AI agent deployment
One agent per workflow, integrated with your tools, with a person keeping the final say. Built in four to twelve weeks, with a written contract on what the agent may do.
Definition
What we mean by "AI agent"
An agent runs a workflow end to end: it reads incoming email, classifies invoices, updates the CRM, chases late payers, prepares filings, and alerts you when something drifts.
Nobody on your team learns a new interface. The agent works inside your tools, and a person keeps the final say on what matters.
Three conditions to deploy
- The workflow is described in a form a machine can follow.
- Your data is reliable enough to be processed without systematic manual checking.
- There is a named person who approves the critical actions.
If one of these conditions is missing, we start with the diagnostic.
Autonomy
Who decides, at each level
An agent's autonomy is set per workflow, on a five-level ladder. Every deployment starts as a consultant. Moving up is conditional: thirty days above 95% accuracy, and only on low-risk actions.
| Level | Who decides |
|---|---|
| Operator | The person does the work. The agent is a tool they invoke. |
| Collaborator | The agent produces pieces. The person assembles and decides. |
| Consultant | The agent proposes a complete action. The person approves every time. This is the starting point of every deployment. |
| Approver | The agent acts on low-risk cases. The person handles the exceptions. |
| Observer | The agent acts alone, the person audits afterwards. Never for an irreversible action. |
Autonomy ladder: Feng, McDonald, Zhang, Knight First Amendment Institute, 2025.
Classify a workflow
Which category is your workflow in?
Three questions. It is the test we apply to every workflow kept in a diagnostic, and it drives the timeline as much as the price.
Architecture
One agent per workflow
We build one agent per workflow. On tasks nothing verifies automatically, every added agent is one more place where an error slips in silently. An analysis published by AlphaSignal in May 2026 measured a 17x error amplification between multi-agent and single-agent systems on the same task.
Multi-agent systems make sense when every result can be verified automatically, like a mathematical proof or a passing test. Your invoices and client emails do not verify like that.
Measurement
What we measure, and when
For the first two or three months, a deployment that works and one that fails look the same from the outside: extra costs, no visible effect on results. So we set leading indicators from the start, before spending. If they have not moved by week eight, we stop and rework.
The time saved has to go somewhere too. We write down with you what the freed hours are reassigned to. Left alone, they dissolve into email and the return stays at zero.
Investment
Investment
Quoted, after the diagnostic. The price depends on the workflow's category (fixed flow or reasoning agent), the number of integrations, and how critical it is.
If a Havoop diagnostic precedes the deployment, 50% of its price is deducted from the first month of the engagement.
Questions