# Your First AI Agent in Production in 6 to 8 Weeks | AvantIT > One AI agent in production, with real users and agreed metrics: discovery, build, pilot and handover to operations in about 6 to 8 weeks. Fixed scope. Canonical: https://www.avantit.pt/en/offers/first-ai-agent/ Source: AvantIT — https://www.avantit.pt (Microsoft Partner, Matosinhos, Portugal) --- Fixed-scope offer # Your team's first AI agent, in production and measured Pick a process that eats hours of your team’s time today and have an AI agent working on it, with real users and metrics agreed up front. In about 6 to 8 weeks, from discovery to the decision to scale. Duration About 6 to 8 weeks Format Fixed scope, 4 phases Price Fixed, agreed after discovery [Request a proposal](https://www.avantit.pt/en/book-a-demo/) [Take the free diagnostic](https://www.avantit.pt/en/maturity-diagnostic/) ## Who it is for - A repetitive, rule-based process eats hours every week: checks, reports, follow-ups, requests. - You have tried ChatGPT or Copilot and want an agent connected to your company systems, following your rules. - You need a first measured case, with security and human oversight, before investing in more agents. ## What you get - A process and rules document for the agent, written with your team - A working agent, connected to your systems and respecting each user’s permissions - An audit log of every check, alert and action: what, when, with which data and who approved it - A dashboard or report of the agreed metrics, measured from the first week of the pilot - An EU AI Act risk classification note for the agent - An operations runbook: monitoring, rule tuning and what to do when something fails - A go/no-go recommendation: scale, widen autonomy, adjust or stop ## How it works: 4 phases in about 6 to 8 weeks - 01 · Weeks 1–2 ### Discovery We choose the process, write the rules with your team, confirm the data sources and access, and fix the success metrics and the agent’s level of autonomy. - 02 · Weeks 2–4 ### Build We connect the agent to your systems (Microsoft 365, Google Workspace, ERP, CRM or APIs) and pick the model per task: OpenAI, Claude, Gemini or open-source. “Draft for approval” mode is switched on. - 03 · Weeks 4–7 ### Pilot A group of real users works with the agent day to day. We review the agreed metrics every week and refine the rules based on what users approve, edit or reject. - 04 · Week 8 ### Handover Security, permissions, audit logging, monitoring and training. We close with the decision: scale to more users, widen autonomy or move on to the next agent. ## Typical agents to start with - Executive Assistant Agent - Financial Controlling Agent - Team Operations Agent - Capex Management Agent - Quality Operations Agent ## What we need from you - A process owner, available to write the rules with us and review the results every week - A group of real users for the pilot - Access to the data sources (Microsoft 365, Google Workspace, ERP, CRM or APIs), granted by your IT - The cloud or servers the agent will run on: Azure, Google Cloud, AWS or your own infrastructure ## What is not included - AI model licences and consumption, billed directly by the provider - More than one agent or more than one process - Automated decisions about people, such as recruitment or performance reviews - Large-scale data clean-up or migration; if the sources are not ready, discovery shows it - Ongoing operation and improvement after handover, which can be contracted separately ## When this is not the right first step If the process mostly depends on case-by-case judgement and nobody can write down the rules, an agent is not the place to start. If the process is always the same and never needs to read text or interpret documents, an automation with Power Automate or n8n does the job at lower cost. ## Frequently Asked Questions Get answers about this service How much does the first agent cost? The price is fixed and agreed at the end of discovery, once the process, the data sources and the autonomy level are defined. AI model licences and consumption are billed by the provider, on consumption; we estimate them during discovery for the expected volume. Which AI model do you use? We pick the model per task: OpenAI, Claude, Gemini or open-source. The architecture is model-agnostic, so you can switch models later, when another is better or cheaper, without rewriting the agent. Where does the data live? In your company’s infrastructure — Microsoft 365 and Azure, Google Workspace and Google Cloud, AWS or your own servers — with EU data residency. The agent only sees what its user can already see, and your data is not used to train models. Does the agent send emails or make decisions on its own? Not by default. Anything that goes outside — replies, follow-ups, opinions, reports — stays as a draft for human approval. Low-risk internal actions can be automatic, and autonomy only grows when the pilot metrics justify it. What about the EU AI Act? We classify the agent’s risk and document its purpose, the data it uses and the human oversight. Productivity agents like the ones in this offer do not make decisions about people and, as a rule, fall outside the high-risk categories. What happens at the end? You receive the agent in production, the runbook, the metrics report and a go/no-go recommendation. You can run the agent with your own team or with AvantIT, scale it to more users or move on to the next agent; the decision is yours. ## Worth reading before you decide - [AI agents for business](https://www.avantit.pt/en/solutions/ai-agents/) - [AI agents in Microsoft 365: what changes in 2026](https://www.avantit.pt/en/blog/ai-agents-microsoft-365-2026/) - [How to calculate AI ROI](https://www.avantit.pt/en/blog/how-to-calculate-ai-roi-company-examples/) - [EU AI Act: what your company must do](https://www.avantit.pt/en/blog/eu-ai-act-company-2026/) [See all offers →](https://www.avantit.pt/en/offers/) ## Want to know if this offer fits your company? In a 30-minute call we confirm the scope, then send you a fixed-price proposal. [Request a proposal](https://www.avantit.pt/en/book-a-demo/)