Question 1

What is an AI agent?

Accepted Answer

An AI agent is a program that gets a goal, decides what steps to take, runs those steps with the tools it has (search, file access, APIs), and returns a result without a human directing every action. An assistant answers your question. An agent does your task. Most of what I build sits between the two: systems that run a defined process start to finish, with a person reviewing the output instead of babysitting every step.

Question 2

What is memory in an AI system?

Accepted Answer

Memory is how an AI system holds onto information between sessions. Without memory, every conversation starts from zero. The system has no context about past interactions, your preferences, or your history. It's like talking to someone with complete amnesia every time you meet. With memory, the system can reference earlier conversations, learn your preferences over time, and build context incrementally. It remembers that you prefer short reports, that Client X always has special requirements, that the policy changed last month. I implement two types: Session memory: Context inside a single conversation. The system remembers what was said earlier in the current session. Persistent memory: Structured storage that survives between sessions. The system remembers things from last week, last month, last year. Persistent memory needs deliberate design decisions about what gets stored, how it's indexed, and who can read it. I work through that with you during the specification phase.

Question 3

How does an AI assistant differ from ChatGPT?

Accepted Answer

ChatGPT is general-purpose AI. It knows a lot about a lot of things, but nothing specific about your business, your clients, your documents, or your tone. A custom AI assistant is built on your context: your products, your pricing, how you communicate, what your policies say, and what to do when edge cases come up. A practical example. You ask ChatGPT: "What's our return policy for enterprise clients?" ChatGPT doesn't know your return policy. It gives you a generic answer about return policies in general, or it makes one up that sounds plausible. Your custom assistant, built on your actual policy documents, says: "Enterprise clients on the Plus plan have 60-day returns with a full refund. Standard enterprise clients have 30-day returns. Both require the original order number. Exceptions for defective products are handled under section 4.2 of the enterprise agreement." ChatGPT gives you a reasonable answer. A custom assistant gives you the answer that's correct for your situation and your rules.

Question 4

What does "working system" mean?

Accepted Answer

A system that runs the target process start to finish on real inputs without needing manual intervention at every step. That means: It runs on your actual documents, not demo data It handles the edge cases I defined in the spec It produces output that meets the acceptance criteria, set before the build started Your team can run it without contacting me for day-to-day use That doesn't mean: Zero errors (I document known limitations) Perfect performance on inputs outside the spec boundary Self-running or self-improving without human oversight

Question 5

What do I need to prepare before we start?

Accepted Answer

The detailed list is in the pre-consultation guide and the integration guide. Short version: Before Discovery: A description of the process you want to automate or improve — written down, not just "we'll talk about it" 5-10 real examples (actual emails, documents, reports — not the polished versions) A list of the tools you use in that process 60 minutes of uninterrupted time Before the build: Signed specification Access credentials for the relevant systems A named point of contact who can test and give feedback within 48 hours The most common mistake: showing up to Discovery with a vague idea instead of concrete examples. "We want to automate email" isn't enough. "Here are 50 emails from the last month, here's how we classify them, and here's where it breaks down" — that's what I need.

Question 6

How long does a project take?

Accepted Answer

Depends on the install flavor: openclaw (an agent in Telegram / WhatsApp / Slack / Discord): 5-10 business days from a signed spec hermes (a background worker: cron, recurring reports, triage): 5-10 business days custom (Claude Code / Codex CLI wired into your stack): 3-7 business days hybrid (presence layer + CLI as one system): 2-4 weeks The clock starts when the spec is signed and initial access is in place. If I'm waiting on your side for more than 5 business days, the deadline shifts accordingly. Fast feedback keeps projects on track.

Question 7

What happens after the project is delivered?

