AI assistant vs AI agent: the four building blocks
The AI-Native Roadmap is built from four blocks: AI assistants, department brains, AI agents and multi-agent systems. This guide defines each one, compares them side by side and explains what each can see and what people still decide.
An AI assistant helps one person with one task when asked. A department brain answers questions across a department’s own data. An AI agent chooses and carries out its own steps towards a goal, within limits a person sets. A multi-agent system is a team of agents coordinated by an orchestrator agent. People keep every judgement.
How do the four building blocks compare?
An AI assistant responds to one person, a department brain answers from a department's own data, an AI agent acts on its own steps within set limits and a multi-agent system coordinates several agents. Each builds on the one before, which is why they arrive in that order on the AI-Native Roadmap.
| Building block | What it is | What it does | Who drives it | Phase | Example | What the person still does |
|---|---|---|---|---|---|---|
| AI assistant | A chat-based tool connected to the apps one person uses | Drafts, summarises and answers when asked | The person using it | Phase 1 | A school office drafting a letter to parents | Checks and sends every output |
| Department brain | AI that answers from one department's own data | Answers questions across documents, records and history, with sources | The department lead | Phase 2 | A charity fundraiser asking what a trust was told last year | Judges whether the answer is right and what to do with it |
| AI agent | Software that pursues a goal by choosing its own steps | Analyses, enriches, monitors and acts within set permissions | The department lead, with Hartz AI | Phase 2, expanding in Phase 3 | A sports club agent that spots lapsed members and drafts a win-back message | Approves anything that reaches a member or a decision |
| Multi-agent system | A team of agents coordinated by an orchestrator agent | Splits a larger job across specialist agents and assembles the result | The department lead, with Hartz AI | Phase 3 | A synagogue office handling a booking from enquiry to invoice | Makes every strategic and customer-facing decision |
The four blocks share one goal, the one the roadmap names on every page: An AI-native business is one where AI assistants, department brains and agents do the repetitive work in every team, with people keeping the judgement.
What is an AI assistant?
An AI assistant is a chat-based tool that helps one person with one task at a time, using only the information it is connected to.
You ask, it drafts, you check. Assistants are built into most office suites and business systems, which is why they are the first thing a team adopts in Phase 1. They are good at writing, summarising, rewording and answering general questions.
They are not good at knowing your organisation. An assistant answers from its training and from whatever files it can see, so it can sound confident about things it has never been told. In a Harvard Business School field experiment with 758 knowledge workers (opens in a new tab), those using AI completed tasks 25.1% more quickly on average, but on a task chosen to sit outside the AI's capabilities they were 19% less likely to produce correct solutions.
What is a department brain?
A department brain is a knowledge layer that answers questions across one department's own documents, records and data, with sources a person can check.
Where an assistant knows the world, a brain knows your department. Ask it what a customer was promised, which policy applies or what happened last time, and it finds the relevant records first and answers from them, citing where each point came from.
Brains belong to Phase 2, because they depend on the clean, well-organised data that Preparation and Phase 1 create. A brain built over messy files gives confident answers from the wrong version.
What is the difference between an automation and an AI agent?
An automation follows fixed rules that a person wrote, while an AI agent decides its own next step within limits a person set.
An AI agent is software that works towards a goal by choosing and carrying out its own steps, within limits and permissions a person sets.
An automation says: when an invoice arrives, save it here. It never varies, which makes it safe and predictable, and it cannot cope with anything its rules did not foresee. Phase 1 uses automations like this.
An agent is given a goal instead: find members who have lapsed, work out why from their history and draft a message for each. It reads, reasons and acts, and a person approves what it produces before anything reaches a member. Agents arrive in Phase 2, once a department brain gives them reliable information to work from.
What is a multi-agent system?
A multi-agent system is a team of AI agents, each with one job, coordinated by an orchestrator agent that assigns work and does none itself.
Think of it as an agent team. One agent gathers information, another checks it against policy, a third drafts the reply and the orchestrator agent decides who does what and in which order. Splitting the work keeps each agent simple enough to test and trust.
Hartz AI's investment committee demonstration in the Property AI Lab shows the shape: seven specialist agents pressure-test one property deal and draft the paperwork behind a recommendation. It is scripted to show the reasoning and uses illustrative figures. In a real build a person signs off the call.
Multi-agent systems belong to Phase 3 and stay inside one department, where the data and the accountability are clear. Most organisations are some way off: the Office for National Statistics (opens in a new tab) found that UK businesses that use AI use 1.6 AI technologies each on average, and McKinsey's 2025 global survey (opens in a new tab) found that nearly two-thirds of organisations have not yet begun scaling AI across the enterprise.
What can each building block see?
Each building block sees only what its connections are scoped to see, and it inherits any permissions that are too loose. Scoping those connections is work Hartz AI does in discovery, not a property that comes free with the tool.
If a shared drive lets everyone open the HR folder, an assistant connected to that drive can quote from it to anyone who asks. Nothing has been hacked; the AI has simply been allowed to read what people were always technically allowed to read and never did.
The risk grows with each block. A brain can search a whole department at once, and an agent can act on what it finds. That is why each phase begins with a check of who can see what, and why the AI governance work runs alongside the build rather than after it.
Which building block do we need first?
Start with AI assistants, almost always. They use tools you already pay for, they build the habits every later block needs and they show quickly which questions need a department brain.
Agents and multi-agent systems are where bespoke build usually enters, because platforms rarely fit a department's exact process. The people who design and build them at Hartz AI are on the team page, and bespoke AI solutions explains how that work is scoped.
Hartz AI does not build an agent for a task an AI assistant and a person can already do well.
Common questions
Frequently Asked Questions
Work out which building block you need
A discovery call looks at your teams and data and tells you whether the next step is an assistant, a brain or an agent.