LusivisionLUSIVISION

AI Agents

AI agents that act, not just answer

An agent queries your systems, carries out actions and knows when to stop and ask for approval. We build agents with clear boundaries and real oversight.

Discuss your use caseSee our work

What an AI agent is

An AI agent is a system that takes a goal, decides what steps to take to reach it and uses tools to carry them out — querying a database, creating a record, sending an email, booking a slot.

That is where it differs from a chatbot. A chatbot replies from a script or a model; an agent acts on your systems. One can tell you which slots are free on Thursday; the other books Thursday, sends the confirmation and updates the calendar.

That ability to act is what makes agents useful and what makes them require care. An agent with write access to real systems needs explicit boundaries, a record of what it did, and points where a person approves before an action happens.

Agents we build

Customer support agents

They answer recurring questions from your actual documentation, look up the status of an order or request, and resolve what can be resolved without a person.

Success is not answering everything. It is handling routine well and escalating the rest quickly, with context attached. An agent that insists on answering when it does not know costs more than it saves.

  • Answers grounded in your documentation
  • Order and request status lookups
  • Escalation with the context already gathered
  • Explicit limits on what it should not attempt

Sales and lead qualification

An agent that replies to a new enquiry within minutes, asks the questions your sales team would ask and records the answers in the CRM.

The value is mostly response time and consistency: qualified leads reach a person already in context, and enquiries that were never a fit consume no selling time.

  • Fast first response to inbound enquiries
  • Consistent qualification questions
  • Automatic CRM logging
  • Hand-off to a person on real intent

Booking and scheduling

Agents that check real availability, propose times, confirm and send reminders — cutting no-shows and back-and-forth messaging.

This needs integration with your existing calendar and clear rules on duration, notice period and service types. It is a case where deterministic logic does most of the work.

  • Integration with existing calendars
  • Automatic confirmations and reminders
  • Duration and notice-period rules
  • Rescheduling and cancellation

Internal knowledge agents

For internal use: answering questions about procedures, contracts or history, and carrying out administrative tasks on behalf of whoever asks.

They respect existing permissions — an agent should never surface information to someone who could not reach it through the normal route.

  • Search across internal documentation
  • Execution of administrative tasks
  • Permissions inherited from the user
  • Complete record of actions taken

RAG and tool integration

An agent is only as good as its access. We connect it to your documentation through RAG so answers rest on verifiable facts, and to your APIs so it can act.

Each tool is defined explicitly: what it does, what data it needs and whether it requires approval. Nothing is left implicit.

  • Documentation access with source citation
  • Integration with CRM, calendar and internal systems
  • Explicit definition of every available tool
  • Read actions separated from write actions

Agent, automation, or person?

Not everything that looks like an agent problem is one. This is the distinction we draw before proposing one.

An agent fits when

  • Requests arrive in natural language and vary case by case
  • Answering requires looking across several systems
  • Volume justifies instant response outside working hours
  • Real documentation exists to ground answers in
  • The actions to be taken can be bounded clearly

Probably not when

  • The process is fixed and predictable — plain automation is cheaper and more reliable
  • Each mistake carries serious financial or legal consequences
  • There is no documentation to base answers on
  • The conversation needs judgement, negotiation or genuine empathy
  • Nobody is going to review what the agent has been doing

How we implement

From the simplest case outward, with supervision from day one.

  1. 01

    Bounding

    We define what the agent may and may not do, which tools it can use and where a person must approve before an action happens.

  2. 02

    Build

    We connect documentation and tools, then test against real cases from your own history — including the awkward and ambiguous ones.

  3. 03

    Supervised launch

    It goes live with human review over its actions, so you learn where it is right and where it is wrong before it gains autonomy.

  4. 04

    Widening

    Based on the real log, we widen what it does unaided and keep approval where approval is still warranted.

Reliability and safety

An agent with access to real systems is a liability as much as an asset. These are the mechanisms we treat as mandatory rather than optional.

Control

Human approvalPer-action limitsRead/write separation

Observability

Action loggingAlertingCost monitoring

Security

Inherited permissionsData minimisationAudit trail

Where it typically applies

Common scenarios where an agent adds value. These are illustrative of the kind of work we do.

First-line support

Resolving routine requests around the clock and passing the rest to the team with context already gathered.

Responding to quote requests

Collecting the missing information, logging the request and notifying the right person, without leaving the enquiry waiting.

Service booking

Proposing times from real availability and confirming them, with automatic reminders before the date.

Internal procedure support

Answering questions about internal processes with the applicable document cited, within the asker's permissions.

Frequently asked questions

What is the difference between an AI agent and a chatbot?

A chatbot converses; an agent acts. The chatbot answers from a script or a language model. The agent queries your systems and performs actions — creating records, booking, updating. In practice: a chatbot tells you Thursday is free; an agent books you in on Thursday.

Can the agent do something wrong in our systems?

Only within what it is permitted to do. We define explicitly which actions are available and which require human approval before they happen. Consequential actions — deleting, invoicing, communicating with clients on the company's behalf — sit behind approval as a matter of course.

What happens when the agent does not know?

It should say so and escalate. We configure the boundaries so it recognises when it is outside its scope and hands over to a person with the context already gathered. An agent that invents answers is worse than no agent at all.

How much does an AI agent cost to build?

It depends on how many tools it must integrate with and the level of reliability required. An agent that only looks information up is substantially simpler than one taking actions across several systems. We estimate once the scope is bounded.

Will customers know they are talking to an agent?

Yes, and we recommend making it explicit. Beyond being the right practice, it avoids the frustration of someone discovering mid-conversation that they were not talking to a person.

Do we need a broader AI project first?

Not necessarily. An agent can be a company's first AI project provided the case is well bounded. If your need is mostly answering and processing rather than acting, the AI development page covers that scope better.

Related areas

Services that often accompany an agent project.

AI Development

AI development services for businesses: language model integration, RAG systems, AI automation and internal assistants — built with clear boundaries, human review and measurable outcomes.

Custom Software

We build custom software around how your business actually works: web applications, business platforms, system integrations and process automation. Discovery, delivery and long-term maintenance.

Software for SMEs

Software for small and medium businesses: replacing spreadsheets and manual processes with management systems, automation and AI, sized to an SME's budget and timeline.

Related reading

Useful background on AI agents.

Which process would you hand to an agent?

Describe the process. We will tell you whether an agent is the right approach, whether plain automation would serve better, and what it would take to test it.

Get in touchSee all services