Skydive launched publicly on August 20, 2026, after a closed beta with more than twenty companies. It comes from Anything, an AI app builder with over a million users, and the pitch is compact: describe the outcome you want and you get a working AI coworker in minutes, with a name, a role, and a computer of its own.
Most agent products give you a chat box wired to some APIs. Skydive gives each agent a browser, a file system, and a terminal, so it can log into a tool that has no API and click through it the way a person would.
We spent time in the product and its documentation to work out what it does, what it costs once usage is counted, and where the edges are.
Key takeaways #
Every agent gets its own cloud computer, with a browser, a file system, and a terminal. That is the feature that separates Skydive from tools limited to whatever their integrations expose.Routines run on a schedule, not on events. Skydive’s own documentation describes a Routine as work an agent runs on its own schedule, with examples like every weekday at 9am. Nothing fires the moment an email arrives.The plan price is a floor. Starter is $20 a month and Team is $200, but model and compute usage is billed on top, pay as you go, at what Skydive states is 0% markup on model cost.Skydive is a product of Anything, which raised an $11M Series A led by Footwork and runs with a team of about fifteen people.
What is Skydive? #
Skydive is a platform for building AI coworkers. You describe a role in plain English, and Skydive assembles an agent to fill it. Each agent gets a name, a job, and persistent memory, and you reach the same agent from the web, Slack, email, iMessage, a desktop app, the command line, or the API. The memory follows it everywhere, so a correction you make in Slack shows up when you email it the next morning.
Templates ship for Chief of Staff, Engineer, Customer Support Manager, Project Manager, and Product Hunt launches. The Chief of Staff template states the ambition plainly: it learns how you work, then runs your week from your inbox.
Two former Google product leaders, Dhruv Amin and Marcus Lowe, founded Anything. Footwork, Bessemer, Uncork, M13, and Shopify CEO Tobi Lütke back it.
How do agents with their own computers work? #
Most AI assistants can only touch what their integrations expose. If a vendor hasn’t built a connector, the assistant can’t reach the tool. Skydive gives each agent a cloud computer with a browser, which means the agent can open a site, log in, navigate, and pull data out of a system that never published an API.
In practice that covers the long tail of internal tools, older vendor portals, and anything behind a login that a connector directory has skipped. It also means an agent can download a file, run a script over it, and finish multi-step work without handing control back to you between steps.
The trade is straightforward. A browser agent clicking through a portal is slower and more fragile than an API call, and it needs credentials for the systems it visits. Both are normal for the category, and both require upfront planning.
What do Skydive Routines actually run on? #
Routines are time-triggered, and that decides which problems Skydive solves.
Skydive’s Routines page is titled “Agents That Run on a Schedule.” Their FAQ describes a Routine as work your agent runs on its own schedule, with no human in the loop, and the examples given are every weekday at 9am, the first of each month, and tomorrow at noon. Their own worked example is a weekday inbox summary.
So Routines are time-triggered. An agent can read your inbox, but it reads it when the clock says to. There’s no event trigger on the page, nothing that fires when a specific email lands, and no webhooks.
For plenty of work, a schedule is exactly right. A morning digest, a monthly report, and a weekly pipeline review all want a clock. For work where the arriving message is the thing that matters, a schedule means the request sits until the next run. Skydive can read your channels. Their integrations page offers to let an agent read and send email from Gmail and read and post in Slack. The question isn’t whether the agent can see your inbox. It’s what makes it act.
Skydive pricing: what $20 and $200 actually buy #
Skydive’s pricing is more legible than most of the category, and it still needs reading carefully.
Starter, $20 a month. Up to five users, unlimited agents.Team, $200 a month. Ten times the usage of Starter, plus priority support.Enterprise, custom. Adds SSO, SAML, and audit logs.
Skydive meters usage, not agents. Build as many as you like on any tier, and what you pay for is what they consume, at a stated 0% markup on model cost.
People are a separate axis. The price doesn’t rise with each person you add, but the tier caps how many you can have: Starter covers five, so the sixth moves you to Team at $200 a month. Team’s listing states no user count of its own.
