01
Manual work eats hours every week
Employees copy data between spreadsheets, retype information from emails into systems, assemble reports by hand. The same process repeats hundreds of times a year and nobody automates it.
// AI implementations for companies
I build AI systems that take over your team's repetitive processes.
I start by mapping your processes. Then a pilot on one or two. Then a system the whole team uses.
// problem
Why AI implementations fail in companies
02
ChatGPT hallucinates and doesn't know your company
Most companies try generic AI tools and hit the same wall: the model knows nothing about your clients, procedures or documents. An AI agent built around company data answers differently. Across eight AI readiness audits of Polish SMEs this was the most common problem.
03
Automation stalls after a week because nobody tends it
Companies set up a workflow in Make or n8n, it runs for a while, then an integration drops, a token expires, someone changes the data structure. Without documentation and a person who knows how to fix it, the automation gathers dust. An AI system needs guardrails and someone on maintenance.
// solution
How I build AI systems for companies
Process mapping as the starting point
I don't quote a project before studying the processes. Together we map what repeats, where employees lose time and where the easiest entry point for a system is. You leave with a process map and a pilot plan before any number is mentioned.
Human in the loop at every step
The system prepares, the human decides. Tomek, a hotel operator, gets reply drafts at 6:00 every morning. He picks what gets sent. The agent removes the prep work without taking away control. Every deployed system has clearly defined points where the decision belongs to a human.
Interfaces your team uses without knowing AI
I build the system so an employee uses it like any other tool in the company. I match the stack to the team: Claude Code, OpenAI Codex or a custom Python interface. The employee doesn't need to know what's underneath.
Tuning on data from real work
After the rollout the system isn't frozen. I collect logs, analyse results, take feedback from employees. It stays clear what works and what needs fixing. The system gets better from real usage, not from launch-day assumptions.
// pricing
What does building an AI system for a company cost
The path looks the same for most companies: a free process study, a pilot on one or two processes, then a system for the whole team. The client pays separately for the LLM subscription or API key, usually ChatGPT Plus at about $20 a month. Solo operator? Installs from 1,500 PLN live in the labs section.
01
Process study and map
Free to start
The free Audyt AI: a PDF with a 0-100 score, a gap map across 5 areas (digital presence, automation, communication, documents, AI readiness) and three priorities with estimated impact. Plus a diagnostic call. You leave with a process map and a pilot plan. No slide deck, no list of trendy tools.
02
AI system for your team
From 10,000 PLN
The starting point is 10-15k PLN, scaling up with the number of roles covered and the complexity of the work. A pilot on one or two processes is the first stage of the project. Then a system for the rest of the team: interfaces, guardrails, training, documentation. The quote comes only after the free process study.
03
Care and tuning
From 1,000 PLN/month
Optional. Covers monitoring, analytics, system tuning and rolling out further processes. The company can also run it on its own after the documentation handover.
// use cases
Where AI process automation pays off most
01
Inbox and communication
The agent reads emails, ranks urgency, drafts replies and sends them for human approval. Tomek Prusak, a hotel operator, gets prepared replies from three inboxes before his day starts. A voice memo recorded in the car comes back as a ready file. A human decides what gets sent.
02
Management reporting
The agent collects data from sales systems, CRM, spreadsheets and the mailbox overnight. In the morning the decision-maker has one report in one place. I shape the format around the questions the owner asks every morning: what is open, what is stuck, where the problem is.
03
Back office and documents
Invoices, contracts, briefs: the agent reads the file, extracts data into structure, syncs it with a spreadsheet or a system. Sometimes automation means leaving an expensive tool. Norbert Buszek, a personal trainer, paid a platform around 500 PLN a month. Today he runs his own system himself and grows a training app for 50 clients under his own brand, with no middleman.
// process
What an AI implementation looks like, step by step
Krok 01
Process study and map (free)
I start with a map. The free Audyt AI shows in a PDF where the company loses the most (a 0-100 score and three priorities), and in a diagnostic call we get down to specific processes: what repeats, where employees lose time, where the easiest entry point is. You leave with a map and a pilot plan.
Krok 02
Pilot on one or two processes
The system goes to production with a human in the loop. Employees work with it live and flag what doesn't fit. For the first weeks I check what works and what needs tuning. The client sees the effect before deciding to roll it out to the rest of the team.
Krok 03
A system for the rest of the team
After the pilot I build interfaces for the remaining employees. Claude Code, OpenAI Codex or a custom Python interface, depending on how the company works. Every rollout ends with training and operational documentation that stays with the client.
Krok 04
Tuning on data
I collect logs and feedback from the team. I improve the system based on real usage. The retainer is optional: monitoring, analytics and rolling out further processes. If the company prefers to run it alone, I hand over everything needed.
// faq
Questions about implementing AI in a company
Where do I start with AI in my company?
Start with the free Audyt AI, which points to three priorities ready for automation in your specific company. A generic answer is not much use, because everything depends on whether your biggest pain is communication, documents, reporting or repetitive back office. The audit maps five areas and gives you specifics instead of a generic "start with a chatbot". You order it via the contact page; the result arrives within 24-48 hours.
How much does implementing AI in a company cost?
Building an AI system for a company starts at 10,000 PLN. The starting point for a typical scope is 10-15k PLN, scaling up with the number of roles covered by the system and the complexity of the work. I don't quote before the free process study, because without a map I don't know what I'm quoting. The client pays separately for the LLM subscription or API key. A solo operator with a different scale of needs? Installs from 1,500 PLN live in the labs section.
n8n, Zapier or a dedicated AI agent?
An off-the-shelf tool is enough for simple, repeatable data flows, and a dedicated AI agent belongs where the system has to make decisions. n8n and Zapier are good tools for connecting apps when the flow between systems is clearly defined. An AI agent classifies, prioritises and answers differently depending on content, instead of only moving data from A to B. In practice I often build a hybrid: n8n does the data flow, the agent steps in where judgement is needed. I match the tool to the problem at the audit stage.
What's the difference between an AI agent and a chatbot?
An AI agent runs a task end to end, while a chatbot answers a question and stops there. An agent uses tools, checks results, fixes its own mistakes and comes back with an outcome. Built around your company's data it knows your procedures, where to pull information from and how to assemble a finished document. The difference comes from the configuration and structure around the model, not the model itself.
How long does an AI implementation take?
A pilot is measured in weeks, not quarters. We set the schedule after the process study, once it is clear what is being deployed. The system grows in stages: the pilot first, then the remaining roles. The total time depends on the number of processes and the availability of the client's team.
Is my company's data safe?
Company data stays in accounts the client owns. You have your own API key or subscription, so requests go straight to the model provider on the terms of your agreement with them, not through my server. Model providers do not use API requests for training by default, and on a consumer subscription I turn off data sharing for training in the account settings. In systems hosted on the client's infrastructure, the agent's work history stays on site, with the client. Human-in-the-loop covers the points where the agent touches sensitive data.
Do I need an IT department to maintain this?
No, you do not need an IT department for day-to-day operation. Norbert Buszek, a personal trainer, runs his system himself after the documentation handover and training. Every implementation includes operational documentation and team training for daily use. A retainer from 1,000 PLN/month covers updates, analytics and technical repairs. It is optional. Technical clients like Norbert usually run it themselves.
// next step
Start with a free process study
In a single call we map where the biggest potential in your company is. You leave with a pilot plan, no commitments. If you'd rather read the audit first, order the free PDF.