Bot, Agent, Copilot or Automation: What Does Your Task Actually Need?

You want to stop copying the same information, prepare a report or answer recurring questions. Someone suggests a bot. Someone else suggests an agent or Copilot. Which one do you need?

Start with the work. A scheduled report may need a flow. A document question may be handled by an existing Copilot feature. A process with changing requests may benefit from an agent. A single solution can combine all three.

This technical companion to Make Existing Tools Useful in Everyday Pharma Work explains the choices. Microsoft product information was checked on 6 October 2026. Product availability, licensing and the settings in your organisation must be checked before implementation.

Compare the behaviour, then choose the product

The terms describe different aspects of a solution. This is a practical comparison, rather than four mutually exclusive product categories.

Term	What it describes	Example	What to ask
Bot	A conversational interface, often in Teams	A chat command retrieves the status of a request	What runs behind the chat?
Programmatic automation	Explicit steps and rules implemented in a flow, script or application	Every Friday, collect figures and prepare a report	Are the steps known and repeatable?
Copilot	Microsoft's assistant products and features	An employee asks for a draft or summary	Which Copilot, with which information and entitlement?
Agent	A specialised assistant configured for a purpose, potentially with tools and actions	Find relevant documents, ask a question and prepare a request	Which decisions can it make, and which actions can it perform?

An AI-generated answer does not prove that a system can take action. A chat window does not prove that it uses AI. And an agent does not necessarily run without a person.

A bot is the conversation, not a guarantee of intelligence

Microsoft's bot documentation describes conversational applications ranging from scripted interactions to AI-powered assistants. A Teams bot can respond to a command, collect information or call another service. An agent can be presented through that same interface. Microsoft bot overview

For example, a user could type “show my open requests”, and the bot could call a predefined query. That does not require a language model to plan the work. If free-form conversation and interpretation are needed, AI may become part of the implementation.

Here, “bot” means a conversational bot. An RPA bot is different: it automates interactions with an application interface.

Programmatic automation follows an authored process

In Power Automate, cloud flows can start from an event, a button or a schedule. Desktop flows automate interactions with desktop and web applications. These are useful options for repeatable steps, including work involving older applications. Power Automate flow types, desktop flows

Scripts and custom applications are another implementation route when the task needs code, specialised calculations or an integration that does not fit the available flow tools. This is an engineering choice, not a requirement to introduce AI.

Within Microsoft, Office Scripts can automate Excel work and be called from Power Automate. Azure Functions can run custom code on events or expose an API. Azure Logic Apps provides another workflow and integration platform, typically evaluated with the team responsible for enterprise integrations. These options need their own access and deployment decisions. Office Scripts, Azure Functions, Azure Logic Apps

Illustrative task: Collect this week's approved figures, calculate totals, create a report and send it for review. Define the sources, calculations and review step explicitly. If summarising comments needs AI, add it to that step and check the result.

Calling a language model inside a flow does not make every other step autonomous. Equally, using Copilot to help write a flow does not mean Copilot runs the process afterwards. Microsoft documents Copilot in Power Automate as assistance for creating and working with automation. Copilot in Power Automate

Copilot is an assistant experience; Copilot Studio builds solutions

Microsoft Copilot Chat provides an organisational AI chat experience for eligible Microsoft 365 users. Microsoft 365 Copilot licensing adds capabilities; access to organisational information depends on the experience, entitlement and configuration. Do not assume that seeing a Copilot button gives access to every company document or every agent feature. Copilot Chat overview

Illustrative task: Prepare a first draft from material the employee is allowed to use. Check whether existing Copilot capabilities cover that task before building another assistant.

Copilot Studio is the building and management platform for agents and automation. It is distinct from the assistant the employee uses. A configured solution can be made available in supported channels, subject to the organisation's controls. Copilot Studio overview

GitHub Copilot is another product, aimed at development work. It should not be used as shorthand for Microsoft 365 Copilot.

Agents range from focused knowledge assistants to adaptive processes

In Microsoft 365 Copilot, declarative agents configure Copilot with instructions, knowledge and, where supported, actions. They use Copilot's models and orchestration. Custom engine agents use a separately controlled engine and can provide more specialised behaviour. These are architectural distinctions, not a ranking of usefulness. Microsoft's agent comparison

Agent Builder and SharePoint offer routes for focused knowledge assistants; Copilot Studio and developer tools support additional scenarios. The capabilities differ by tool. A simple document assistant should not automatically become a custom development project. Agent-building tools

Illustrative task: An employee asks about an internal procedure. A focused agent retrieves relevant information. If it can also prepare a request, that action needs its own access, validation and approval design.

