Loading…
Systems / Custom AI Agent
NEW ON THE PLATFORMtime · money
Custom AI Agent
Real work, behind approval gates.
- What it is
- An agent that does real work — qualifying, routing, drafting — with tool use, guardrails and human sign-off.
- Typically gives back
- ~6–12 h/week of the work you can already describe step by step
- Replaces
- Copy-paste evenings and “I’ll do it after dinner.”
What it costs you without this
Calm arithmetic, not fear. Methodology.
Work you can describe step by step is work you are still doing by hand.
Illustrative example
Every manual handoff is a place where a busy week becomes a lost lead.
Illustrative example
Agency-grade agent builds are quoted at $40–90k (FDE market research, 2026) — most SMBs never get one.
Illustrative example
| Week | Month | At €40/h |
|---|---|---|
| 10 h | 40 h | €1,600/mo |
10 h/week × €40/h × 4 weeks
Illustrative composite example — 10 h/week of manual repeated work ≈ €1,600/month at €40/h. FDE agencies quote $40–90k for agent builds (market research, 2026); our SMB version lands at €4–9k. Your scan gives the real numbers.
A firm like this
Van Leeuwen Administraties
8-person accounting office drowning in intake emails and document chase
Without the system
- New client emails are read, sorted and answered by hand.
- Missing documents are chased in threads.
- Nothing moves after 6 pm or on Friday.
With the system
- The agent qualifies intake and routes it to the right colleague.
- Document chase runs as drafts — a human approves every send.
- A log shows what the agent did and why.
| Metric | Before | After |
|---|---|---|
| Intake handling | Same-day, by hand | Minutes, routed |
| Hours/week on chase | ~10 h | Exceptions only |
| €/month at €40/h | €1,600 | Scan gives your number |
Illustrative composite example — not a real client
What is in the system
Only what we can point to. Missing pieces stay off this page.
Approval gates
Nothing customer-facing sends, publishes or deploys without your click.
You see: The agent proposes. You decide.
Six hard boundaries
Scope limits in writing — what the agent may touch, and what it never will.
You see: Guardrails you can read, not vibes.
Tool use with a trail
Every action the agent takes is logged with reason and result.
You see: An audit trail, not a black box.
Rollback path
Each release ships with a snapshot and a way back.
You see: Courage to automate, because undo exists.
Your IP from day one
The agent lives in your repository with docs and handover.
You see: Replaceable by design — including me.
How it works
- 1. Pick one workflow you can describe end to end.
- 2. Define the six hard boundaries and the approval gates.
- 3. Build the agent with tool use, logs and rollback.
- 4. Run it in shadow, then live — on your click.
In the build
- Agent scoped to one real workflow
- Six hard boundaries in writing
- Approval gates on everything customer-facing
- Full action log with reasons
- Snapshot and rollback on every release
- Your repository and IP from day one
Good to know
No. Nothing customer-facing goes live without your approval. Internal drafts and research yes; external actions never without a gate.
Not a chatbot demo, not “full autonomy”, not a black box. It is one workflow, bounded, logged, rollback-able.
You. Code, prompts, docs and IP live in your repository from day one.
Works well with
Start with the scan
90 minutes. Written report is yours. Then a fixed price — or a clear no.