Airtable AI in 2026: Omni, Field Agents, and Automations

What is Airtable AI Today?

In 2026, AI is integrated into several areas of the Airtable platform.

This distinction is important.

For example, lead qualification may require AI if it involves understanding the company's website or a free-text description.

Conversely:

If the amount is greater than €10,000, change the status to "Approval required".

requires no AI model. A classic formula or automation will be simpler, less costly, and above all, more predictable.

The right reflex remains:

business need → process → architecture → AI only if it adds value.

Omni: Build and Use Airtable in Natural Language

Omni is the AI assistant built directly into Airtable.

You can describe what you want to build or ask questions about the existing system. Depending on your permissions, Omni can create or modify tables and fields, work on records, analyze documents, search for information on the web, and help create Interfaces or automations.

For example, you could ask it to:

Create a Contracts table linked to Clients, with a start date, a renewal date, an owner, and a status.

Or:

Which active projects are delayed by more than a week?

Or even:

Create an Interface for project managers showing their active projects, overdue tasks, and pending approvals.

Omni primarily speeds up "how to build"

Creating a table, adding multiple fields, or producing a first draft of an Interface can be much faster than before.

But this ease of use can also encourage adding new tables, statuses, or automations too quickly.

The problem is gradually shifting from:

How do I build this in Airtable?

to:

What should actually be built?

Omni can easily create a Contracts table. However, you still need to determine if contracts really deserve a separate entity or if a few fields in Clients would suffice.

If each client has multiple contracts with their own amounts, services, and deadlines, a dedicated table is probably justified.

If there is systematically only one very simple contract per client, it may on the contrary make the system unnecessarily complex.

AI speeds up the execution of an architecture. It does not guarantee that the architecture is good.

Interfaces also generated with AI

Omni can also help build Interfaces, and Airtable now offers Interface elements generated with AI.

This is useful for quickly producing a first version of a dashboard or workspace.

But a good Interface is not about displaying the maximum amount of available information. It should allow the user to quickly understand:

• what requires their attention;

• the information useful for their task;

• the actions they can take;

• what they do not need to see.

Our Airtable Interfaces guide explains this user layer in detail and how to decouple it from the data model.

Field Agents: Putting AI to work directly on your records

AI features at the field level are now grouped by Airtable under Field Agents.

A Field Agent can search, analyze, or generate information based on the context available for a record.

Unlike a classic formula, it can work with unstructured data and use an AI model to produce a result.

This makes it possible to:

• analyze a document;

• summarize notes;

• classify a request;

• search for information on the Internet;

• enrich company data;

• extract structured data;

• generate content.

The result is then directly available in the record.

Airtable details the current features and possibilities of Field Agents in its official documentation.

Example: Qualifying an incoming lead

Imagine a Leads table containing:

• company;

• website;

• size;

• description of need;

• source;

• status.

A Field Agent could analyze the company and its website to identify its industry and suggest a qualification.

The workflow would look like this:

New lead

→ information search
→ analysis
→ industry identified
→ suggested qualification
→ sales rep reviews
→ next steps in process.

The value of AI here is real because qualification depends on information that is not perfectly structured at the start.

On the other hand, if the criterion is simply:

Number of employees greater than 20.

a formula is enough.

Example: Analyzing incoming documents

Another interesting case: documents arrive in Airtable as attachments.

These could be quotes, invoices, briefs, contracts, or job applications.

A Field Agent can analyze the document and extract specific information. For a vendor quote, we could extract:

• vendor;

• amount;

• date;

• reference;

• type of service.

This data can then be verified before powering the rest of the process.

This results in an architecture that is generally more robust than:

document → AI → all actions executed immediately.

Example: Classifying a request

Take a table of customer requests where each request contains free text.

A Field Agent can determine:

• the topic;

• the likely priority;

• the relevant team;

• a summary;

• possibly a draft response.

The AI then enriches the record.

An automation can then use this information to run the workflow.

This is precisely the major difference between a Field Agent and Automation.

Automations + AI: When intelligence is part of a workflow

Airtable automations execute actions when a specific event occurs.

For example:

Form submitted

→ create the record
→ assign an owner
→ create a task
→ send a Slack notification.

No AI is needed if all the rules are defined.

However, Airtable allows you to add an AI generation action within an automation.

You can then build:

New customer request

→ analyze the request
→ produce a structured output
→ determine the category
→ update the record
→ notify the right team.

