sessions pause at a safe point with their state saved, rather than failing halfway through a job.
You can also bring the harness you already use. DigitalOcean lists Claude Code, Codex CLI, OpenCode and Hermes, plus agents built with CrewAI or LangGraph.
Why the pricing model matters more than the price
Agents are bursty. They compile, test or fetch for a few seconds, then sit waiting on a model, a tool or a human. Traditional cloud billing charges for the whole box from boot to shutdown. Managed Agents bills CPU per second of actual use ($0.044 per vCPU-hour) and pauses idle sessions automatically, resuming in about 300 milliseconds.
DigitalOcean's own worked example is useful: a 2 vCPU, 4 GB session costs $0.126 for a fully used hour, but about $0.060 at a typical 25% activity level. Its point is that infrastructure under an agent run rounds to pennies. Tokens are what drive the spend. That is the right mental model for any UK team budgeting for agents, whichever cloud you use.
Read the small print before you build on it
This is a promising launch, and DigitalOcean is a partner we like. It is also a week old, so a few honest checks:
- It is a public preview. DigitalOcean says it expects, but does not guarantee, that preview services will perform for production workloads. Fine for pilots and internal tools; think twice before putting a customer-facing process on it.
- Check where the work runs. At the time of writing, DigitalOcean's availability page lists Harness Runtime in one region, RIC1, and not London (LON1). If your agent will touch personal data, factor that into your UK GDPR assessment and international transfer paperwork, or keep personal data out of the pilot.
- Frontier models are not in the discount. Anthropic Claude and OpenAI GPT models are charged at standard rates. The discount applies to DigitalOcean-hosted models such as Kimi K3 and GLM 5.3, plus compute, storage and tools.
- It is priced in dollars. Plan fees are billed in advance, exclude taxes and renew monthly, so budget for exchange rates and VAT.
- If you built on DigitalOcean's older Agent Development Kit, note the change. The ADK was deprecated on 1 October. Existing agents keep running, but you can't deploy new ones with it, and DigitalOcean says migration guidance will follow.
On the plus side, DigitalOcean also put its Insights observability service (dashboards, metrics, logs, traces and alerts) into public preview the same day. Watching what your agents do is half of trusting them.
A sensible first agent: audit your website content
The best first agent is boring, bounded and read-only. A content audit of your own website is ideal, and this is where Statamic shines.
Statamic keeps content in clearly defined blueprints, so an AI assistant can see exactly which fields exist and what each one is for, instead of guessing at a page builder. Add-ons such as Cbox's MCP Server for Statamic (free, MIT-licensed, Statamic 6) give assistants like Claude, ChatGPT and Cursor structured access to entries, blueprints and assets. Access runs through scoped tokens with 21 separate permissions, OAuth 2.1, rate limiting and an audit log you can read in the control panel. It also ships content-workflow actions such as content_audit and cross_reference.
That combination lets you hand an agent a narrow job: find entries with missing meta descriptions, broken internal links or thin pages, and report back. Give it a read-only token, keep publishing with a human, and you have a useful agent on a predictable bill without giving it the keys to your live site.
Two habits to adopt from day one:
- Keep your add-ons and core patched. Cbox's 3.1.0 release in September fixed a bug where one blueprint action could write files with only a read permission. Statamic itself shipped a security release, 6.35.1, on 6 October. Agents raise the stakes on keeping up to date.
- Scope every token to the job. Read-only for audits, write access only for named collections, and never a shared admin token.
A two-week plan for UK teams
- Pick one repetitive, low-risk task, such as a content audit, a weekly report or triaging form submissions.
- Run it on the free $5 credit and log what each run costs and how long it takes.
- Decide where the data can live before you include anything personal.
- If the numbers stack up, move to Pro at $50 a month and set a hard stop, so the bill can't surprise you.
- Turn on Insights and review what the agent did every week.
Why this matters for Clarity Growth
Clarity Growth is a DigitalOcean and Statamic partner, and this is the kind of setup we help UK teams get right: choosing which job an agent should do first, connecting it safely to the website and systems it needs, scoping access so it can't do damage, and keeping costs readable as it scales. Predictable agent pricing makes the business case easier. Good structure, tight permissions and sensible data decisions make it safe to say yes.
Sources
- Introducing Agent Droplets: everything an agent needs, one price, one bill — DigitalOcean (1 October 2026)
- DigitalOcean Introduces Agent Droplets — DigitalOcean Investor Relations / Business Wire (1 October 2026)
- Introducing DigitalOcean Managed Agents — DigitalOcean (22 September 2026)
- The agent-first cloud: why we built Managed Agents — DigitalOcean (25 September 2026)
- Harness Runtime pricing — DigitalOcean Documentation
- Harness Runtime availability — DigitalOcean Documentation
- DigitalOcean release notes — DigitalOcean Documentation
- MCP Server for Statamic — Statamic Marketplace
- MCP Server for Statamic release notes — Statamic Marketplace
- Statamic release notes — Statamic