AI agent development and process automation
An AI agent is a program that plans and runs several steps to deliver a result, instead of answering a single question. I build agents that take over repetitive work: they collect data, analyze it, and prepare a result for a person to approve.
Who it is for
- Companies where someone copies, cleans, or reports data from several sources every day.
- Product teams that want automated QA or feedback analysis.
- People who built a prototype in ChatGPT or n8n and need a version that runs reliably.
What I can do
- Picking the process where an agent saves the most time and writing down its steps.
- One agent or a multi-agent system with tools (APIs, databases, browser, files).
- Context and memory: the agent sees what it needs and nothing more.
- Evals, traces, and logs so you can see what the agent did and where it fails.
- Human approval before actions that cannot be undone.
How I work
01
Workshop and process map
We write down what a person does today, step by step, and where the data lives.
02
Scope
We agree on what the agent does on its own and what it leaves for a person to approve.
03
Two-week sprints
I build the agent on real data, and after each sprint I show how it handles real cases, not a demo.
04
First working version
Your team can run the agent on its own cases. It ships with tests, scoped permissions, logs, and human approval.
05
Iterations and rollout
The agent runs in your environment and reports where your team already works. Later sprints fix the places where it gets things wrong.
- I do not give fixed timelines, because they depend on the size of the app. After every sprint you see what is done and what is left.
- I bill hourly and invoice as a B2B contractor.
- I design the UX and UI myself, alongside the code.
- I reply to messages within 24 hours.
Projects that back it up
FlowTap
Multi-agent QA: agents walk through flows in a React Native app, take screenshots, and a vision model reviews the UI.
See project: FlowTapBoard of Advisors
A multi-agent system in AutoGen where agents look at a problem from different perspectives and return structured output.
See project: Board of AdvisorsSamantha
A private assistant with long-term memory, notes, and integrations: desktop, Slack, Telegram, Siri.
See project: SamanthaTechnologies
FAQ
How is an AI agent different from a chatbot?
A chatbot answers a question. An agent gets a goal, splits it into steps, uses tools, and returns a finished result. I explain more in my article on AI agents.
Which process should we start with?
One that repeats, has a clear output, and takes someone a lot of time. The agent lab lists 70 such workflows for sales, finance, operations, HR, and IT.
Can an agent make mistakes?
Yes. That is why I build it with tests on real cases, logs of every step, and human approval before important actions.
Is n8n or Make enough?
For simple, predictable automations it often is. A coded agent makes sense when the steps depend on the content of the data, you need tests, or the process is too complex for a scenario.
What does working together on an AI agent look like?
We start with a workshop where we map the process. Then I work in two-week sprints on real data. I do not give a fixed timeline, because it depends on the size of the process. I bill hourly on a B2B invoice.
Read next
Tell me what you want to build
A few sentences are enough: what problem you solve, for whom, and what you already have. I reply within 24 hours with questions or a workshop proposal.