AI-leveraged, not AI dependent.
Custom AI agents built around how your team works
I build custom AI agents for the work your team still does by hand, from recurring reports to document checks. I review the process first, connect the systems it needs and agree which actions require human approval.
From 2,000 USDexcl. VAT
- A fixed price quoted after the first call.
- Agents run in the cloud - you don't need a laptop to run them.
- You decide what it sends alone and what waits for approval.
Examples
Give each agent a clear role
Start with one job your team wants off its desk. These example agents pair that job with tools and instructions for how your team works.
Mara
Marketing agent
Prepares a content brief from your search and website data.
Leon
Sales agent
Researches an account and drafts a follow-up for approval.
Rory
Reporting agent
Posts a weekly client report with changes worth reviewing.
Nora
Delivery agent
Turns meeting notes into tasks and drafts project updates.
Felix
Finance agent
Checks invoices against orders and drafts payment reminders for approval.
Start with the process your team actually runs
I walk through real work with the person who does it today, including the exceptions. I then recommend a first workflow based on the manual effort, the available data and the cost of getting it wrong.
Follow a real case
I trace a report, document or request from arrival to completion, including every handoff and manual check.
Choose a useful first task
I recommend one task with a clear owner and a result your team can check before expanding the work.
Check the existing systems
I confirm which APIs, connectors and permissions are available before promising an integration.
Define the controls
I agree what the agent may read or change, what needs approval and who handles exceptions.
The work runs while your laptop is off
I configure scheduled tasks in the cloud workspace, such as a weekly report or an overdue-invoice check. Each task has an owner, agreed access and a destination for the result, with exceptions sent for review.
How the setup runs
A setup starts from 2,000 USD excl. VAT. After the first call I quote a fixed price for the POC, MVP and 8 hours of iterative improvements.
Review the real process
You show me real examples, the systems involved and the person doing the work today. I map the steps, exceptions and approval points, then identify what needs AI and what can use straightforward automation.
Prove the task with a POC
I scope the proof of concept (POC), MVP and initial improvement hours before starting. I test the POC on your examples to establish whether the data, integrations and output are good enough to continue.
Build an MVP for daily use
I turn the proven task into a minimum viable product (MVP) in the cloud workspace. I connect the agreed tools, set access and approval rules, and add scheduling and error handling.
Use the first 8 hours to improve it
Your team uses the MVP on a limited set of real work. I use the included 8 hours to review results, refine the skills and address failures before recommending a wider rollout.
Hand over the working setup
You get documentation, an owner for each workflow and instructions for reviewing, changing or stopping the agents.
Continue with a retainer or more hours
After the initial improvement block, I can keep developing the setup through a retainer or an agreed block of hours. I agree the priorities, scope and fee with you before the next work starts.
What clients say about working with me
These recommendations cover my engineering work and how I communicate during a project.
01 / 03
Before you set up an agent
Which AI do the agents run on?
What is an MCP gateway for?
It gives supported AI apps a shared way to reach the tools and skills your team has configured. For example, Duckpin gateways connect apps such as Claude, ChatGPT and Cursor, with access set for the person signing in. Duckpin's cloud agents have their own assigned connectors and skills, so they do not need to go through that gateway.
Do I need software such as Duckpin or Dust?
A managed workspace runs the cloud agents and holds the connections, skills, schedules and access settings. I choose the platform with you based on the systems involved and the controls you need. Platform subscriptions and model usage are ongoing costs, quoted separately from my setup fee.
Whose account does an agent use?
I agree that per connection - a team account for shared records, or the requesting person's account where the platform supports it. A shared connection exposes what that account can reach, so I keep personal inboxes separate and limit who can use each agent or gateway.
What jobs can an agent take over?
Work where someone moves data between tools or turns it into something the team reads, such as a summary, a draft or a ticket. Document checks, order matching and payment reminder drafts are other workflows I can assess. Where the work needs judgement, the agent prepares it and a person decides.
Can you work with invoices and transport documents?
I can assess a workflow that reads documents, extracts fields, proposes matches to orders and flags discrepancies. I start with representative documents and confirm access to your order and accounting systems before quoting the integration. Posting records, changing accounting data and sending customer reminders need explicit permissions and agreed review rules.
Can you handle thousands of documents each month?
I confirm that through a pilot using your document types and peak workloads. I check extraction quality, exception rates, system limits and running costs before agreeing the production scope.
What if our software has no ready-made connector?
I check whether it exposes an API, supports an export or allows another practical integration. I confirm feasibility and cost before promising that the agent can work in that system. Existing systems help define the solution.
How do I keep control of the agent?
I agree which data it can access, which records it may change and which actions need approval. The setup includes a record of its actions, exception handling and instructions for stopping it. Missing data, conflicting documents and failed requests go to a person for review.
How much does an agent setup cost?
A setup starts from 2,000 USD excl. VAT. After the first call I quote a fixed price for the POC, MVP and 8 hours of iterative improvements, and I start once you approve the quote. Further improvements use a retainer or an agreed block of hours. A workflow spanning several business systems needs its own scope and estimate.
Can an agent send messages to my clients?
Yes, if you want it to. You decide for each agent what it sends on its own and what waits for someone to approve. A routine status update can go straight out while a reply to a client waits as a draft.
Why one agent per job instead of one assistant for everything?
An agent with one job and a few tools is easier to trigger, test and trust. An assistant with hundreds of skills ends up with nobody knowing which skill runs when.
Will the skills go out of date?
Yes, when your business changes. Skills store only what changes slowly, such as where things are, who owns what, how your team works and decisions with a link to where they were made. Status, dates and open work come from the connected tools each time the agent runs. When a process or an owner changes, the skill needs an update. You own the setup, so your team can edit the skills itself, or I can update them for you.
Where does the agent's memory live?
In the agreed workspace or connected storage, where the right people can read and correct it. I document what the agent keeps, who can access it and how to remove or update it. Reusable instructions can also be kept in a versioned repository.
Does my laptop need to stay on?
No. The agents run in the cloud, so a scheduled task runs while your laptop is closed and nobody needs a powerful machine to use them.
Do you use agents in your own work?
Yes. I run project agents on client work for reports, onboarding and status updates, and I edit this site through Claude Code.
Can you show an existing deployment like the one I need?
I can discuss the reporting, onboarding and status-update agents I use in project work, within client confidentiality limits. For a document or accounting workflow, I confirm whether I have a relevant example to share. If I do not, I propose a small pilot with agreed checks so you can assess the results before commissioning a larger rollout.
Is this the same as the chatbot on your site?
No. The website chatbot answers your visitors from content you approve. An agent from this service works inside your team's own tools.
Something else on your mind? The chatbot answers questions about my work from this site. Ask in English, Polish or your own language.
Ask the agentPlan your first agent
Describe the problem
Send a short brief of the process your team runs by hand today and the tools that hold its data.
Agree the plan
After a call I propose the agent, its connectors and a fixed price.
Run it with your team
I set it up, run it next to the person who does the job today, then hand it over.