The AI output can, for example, take the following form:

• Category = Billing

• Priority = High

• Escalate = Yes

The subsequent steps then become deterministic again.

Field Agent or Automation with AI?

Both can sometimes produce similar results.

Prefer using a Field Agent when the AI's output is information that naturally belongs to the record:

category, summary, score, extracted data, enriched company info.

Prefer using an Automation with AI when the analysis is a step within a sequence of actions:

event → analysis → structured decision → email → Slack → task.

In some systems, both work together.

The Field Agent enriches the data.

The Automation uses this data to move the process forward.

Omni, Field Agent, Automation, or MCP: Which one to choose?

These different building blocks are not in competition with each other.

MCP deserves a special distinction.

With Omni, the user asks for something from inside Airtable.

With MCP, an external assistant like ChatGPT or Claude can work with Airtable.

This allows you to ask, for example:

Which active projects are over budget and still have blocked tasks?

The assistant can determine which Airtable data to query to provide the answer.

This behavior is very different from an automation that must run the exact same sequence every time. Our Airtable MCP guide details this architecture and its limitations.

And when an application simply needs to read or write to Airtable deterministically, the Airtable API often remains the best choice.

How much does AI cost in Airtable?

Airtable operates with pooled AI credits at the workspace or organization level, depending on the plan.

Currently, Airtable includes monthly:

Usage depends on the action, the volume of data processed, the model used, and the size of inputs and outputs.

A short classification will therefore not consume as much as analyzing long documents across thousands of records.

Also note: building apps and agents with Omni does not currently consume AI credits. Using them, however, may.

The cost must therefore be evaluated at the workflow level.

Running an agent occasionally on 50 opportunities does not have the same impact as triggering automated analysis on 50,000 records.

Airtable maintains quotas and consumption rules in its documentation on AI credits.

For other platform limits and plans, see our Airtable pricing guide.

Limitations of AI in Airtable to Keep in Mind

Native integration reduces the need to connect external AI tools for many workflows.

However, it does not remove the typical limitations of AI models.

An AI output is not a business rule

A formula will always respond deterministically to:

If the due date has passed and the status is not Complete, then the project is overdue.

An AI model should not replace this logic.

AI becomes interesting when interpretation is needed:

Does this customer request seem urgent?

These two types of problems should not be handled the same way.

You must plan for errors

An agent might misinterpret a document, produce a wrong classification, or generate an inaccurate response.

The more critical the subsequent action, the more important control becomes.

Depending on the context, the right architecture might be:

AI suggests → human approves → system executes

rather than:

AI decides → system executes immediately.

An error in an internal summary does not have the same consequences as a bad decision triggering billing or modifying a contract.

More AI can also mean more complexity

A workflow made of five agents passing answers to each other might look advanced.

But if it replaces three simple rules, it is probably inferior.

It becomes harder to understand, test, maintain, explain to teams, and budget for.

The goal should not be to maximize the number of agents, but to reduce manual work without making the system more fragile.

Accessible data must be controlled

The data required for AI features is processed under the services and models configured by Airtable.

Permissions, sensitive information accessible to agents, and data included in prompts must therefore be carefully designed from the start.

Giving "all available context" to an agent is not necessarily a good practice. Restricting context to the data actually needed is often easier to control.

How to Intelligently Integrate AI into an Airtable System

The best starting point is generally not:

Where can we add AI?

But:

Which task currently takes time because it requires reading, understanding, searching, or interpreting information?

Then:

1. Verify if AI is actually necessary

If the rules can be clearly expressed with Airtable fields, formulas, or conditions, it is generally better to remain deterministic.

If the task requires understanding natural language, analyzing a document, or searching for information, AI becomes more relevant.

2. Choose where intelligence should step in

Should the output become data in the record?

→ Field Agent

Does the analysis occur in an automated sequence?

→ Automation + AI

Does the user want to occasionally ask a question in natural language?

→ Omni or MCP

3. Test on a limited scope

Before rolling it out broadly, monitor the quality of results, edge cases, errors, and credit consumption.

A workflow that performs well on 30 examples is not necessarily ready to run uncontrolled on thousands of records.

4. Keep human validation where it adds value

The goal is not necessarily to remove all human intervention.

Going from:

20 minutes of manual work

to:

2 minutes of validation

is already a highly successful automation.

This is also the approach we follow when we need to automate a process and integrate AI: simplify the workflow first, and then use AI only in places where it brings concrete value.

FAQ on Airtable and AI