How it works
Dashboard, command centre, agent operations.
We build in three steps. Each one adds something to the one before it, and in each one a person decides. Before the first step, a few things have to be in place. After the last, you start on Monday.
Step 1 · Dashboard
One live view across your systems.
What it shows. Information from two or more systems in one live view that belongs to your company. Open orders from the ERP next to open complaints from the support mailbox. Project hours in one tool next to the budget in another. The screen uses your own words. Customer means one thing on it, whatever each system behind it calls a customer.
What it adds. It closes the gap between systems that each know part of the truth. Nothing has to be replaced for that. The ERP stays, the CRM stays, and the view sits on top of them.
What it does. It shows. It does not act.
Who decides. The person looking at the view. A dashboard gives the whole picture, and the person chooses what to do about it.
Step 2 · Command centre
It also acts, and it also judges.
What it shows. The same view, with a proposed next action beside every row and a one-line reason for it. In a supplier invoice process, each row shows the supplier, the amount, the matched purchase order, the approver and the days the invoice has waited. The proposal is one of four: approve and pass on, ask the supplier, ask the buyer, hold.
What it adds. Two things. Actions: the tool can change something in the process. Judgement: the tool uses AI to assess something, such as whether an invoice looks right, how urgent a ticket is or what the next sensible step is for a contract.
What it does. It sends the reminder, sets the status to on hold, creates the task and writes the draft reply. Code does everything that has to be exact: matching an invoice to an order number, comparing amounts, counting days. The model is never allowed to do the sums. It reads the PDF into fields and judges a mismatch, such as an invoice for forty units against a delivery note for thirty-six.
Who decides. The person. They get a proposal with a reason, and they approve it or change it. The tool does the reading, the matching and the preparing.
Step 3 · Agent operations
An agent runs the process between your checks.
What it shows. A process that keeps moving on its own, and a view of what it did and what it left for a person.
What it adds. Continuous running. An agent is not smarter than a prompt. It is a prompt in a loop, with tools and rules: something happens, it checks a condition, takes an action, verifies the result and waits for the next event.
What it does. It runs one or more business processes and saves people's time.
Who decides. We move a process to an agent only once the command centre has run long enough that people trust the proposals and the exceptions are rare. Then the person goes from approving every action to watching the process and stepping in when something looks off.
Most of the value sits in the first two steps, which are cheaper and safer.
Automation levels
Workflow, copilot, agent.
Three ways to automate a step. Each one leaves the person somewhere different.
| Level | What runs | Where the person is |
|---|---|---|
| Workflow | Plain code, steps in a fixed order. Nothing is judged. This is where the exact work goes: matching, comparing amounts, counting days. | Wrote the rules in the specification and owns the process. |
| Copilot | The command centre. Code does the exact steps, a model reads and judges, and the tool proposes the next action with a reason. | Approves or changes every proposal. |
| Agent | A prompt in a loop, with tools and rules. It acts on events, checks the result and waits for the next one. | Watches the process and steps in when something looks off. |
What has to happen first
Three prerequisites, in this order.
The order matters. Each one is a part of the platform.
People learn to order code
Basic code generation and the process around it: how to describe what you want, how to read what comes back and how to ask for a change. It is a skill of the same size as learning Excel once was. Without it, someone outside the business stays the bottleneck. This is FastTrack.
Guardrails
Generated code has to be tested and has to meet the requirements you set: automated checks on every change, review by a person who can read the result, and a clear rule for what may reach the people who use it. The glossary and the process description are what the checks are written against. This is GuardRails.
A safe zone
A place where data is stored safely and access is controlled. When people start building their own tools, the data they point those tools at needs rules about who may see what. Without it, the first useful dashboard is also the first leak. This is SafeZone.
Start here
What to do on Monday.
Pick one process
One spreadsheet, or one process that someone repeats every week. Choose one that a person updates by hand from two other places.
ChooseWrite it in ten lines
Where the information comes from, which words it uses, what the person does with it and what happens next. If the five words (customer, order, case, product, project) are not agreed yet, write down the disagreement. Then add one sentence per step: code or judgement.
DescribeName the owner
One person who knows the process and is allowed to change it. If you cannot name one, that is the first finding.
OwnDashboard or command centre
If people only need to see the situation, build a dashboard. If someone takes the same kind of action every time they look, build a command centre with a proposed action and a reason.
DecideTen lines are a complete specification. Bring them to a call, and we build the first version with the owner.