How to create an AI agent for your business (without being a programmer)

01 What you need before you start

Before you begin, identify the process, available data, permissions and a responsible owner.

A process that already exists. You can't automate a decision nobody has made yet. If two people on your team handle the same task differently, an agent will make that inconsistency faster, not better. Write down how the process runs today before you build anything.

Data the agent can reach. This is the requirement that kills most projects. An agent that can't query your orders, your invoices or your customer history has nothing to reason about. If your data lives in a spreadsheet on someone's laptop, that's step zero.

A decision about permissions. What can the agent read? What can it change? What must a human approve before it happens? Answer this before building, not after the first incident.

Someone who owns it. Not a developer, an owner. Someone who reviews what the agent did during the first weeks and decides whether it's working. Agents that nobody reviews get quietly ignored, which is the most expensive outcome of all.

​

When technical help may be needed

Some document and workflow tools let teams configure a narrow task without programming. Connecting systems, managing permissions or automating a complex process may need technical support.

If you're new to the category itself, it's worth understanding what agentic AI is and how it differs from the generative AI you already use before choosing a tool. The distinction determines what you should be building.

How to create an AI agent step by step

Work through these seven steps in order, beginning with a task whose output someone can review.

Use the list as a practical checklist for a first agent:

1. Define the task. Choose one repeatable task with a clear trigger and a result a person can check. Write down the input, expected output and cases that should be escalated.
   
2. Choose the tool and model. Compare tools against the data you need, the actions allowed and the way your team will review results. Pick a model that handles representative examples, rather than choosing on name alone.
   
3. Connect data and tools. Identify the source records or documents the agent needs and confirm they are current and accessible. Start with read access and check what each connection actually exposes.
   
4. Write instructions and boundaries. Specify the task, output format, source of truth and when the agent must stop or ask for help. Include examples of ambiguous or incomplete inputs.
5. Set permissions and approvals. Give the agent only the access needed for its task and identify actions requiring human approval. Keep sending, posting, purchasing and record changes behind an explicit review.
6. Test representative cases. Try normal, messy and edge cases from the real workflow, including situations where the agent should refuse or escalate. Review outputs with the process owner before allowing any consequential action.
7. Measure, improve and expand. Track useful drafts, incorrect outputs and escalations, then adjust instructions, data or scope. Expand to another task only after the owner has reviewed performance and the required approvals.

Where to create AI agents

The right environment depends on where the relevant data lives, what actions are needed and who will maintain the setup. Compare these three common approaches against your defined task.

Document-based tools

Document-based tools let you ask questions over manuals, policies and other files, and can help retrieve or summarise information. Check how documents are updated and whether a static file is suitable for a task that depends on current records.

Automation platforms

Automation platforms can link workflow steps with a model and connected applications when those connections are available and authorised. Map the data passed between steps, the actions each step may take and the person responsible for reviewing changes.

Flows and connections need an owner who can update them when processes, permissions or application interfaces change. Include monitoring and maintenance in the decision, especially when several systems are involved.

Agents within business software

An agent embedded in business software may be able to work with records and permissions in that system. Check the specific product, available actions, audit trail and approval controls before relying on it; some workflows still need configuration or integration.

This is also where the MCP standard matters. Model Context Protocol is an open specification that lets AI models query a business system in a structured way, and it's what makes it possible to connect ChatGPT or Claude to your own ERP data rather than pasting exports into a chat window.

How much does it cost to create an AI agent?

Cost depends on the task, the state of your data and the systems the agent must use. Estimate the work to configure, test and maintain the complete workflow, including human review.

Configuration and data

Allow for defining the task, writing instructions, preparing source data and setting permissions. Incomplete or inconsistent data adds work before the agent can be tested reliably.

Connections, usage and testing

The estimate should include any approved connections, model usage and tests with representative cases. Check how usage is billed by the chosen tools and who will review the outputs.

• The messy ones, not just the clean ones
• The edge cases your team already argues about internally
• At least three cases where the agent should refuse and escalate

Include normal, messy and refusal cases so the testing effort reflects the real workflow.

Maintenance and ownership

• Right: the agent produced the correct outcome
• Wrong: it produced a confident, incorrect outcome
• Didn't know: it stopped and escalated

Budget for monitoring, correcting instructions or data, updating connections and reviewing the results as the business process changes. Assign an owner to decide when another change is needed.

Estimate the full workflow

A read-only drafting task generally has a different implementation and review burden from a workflow that can send, post, buy or change records. Scope permissions and approval steps before comparing options.

Include any logging, review and escalation requirements in the estimate so the owner can investigate outcomes and improve the process.

Expect to refine the scope, instructions and data after testing. Maintenance needs depend on the systems involved and how often the workflow changes.

Common mistakes when building your first agent

Five mistakes to check

Review these risks before granting an agent access to business data or actions:

• Giving the agent a task that is too broad to define or review.
• Testing only easy examples and missing ambiguous or incomplete cases.
• Granting more permissions than the task requires.
• Letting consequential outputs or actions proceed without human review.
• Failing to assign an owner for monitoring and changes.

Keep the first task narrow enough for the owner to check each result and correct the workflow.

Practical safeguards

Put the following controls in place before expanding the agent:

• Document the task, its boundaries and the source of truth.
• Test representative inputs, errors and escalation cases.
• Restrict access and require approval for consequential actions.
• Assign an owner to review outputs and maintain the workflow.

Revisit these controls when the process, data or connected tools change.

Choose a manageable first task

Compare tools against your task, source data, permissions and support needs. An embedded agent or an automation platform may reduce some setup work, but the exact capabilities and ongoing maintenance depend on the implementation.

Start with read-only research, extraction, summaries or drafts. Keep a person responsible for checking outputs and approving actions that affect customers, finances or records.

If the task needs several systems or complex decisions, involve technical and process owners before expanding access or automation.

Use the cost factors above to compare the complete workflow, then review the chosen platform’s documented capabilities and support terms.