Accepted Answer

You get a handoff package with full documentation. Whatever the flavor: a free first month of support and retainer-level coverage (prompt tuning, small fixes, priority contact). During that month I check with you that the system actually changes your week. After the first month, the retainer is optional: 1,000 PLN/mo for single installs (openclaw / hermes / custom), 2,000 PLN/mo for hybrid. It covers model upgrades, tuning for new tasks, monitoring, and priority response. Cancel anytime with 30 days' notice. If you don't want a retainer, the system stays with you with full documentation and code. Support terms are in the legal documents.

Question 8

Can the AI make mistakes?

Accepted Answer

Yes. Every AI system makes mistakes. That's not a flaw in my implementation — it's how language models work. LLMs hallucinate: they generate text that sounds right but is factually wrong. What matters isn't whether the AI makes mistakes, but whether those mistakes get caught before they do damage. I build the workflow around that reality: multi-step verification, fact-checking against source material, confidence scoring, and human review on every high-stakes output path. How that works in practice is in how I work. I don't build systems where the AI has unreviewed authority over consequential actions: financial transactions, client-facing communications without approval, or deleting data. I document known failure modes in the handoff package and test against edge cases, not just the happy path.

Question 9

Do you sign NDAs?

Accepted Answer

Yes, with one limit: I won't sign an NDA that bars me from describing the general category of work I did (e.g. "an AI assistant for internal knowledge management"). I protect your specific documents, data, and business logic. I can't agree to hide that a system of a given type even exists.

Question 10

Do you work with clients outside Poland?

Accepted Answer

Yes. I work in Polish and English. All documentation, communication, and deliverables can be in either language. Location isn't a constraint — the whole process runs remotely.

Question 11

What if I'm not technical?

Accepted Answer

That's the norm. The process is built for it. I don't expect you to understand how the system works under the hood. I expect you to know what your process looks like today and what success looks like. You're the domain expert — you know your business. I handle the technical side. The specification phase translates your descriptions into technical requirements. If something needs technical knowledge from you, I'll tell you upfront and help you get ready. At handoff, I train your team to run the system. The documentation is in plain language, not technical jargon.

Question 12

What's the smallest project you take on?

Accepted Answer

The smallest install is a single Custom Agent Install from 1,500 PLN, in the flavor you pick (openclaw / hermes / custom). The discovery call before any install is free. Before that, you can start with the free AI Audit or geo-scan — no call, no obligation.

Question 13

Can I start small and expand later?

Accepted Answer

Yes. Start with one install, test it on real work. Once you've seen the results, you decide whether to add a skill (SEO Kit, Research Lab), pick a second flavor, or move to hybrid as one integrated system. That's the recommended path for people new to AI in their business. The first install gives you confidence in the technology, a working relationship with me, and real data on what works. When I see during Discovery that expansion is likely, I design the install with that in mind — I note it in the spec so the component can become part of a bigger pipeline without a rebuild.

Question 14

Do I need to understand programming?

Accepted Answer

No. You need to understand your process and your goals. I handle the technical side entirely. What you need: Access to your tools. I'll tell you exactly what credentials or permissions I need and walk you through granting them. Availability for feedback. During the build you'll test the system on real work and tell me what works and what doesn't. That's a few hours a week, not full days. Willingness to answer questions. I'll ask about your processes, edge cases, and preferences. Specific questions — not technical ones. What you don't need: Programming knowledge Understanding of AI models or APIs Technical vocabulary A "technical person" on the team (though it doesn't hurt) The process is designed for people who describe their tasks in plain language. If you can explain what you do and what you want, I can build it.

Question 15

What happens when the AI model gets updated?

Accepted Answer

Your system gets better. I design every system to be model-upgradeable. When Anthropic ships a better version of Claude or OpenAI ships an improved GPT, your assistant gets smarter without a rebuild. What a model upgrade looks like in practice: I swap the base model, run the test suite against your acceptance criteria, confirm everything works as expected (or better), and ship the update. Your system prompts, knowledge base, and integrations stay the same. Only the model underneath changes. If you're on a retainer, model upgrades are included. I handle the upgrade, test it, and ship it. You get a note that your system is now on the next model version, and you carry on. If you're not on a retainer, I can do the upgrade as a one-time job with a quote upfront.

