https://ragstone.ai/en/enNew answers and explanations about AI agents by Hesham Elmahdy. https://ragstone.ai/en/questions/stop-agent-repeating-mistakes/https://ragstone.ai/en/questions/stop-agent-repeating-mistakes/Tue, 27 Oct 2026 08:00:00 +0000Write the correction down where the agent reads it every time, not only in the conversation where you made it. A correction said once is forgotten when the conversation ends. A correction written as a standing rule applies from then on. https://ragstone.ai/en/questions/instructions-by-voice/https://ragstone.ai/en/questions/instructions-by-voice/Tue, 27 Oct 2026 08:00:00 +0000Yes. Most of my own instructions are dictated. But speech-to-text gets names wrong, so the agent must read through the errors, and must check with you before acting on any name where a wrong guess would change the result. https://ragstone.ai/en/questions/clean-data-before-ai-agent/https://ragstone.ai/en/questions/clean-data-before-ai-agent/Tue, 27 Oct 2026 08:00:00 +0000No, not all of it. Only the data your first task depends on has to be in order. If you wait until every system is clean, you will never start. But an agent will not fix messy data for you. It will repeat the mess, and sound sure while doing it. https://ragstone.ai/en/how-i-work/https://ragstone.ai/en/how-i-work/Tue, 27 Oct 2026 08:00:00 +0000I run my own follow-up work on a small set of AI agents. They read my email, calendar, meeting notes and messages, keep track of what was promised and by whom, and prepare drafts. They never send or delete anything, and never change anything already in my mailbox or in our company system. A separate agent checks the whole system every day. Here is how it is built, what went wrong, and what I would tell any leader who wants to do the same. https://ragstone.ai/en/basics/why-instructions-matter/https://ragstone.ai/en/basics/why-instructions-matter/Tue, 27 Oct 2026 08:00:00 +0000Because the agent knows nothing about how your company works except what you tell it. Its instructions, often called the prompt, are its job description. Clear instructions that explain the reasons behind them produce consistent work. Vague ones produce the agent's best guess at what you meant. https://ragstone.ai/en/basics/why-ai-makes-things-up/https://ragstone.ai/en/basics/why-ai-makes-things-up/Tue, 27 Oct 2026 08:00:00 +0000Because a language model is built to produce a likely answer, and the way models are trained and tested rewards a confident guess more than an honest "I don't know". The industry calls these invented answers hallucinations. They come from how the model works, so you deal with them through the way you design the work, not by hoping the next model will be free of them. https://ragstone.ai/en/basics/when-agent-should-ask-a-human/https://ragstone.ai/en/basics/when-agent-should-ask-a-human/Tue, 27 Oct 2026 08:00:00 +0000Before any action that is hard to undo or that commits the company: sending, paying, deleting, signing, or changing a record that others rely on. Below that line, the agent can work alone and report what it did. Deciding where that line sits is the leader's job, and it should be written down before the agent starts. https://ragstone.ai/en/basics/what-is-an-ai-agent/https://ragstone.ai/en/basics/what-is-an-ai-agent/Tue, 27 Oct 2026 08:00:00 +0000A chatbot answers your question and stops. An agent is given a goal and works towards it: it decides its own next step, uses tools such as your email or your business system, looks at the result, and carries on until the job is done or it needs you. The real difference is who decides the next step. https://ragstone.ai/en/basics/what-is-a-language-model/https://ragstone.ai/en/basics/what-is-a-language-model/Tue, 27 Oct 2026 08:00:00 +0000A language model is a program trained on a very large amount of text, learning to predict what comes next. That is how it learned to write, summarise and reason. But your contracts, prices and clients were never in that text, and its knowledge stops at a fixed date. Whatever it needs to know about your company, someone has to give it. https://ragstone.ai/en/basics/what-is-a-context-window/https://ragstone.ai/en/basics/what-is-a-context-window/Tue, 27 Oct 2026 08:00:00 +0000The context window is everything the model can see while it works on your request: its instructions, the documents it was given, the conversation so far, and its own answer. It is the model's working memory. It has a fixed size, and as it fills up, the model pays less careful attention to what is in it. What falls outside it, the model does not know. https://ragstone.ai/en/basics/how-agents-use-tools/https://ragstone.ai/en/basics/how-agents-use-tools/Tue, 27 Oct 2026 08:00:00 +0000The model itself cannot open anything. It can only ask. When it needs data or an action, it writes a structured