# AI Agent Builds _Software that acts on your behalf_ > We design, build and operate AI agents that finish real work, from qualifying leads to back-office operations, with clear limits on what they do without asking. - Tags: AI, Build, Service - Source: https://www.insidethebox.dev/services/ai-agents ## Overview An agent is software that works towards a goal on its own: it reads a request, decides what to do, uses your tools to do it and reports back. The useful ones do one job well. We build agents around a single workflow at a time, such as qualifying inbound leads, triaging support tickets or preparing research briefs, and connect them to the systems your business and team already uses, with clear limits on what they may do without asking. Our guide [AI agents in production](/insights/ai-agents-in-production) covers what it takes to run one safely. ## What we build Most agent projects are one of four kinds. | Agent | What it does | Typical tools | | ---------------- | ---------------------------------------------------------------------------------------------- | ------------------------------------ | | Sales assistant | Researches inbound leads, answers their questions, books meetings and updates the CRM | CRM, calendar, email, Slack | | Support agent | Answers customers from your help centre and account data, and hands over what it can't resolve | Help desk, order and account systems | | Operations agent | Reads documents and moves data between systems: invoices, onboarding, reporting | ERP, spreadsheets, internal APIs | | Research agent | Gathers sources, compares them and writes a brief with citations | Web search, internal documents | ## How an agent works Every agent runs the same loop: it reads the request, plans, calls a tool, reads the result and decides the next step, until the job is done or it needs a person. The scene follows one inbound lead through that loop: a look-up of the company, a check of the CRM, a search of the documentation, then a reply that answers both questions and offers a call. Each step is logged with its inputs, its output and how long it took, so any decision the agent made can be traced afterwards. ## Knowledge Models don't know your prices, products or policies, and what they learned in training goes out of date. So the agent looks things up when it needs them. Your documents are split into passages and indexed by meaning, the question is matched against them, and the best few passages go into the model's context. The reply cites the passages it used, so your team can check where each claim came from. When nothing relevant turns up, the agent says so instead of guessing. ## Tools and actions An agent acts through tools: functions with a name, a description and typed inputs that call your CRM, your calendar or your email. Each tool gets only the access its job needs, read-only where reading is enough and one named channel rather than the whole workspace. A call outside that list is refused, not attempted. Where it fits, we connect tools through the Model Context Protocol (MCP), an open standard now governed under the Linux Foundation, so the same integration works across models and clients. Tool descriptions from third-party servers are treated as untrusted input, and every connection gets the narrowest scopes it needs. ## Approvals Some actions should wait for a person: a discount above the agent's limit, a refund, a first email to a new contact. The agent prepares the action with its reasons and asks where your team already works, such as Slack, and nothing happens until someone approves it. As the agent proves itself in review and on the test set, its limits can rise step by step. ## Evals Before an agent ships, and after every change to its prompt, model or tools, it runs a test set of real cases with known good outcomes. Each answer is scored by code where it can be checked exactly, by a grading model where it can't, and by people for a sample. A change ships only if it clears the gate. The chart shows one test set run across many configurations: most fall short, a few pass, and the best one ships. ## Process Start with one workflow and a measured baseline, prototype on real cases, evaluate before go-live, then operate with review and monitoring. The agent's autonomy grows as its scores hold. 1. **Discover** — One workflow with clear inputs, outputs and a measured baseline. 2. **Prototype** — A working agent on real cases within weeks, not months. 3. **Evaluate** — Accuracy, cost and failure modes scored before go-live. 4. **Operate** — Human review and monitoring from day one, and limits that widen as the scores hold. ## Deliverables A production agent, its integrations and knowledge sources, an evaluation suite, a review console for your team, and monitoring with runbooks. You own the prompts, the code and the data. - Agent design brief - Tool and API integrations - Knowledge retrieval - Evaluation suite - Human review console - Monitoring and runbooks ## Stack Models from several providers, chosen per task on the evaluation scores; retrieval over your documents and data; typed tools, connected through MCP where it fits; tracing, evaluation runs and cost monitoring in production. ## FAQ Answers to the questions we hear most often about agent projects. ### What's the difference between an agent and a chatbot? A chatbot answers; an agent acts. An agent decides which steps to take and uses tools to take them, such as updating a record or booking a meeting, within limits you set. ### How long does an agent build take? Usually four to ten weeks for a first workflow: one or two weeks to map it and collect test cases, a prototype on real cases soon after, then evaluation and a supervised launch. ### What stops an agent from making costly mistakes? Every tool gets only the access its job needs, high-impact actions wait for a person to approve them, inputs are checked for instructions hidden in emails or documents, and every release has to pass the evaluation set first. ### Can agents use our existing systems? Yes. Agents act through the APIs your software already offers, and we build a connector where there isn't one. Where it fits we use the Model Context Protocol (MCP), an open standard that ChatGPT, Gemini, Microsoft Copilot and others support, so one integration works across models. ### Do people have to be told they're talking to an AI? In the EU, yes: the AI Act requires that people are told when they're interacting with an AI system unless it's obvious, and that duty applies from 2 August 2026. We make the agent say so in its first message wherever it talks to customers or prospects. ### Can an agent email prospects for us? Within the same rules your business. Consent rules for business email differ by country (Germany generally requires consent, France and the UK allow relevant B2B email with an opt-out), and in the US the CAN-SPAM Act applies to B2B email too. We build in the sender's identity, an opt-out, and approval for first contact. ### Which model do you use? The one that scores best on your test set at an acceptable cost and speed. We often pair a larger model for planning with smaller, faster ones for routine steps, and the agent can switch providers without a rebuild. ### How do you measure whether it's working? Against the baseline we agree at the start: response time, the share of work finished without a handoff, the error rate in review and the cost per task. The evaluation set runs again on every change. ### Who maintains the agent after launch? We can operate it for you, with monitoring, evaluation runs and a monthly review, or hand it over with runbooks and train your team to run it. ## Capabilities - Orchestration (Agents) — proficiency 4/5 - Retrieval (Data) — proficiency 5/5 - Tool integration (Engineering) — proficiency 5/5 - Evaluation (Quality) — proficiency 4/5 - Observability (Platform) — proficiency 4/5 ## In numbers - Typical build: 6 wks - Eval cases per agent: 200 +