The part to plan for: the plan fee is a floor, not a total. Model and compute usage is billed on top, pay as you go. What you actually spend depends on how much your agents run, which is not knowable from the pricing page. A team running a handful of scheduled digests will look nothing like a team running browser agents through a portal all day.
Skydive’s own pricing page compares an agent against a $190,000 a year human hire and puts a mid-range agent at $6,000 a year. That figure is more useful than the plan fee, because it’s the only place they publish a number with usage included.
Integrations and model choice #
Skydive’s integrations page lists well over a hundred connectors, organized across communication, dev tools, project management, productivity, CRM and sales, finance, support, docs and data, meetings, and analytics. Gmail, Slack, GitHub, Notion, Linear, Google Drive, Figma, and Intercom are all there.
The platform supports more than 170 models, and because model cost is passed through, choosing a cheaper model for a routine job actually shows up in your bill.
Who should use Skydive? #
Skydive fits well if:
- You want capacity you can hand a job to. A named agent with one role is easy to reason about and easy to hold to a standard. - Your work involves tools with no good API, where a browser agent earns its keep. - Your automations are naturally scheduled: digests, reports, recurring reviews. - You want an agent anyone can email. Giving an agent an address on your domain means people outside your company can reach it. - You want more agents rather than more people, since agents are unlimited on every tier while Starter covers five users.
It fits less well if:
- Your problem is work arriving and getting missed. A schedule doesn’t solve that. - You need shared team memory. Skydive’s memory belongs to each agent and its templates are per role, so what your support agent learns stays with your support agent. - You need predictable monthly cost, since usage billing makes the total variable by design. - You’re a small team without anyone to own agent design. Something has to define the roles.
The alternative: work that starts when the message does #
Skydive and this+that overlap on the idea that an AI coworker should finish work rather than draft it. We differ on what starts the work.
this+that reads Gmail, Outlook, Slack, Microsoft Teams, Google Chat, and WhatsApp Business as one stream of incoming work. When a message lands, that’s the trigger. A request buried in a long email thread becomes a task in DoBox, assignable and visible to the whole team, and a workflow can act on it immediately rather than at the next scheduled run. Nobody has to forward anything to an agent, because we’re already reading the mailbox.
The other difference is where knowledge lives. Our brain is one shared layer that every teammate and every workflow reads and writes, and it fills itself from the conversations your team is already having. In Skydive, memory belongs to the individual agent.
If you want a roster of coworkers you assign jobs to, Skydive is a good answer. If your problem is that things arrive and get missed, that’s a different shape of product.
Frequently asked questions #
What is Skydive AI?
Skydive is a platform from Anything for building AI coworkers: cloud agents that each have a name, a role, and their own computer with a browser, file system, and terminal. It launched publicly on August 20, 2026.
How much does Skydive cost?
Starter is $20 a month for up to five users and Team is $200 a month with ten times the usage. Agents are unlimited on both, and the Yearly toggle shows the same prices, so there’s no annual discount. Skydive bills model and compute usage separately, pay as you go, at a stated 0% markup on model cost, so the plan fee is a floor rather than a total.
Can Skydive read my email?
Yes. Skydive’s integrations let an agent read and send email from Gmail and read and post in Slack. Each agent can also have its own email address on your domain, so you can email it, forward it a thread, or cc it.
Do Skydive Routines run when something happens?
No. Skydive describes a Routine as work an agent runs on its own schedule, with examples such as every weekday at 9am or the first of each month. Routines are time-triggered, so an agent reads your inbox on a clock rather than reacting the moment a message arrives.
Who is behind Skydive?
Skydive is built by Anything, founded by former Google product leaders Dhruv Amin and Marcus Lowe. Anything raised an $11M Series A led by Footwork, with Bessemer, Uncork, M13, and Tobi Lütke also backing it.
Is Skydive good for teams?
Agents are unlimited on every tier, though Starter caps users at five. The thing to check first is memory: Skydive’s memory belongs to each agent, so knowledge doesn’t automatically pool across a team the way it does in a product built around a shared brain.