Autonomy is a separate decision. A user-invoked knowledge assistant, an event-triggered agent and an agent allowed to change records have different responsibilities. Specify what starts it, which tools it may use and when a person must intervene.

Agent flows and workflows are not the same as agents

Copilot Studio's agent flows belong to the standard-harness experience. They define triggers and actions, can run independently or serve as tools for agents, and can include prompts, connectors and human approvals. The name does not mean the flow independently decides how to achieve a goal. Agent flows overview

The newer workflows experience runs on the GitHub Copilot harness. It provides a visual automation canvas with AI actions and agent handoffs. Workflows can also start from events or schedules, or be called by agents. Workflows overview

Microsoft describes these flows as deterministic. For implementation, distinguish the defined control path from the output of an AI step: a fixed sequence containing a generated summary does not guarantee identical summaries or error-free results. External information and services can change too. That is an engineering qualification of the product terminology.

Illustrative combined solution: An agent interprets a request and gathers missing information. A flow checks required fields and creates a draft record. A person approves the submission. Chat, reasoning and explicit execution each have a role.

What changed recently in Copilot Studio?

Current Microsoft documentation distinguishes three harnesses, the runtime foundations behind solutions: standard, GitHub Copilot and Copilot chat. They differ in orchestration, supported features and billing. The GitHub Copilot harness in Copilot Studio is not the GitHub Copilot service. Harness comparison, updated 1 October 2026

Microsoft announced the GitHub Copilot harness as generally available. Its September update separately labels capabilities such as Work IQ context, memory and connected agents as preview. General availability of the foundation does not establish availability of every feature in every tenant. Microsoft's September update

Some implementation documentation, including adding a workflow as an agent tool, still carries a prerelease notice. Check the status of the exact capability before using it in production, rather than inferring it from the overall announcement. Workflow tool documentation

When reading a tutorial, check which experience and harness it covers. “Agent flow” and the newer “workflow” should not be treated as interchangeable names for the same deployable asset.

Check access, running costs and handover before building

Licensing follows the actual implementation and usage. A Microsoft 365 subscription is not blanket permission to use every connector, automation or agent. Copilot Studio has capacity and consumption rules; these depend on the selected experience. Check the current entitlement and expected volume with the administrator rather than publishing a universal “included” claim. Copilot Studio billing

There is a practical cost distinction in the newer experience: Microsoft documents consumption charges from authoring, previewing and testing on the GitHub Copilot harness, rather than only after publication. Include development usage in the estimate. Usage-based billing

Power Automate has its own entitlements. Premium connectors and attended or unattended desktop use require checking the relevant rights. An agent flow uses Copilot Studio capacity. Converting a Power Automate flow to an agent flow changes its billing and is documented as one-way; it does not convert it into the newer workflow format. A new name alone is no reason to migrate. Power Automate licensing FAQ, flow conversion

For each solution, agree:

• Who uses it and which identity accesses the sources or performs actions.
• Which tools, connectors and information IT permits.
• What requires a person's approval, and how missing or incorrect information is handled.
• Where it is developed, tested and deployed.
• Who owns it after handover, monitors failures and pays for its use.

The identity question matters. Copilot Studio tools can use maker-provided credentials instead of the chatting user's credentials. Do not assume every action automatically uses that user's permissions. Administrators can control this choice, and requiring a live user's authentication affects whether background execution is possible. Credential controls

Developing from my office means working in the setup agreed with the company's IT team. It does not mean moving company information into a personal account. Deployment may involve approved configuration and solution transfer, reconfigured connections or a hosted application, depending on what we build.

A practical starting decision

My recommendation is to choose the smallest arrangement that completes the task reliably:

1. Check whether an existing feature already does the work.
2. Use a flow or code when the steps and rules are known.
3. Use an assistant when the employee needs help interpreting, searching or drafting.
4. Add an agent when a repeatable specialist role needs configured knowledge or tools.
5. Combine these where the task requires both flexible interpretation and explicit execution.

These are choices to test against a real task. In a desk-side working session, we can establish what is available and make the first improvement. A larger agent or integration is separately scoped work.