Question 16

Can I switch the AI model later?

Accepted Answer

In most cases, yes. I design systems to be as model-agnostic as I can. The knowledge base, configuration, integration layer, and business logic don't change when you switch models. Only the AI model underneath changes. For standard installs, switching from Claude to GPT or back is straightforward. The system prompts may need tuning — different models react differently to the same instructions — but the core of the system stays intact. For deeper integrations that lean on model-specific features, switching can take more work. I'll tell you during the spec if the project has model-specific dependencies. I never lock you into one AI provider. If a better model shows up, you should be able to use it.

Question 17

How is my data protected?

Accepted Answer

Data protection is built into every project from day one, not bolted on at the end. The detailed rules are in the integration guide and the Privacy Policy. Short version: Encryption at rest and in transit. No exceptions. API keys and tokens kept in secure storage, never in source code. Access only to the systems and data the project needs, with the minimum permissions required. When the project ends, my access is revoked. I don't keep your data after the project ends unless you specifically ask for ongoing maintenance.

Question 18

What does "human in the loop" mean?

Accepted Answer

For any decision with real consequences — financial, customer-facing, data deletion, legal — a human checks before the AI acts. The AI prepares: it analyzes the data, drafts the version, generates the recommendation. Then it stops and waits for a human to approve, modify, or reject. For routine, low-risk tasks the AI runs on its own. The line between "autonomous" and "human-reviewed" is set during the spec — you decide where it falls. The detailed process is in how I work.

Question 19

Can I see the code?

Accepted Answer

Yes. Every project ships as a GitHub repository with full access. The code is yours — not licensed to you, not hosted on my infrastructure with limited access. Yours. Fully. You get the complete source code with change history, configuration files, documentation, and the test suite. You can hire another developer to modify it, move to a different provider, fork the project. IP ownership details are in the legal documents.

Question 20

What about GDPR?

Accepted Answer

I design every system for GDPR compliance from the start. It's not an add-on or an afterthought. Key points: Personal data processing is documented explicitly in the spec. I don't use your data to train models. Your documents and business data are used only for your project. Retention policies are defined per project during the spec. For projects that process personal data, I provide a standard data processing agreement (DPA). Systems that handle personal data include the ability to delete specific records on request (GDPR Article 17). Full details in the Privacy Policy.

Question 21

How do you handle confidential business information?

Accepted Answer

Everything you share during Discovery, the spec, and the build is treated as confidential by default. I don't share your business processes, documents, or data with other clients. I don't use your project as a case study without explicit written permission. For clients who need formal confidentiality agreements, I sign NDAs (see the NDA question above for the one carve-out).

Question 22

What if the project fails?

Accepted Answer

I have a process designed to prevent that, but I'll answer straight. Before the build starts: Discovery identifies whether AI is the right tool. The spec defines exactly what I'm building. If I have doubts about feasibility during Discovery, I say so. During the build: I test on real data, iterate on your feedback, and measure against the acceptance criteria. If the acceptance criteria aren't met: I keep working until they are. The spec defines what "done" looks like, and both sides are bound by it. If I find mid-build that the approach won't work: I tell you right away. I propose alternatives, adjust the spec, or in rare cases recommend stopping the project. In that case you don't pay for work that doesn't deliver value. I don't start projects I don't believe I can deliver. That's what Discovery is for.

Describe the problem. Get a system.

AI agent on your machine.

Input data

Agents at work

Human in the loop

Output lands where it should

Your phone as the control panel.

Research brief before every meeting.

Team operator in Telegram.

Pick a system layer.

OpenClaw

Hermes

Custom CLI

Hybrid bundle

Two operators. Two directions. Both shipping.

Tomek

Norbert Buszek

Got a process that keeps coming back?