AI agents · automation
Freelance AI engineer · Forward Deployed EngineerAn AI agent that does the work
An agent that takes over a repetitive task in your tools (tickets, emails, quotes, data entry), under a policy it cannot get around. First in shadow mode, then with human approval, measured before and after; the code, tests and runbook stay in your repository.
What you get
The outcome
- System running on your infrastructure
- Acceptance report in numbers
- Code, tests, runbook in your repository
- Timing
- 3 to 5 weeks
- You provide
- A business owner and read access to the data
- Price
- Quoted after the brief
What it is for
Three uses
The tasks that come up most often when a team gets in touch.
- 01
Customer support
The agent reads the ticket, checks the order and proposes the action; refunds go through a written rule and human approval.
- 02
Back office
Data entry, reconciliation, quote preparation: the tasks that take hours every week, done in your existing tools.
- 03
Incoming email
Sorting, drafted replies, routing to the right person, with a log of every decision.
How I work
Four steps
One contact, who does the work himself. You sign off before each next step.
- 01
Pick the task
One repetitive task, a business owner, read access to the data. We measure the starting point.
- 02
Shadow mode
The agent runs alongside your teams without executing anything: its decisions are compared with theirs.
- 03
With approval
It acts, a person approves; the policy sets what it may do alone and what stays human.
- 04
In service
Acceptance report in numbers, code, tests and runbook in your repository; the system is handed to your teams.
Examples
See the work
Architecture, tests and measurements detailed on each page.
AI engineeringRELAISA support agent that acts in your systems, under a policy it cannot get aroundRun as a client mission, fictional client.
AI engineeringATELIER OpsFrom brief to production file, sourced, verified and approvedRun as a client mission, fictional client.
AI engineeringStudio FlowProduction orchestrator wired to ComfyUI, from brief to exportStudio project, detailed on its page.
FAQ
The questions
Not the answer? Write to me, I reply myself.
What is an AI agent?
A program that relies on a language model to decide which actions to take (read a ticket, query a database, draft a reply) and carries them out in your tools. Unlike a simple chatbot, it acts: hence the rules, human approval and the log.
How long for a first agent?
Three to five weeks for a pilot in service, possibly after a fixed-price assessment to choose the right task.
What if the agent gets it wrong?
It is measured on an evaluation set before going live, it cannot step outside its policy and every decision is logged. Sensitive actions require human approval.
Does our data stay with us?
Models can run on your servers, with the host of your choice or through an API; no data goes to an external service without your written consent.