what are enterprise ai agents?

the six things that make an agent enterprise

None of them are about the model.

1. Identity. The agent acts as an identity your directory knows, so access follows the same rules as a person's. Sso, scim, group membership.
2. Scoped permissions. Read-only on one repository is a different risk from write access to a workspace, and someone needs to be able to set that without a support ticket.
3. Audit. Every action recorded, attributable, and exportable. Not a chat transcript, an action log.
4. Isolation. One team's data does not reach another's context. This is where most homegrown deployments quietly fail.
5. Data handling. Where it runs, where it is stored, whether it trains anything, and what a subprocessor list looks like.
6. A contract. Sla, liability, and a security review that a real reviewer will accept.

A vendor that leads with model quality and cannot answer four of these is selling a consumer product with an enterprise price.

why deployments fail

Rarely because the agent was not smart enough. The recurring causes:

• No owner. An agent with no named human responsible for its output drifts until someone shuts it off.
• Permissions too wide, granted once. It gets full access on day one because narrowing it was hard, and then nobody revisits it.
• No review capacity. Output arrives faster than anyone can check, so it is either rubber-stamped or ignored.
• Pilot chosen for demo value. The impressive use case is usually the one with the least measurable outcome.

The successful pattern is boringly consistent: one narrow high-volume process, a defined check on the output, a named owner, and a number you agreed to measure before you started.

what it costs

Enterprise vendors typically quote $25,000 to $75,000 for a pilot and six figures a year at scale, and most of that is integration and support rather than compute. Platform-tier products cover the same requirements at $20 to $250 per user per month plus usage, which is where a lot of teams land once they realise the pilot was mostly a services engagement.

how this works in aldena

Aldena covers the six from the architecture rather than as add-ons. Every project runs in an isolated room with its own server and its own credentials, so tenant isolation is physical rather than a filter. Tool permissions decide exactly what each agent may reach, per room and per agent.

The review capacity problem is handled with a gate: nothing merges, deploys, or ships until a person signs off, because approval is on by default and the run resumes where it paused. Ownership maps to the org chart, so every agent reports to someone.

Pricing is per plan, not per agent: $249 a month for business with 100 rooms, 20 agents per room, and 50 members, with model usage billed separately at published rates.