request; your software receives it, carries it out, and hands back the result. So what an agent can do in your company is decided by the tools you give it, not by the model. https://ragstone.ai/en/basics/fine-tuning-or-retrieval/https://ragstone.ai/en/basics/fine-tuning-or-retrieval/Tue, 27 Oct 2026 08:00:00 +0000For knowledge about your business, give it the data when it answers. This method is called retrieval: the agent first searches your own documents, then answers from what it found. Training the model further on your data, called fine-tuning, is in my view better suited to shaping how it writes and behaves. A 2023 study found it a weak way to teach it new facts. https://ragstone.ai/en/questions/who-is-liable-when-agent-errs/https://ragstone.ai/en/questions/who-is-liable-when-agent-errs/Mon, 26 Oct 2026 08:00:00 +0000In most cases, your company is. The agent is not a person the law can hold to account, so what it says or does on your behalf is treated as what your company said or did. A contract with the vendor may share the cost with you, but your client will hold you responsible. The real decision is not who pays after the mistake. It is what the agent is allowed to do before it. https://ragstone.ai/en/questions/where-to-start-with-ai-agents/https://ragstone.ai/en/questions/where-to-start-with-ai-agents/Mon, 26 Oct 2026 08:00:00 +0000Start with one task, not a company-wide program. Pick work that repeats often, whose information is already in order, whose result a person can judge in a minute, and whose mistakes are cheap to correct. Let the agent prepare and a person approve. Widen its work only after it has earned your trust. https://ragstone.ai/en/questions/where-does-my-data-go/https://ragstone.ai/en/questions/where-does-my-data-go/Mon, 26 Oct 2026 08:00:00 +0000Whatever the agent reads is sent to the model to be processed. Unless you run the model yourself, that means it goes to the company that runs the model, and is processed and often stored wherever that company does it. Some providers now let business clients choose the region. Whether that is enough depends on which of your data is sensitive, and on the law in your country and your sector. https://ragstone.ai/en/questions/build-or-buy-ai-agent/https://ragstone.ai/en/questions/build-or-buy-ai-agent/Mon, 26 Oct 2026 08:00:00 +0000Almost no company should build the AI model itself. That part you buy. The real decision is who owns the part that knows your business: your instructions, your rules and the connection to your systems. Buy where the work is the same in every company. Own the part that depends on how your company works. https://ragstone.ai/en/questions/ai-agent-read-erp-safely/https://ragstone.ai/en/questions/ai-agent-read-erp-safely/Mon, 26 Oct 2026 08:00:00 +0000Yes, if you treat the agent like a new employee you do not fully trust yet. Give it its own account, let it read only, and let the ERP system itself decide what it may see. The danger is not that the agent reads your data. The danger is giving it more access than any person in that role would get. https://ragstone.ai/en/questions/agent-forgets-what-we-agreed/https://ragstone.ai/en/questions/agent-forgets-what-we-agreed/Mon, 26 Oct 2026 08:00:00 +0000It carries on without the agreement, and it does not tell you. An agent remembers only what is written somewhere it reads before it starts. Anything agreed only inside a conversation is lost when that conversation ends or grows too long. The fix is to write decisions down in a place the agent must read, not to hope it remembers. https://ragstone.ai/en/questions/who-checks-the-agent/https://ragstone.ai/en/questions/who-checks-the-agent/Sun, 25 Oct 2026 08:00:00 +0000Another agent checks the work, and you check the decisions. An agent left to grade its own work will tell you everything is fine. You need a second agent whose only job is to look for what went wrong, and a clear line between what that checker may fix on its own and what it must bring to you. https://ragstone.ai/en/questions/normal-report-or-ai/https://ragstone.ai/en/questions/normal-report-or-ai/Sun, 25 Oct 2026 08:00:00 +0000If a programmer could write an ordinary report against your database and the problem would go away, you do not need AI. You need the report. AI earns its place only where a report cannot reach. https://ragstone.ai/en/questions/confident-wrong-answers/https://ragstone.ai/en/questions/confident-wrong-answers/Sun, 25 Oct 2026 08:00:00 +0000Because it is answering from what it learned about the world in general, not from what is true in your company. A model trained on the whole internet knows a lot about business in general and nothing about your contracts, your prices or your clients. Unless you give it your own information at the moment it answers, it fills the gap with something that sounds right.