AI Pulse Check
← Back to blog

AI Agents for Business: 11 Options and What They Cost (2026)

By Ethan Halfhide · Updated October 12, 2026

An AI agent for business is software that takes a goal, decides the steps itself and acts in your apps, such as updating a record, answering a customer or sending a report, within limits you set. For most small businesses the practical start is an agent feature inside a tool you already pay for.

This guide explains how an agent differs from a chatbot or a fixed automation, compares 11 agent products a small business can buy today with the prices their makers publish (checked October 12, 2026), shows what a small team would pay, and covers the risks and controls to set up first.

AI agent vs chatbot vs automation

Anthropic, which makes Claude, draws the line this way in its guide to building agents: workflows are "systems where LLMs and tools are orchestrated through predefined code paths", while agents are "systems where LLMs dynamically direct their own processes and tool usage, maintaining control over how they accomplish tasks."

ChatbotFixed automationAI agent
Who decides the stepsYou write the answers or the AI answers from your contentYou define every step in advanceThe AI picks the next step toward a goal you set
Can it act in your apps?Usually only answers questionsYes, exactly the actions you set upYes, within the apps and permissions you connect
PredictabilityHigh for scripted bots, lower for AI answersHigh: same input, same resultLower: Zapier's own FAQ says agents can give different outcomes on the same task because LLMs are non-deterministic
Good forAnswering repeat questionsMoving data the same way every timeMessy, multi-step work where the path changes each time

The line between the three is blurring. Make says its agents "work alongside deterministic logic, not instead of it", and HubSpot's agent builder puts agents and workflows on one canvas. Anthropic's advice is to find "the simplest solution possible", which "might mean not building agentic systems at all". If a fixed workflow does the job, our list of the best AI automation tools is the better place to start.

AI agents for business compared (as of October 2026)

ProductWhat the agent doesPublished price (maker's own page)Free option on that page
Microsoft 365 Copilot BusinessPre-built Researcher, Analyst and Facilitator agents; build your own with Agent Builder and Copilot Studio$18 per user/month paid yearly, a sale price (was $21). Needs a qualifying Microsoft 365 Business plan"Try for free" link with trial terms
Microsoft Copilot StudioBuild agents, including ones published to websites, apps and other outside channels$200 per pack of 25,000 Copilot Credits per month, or pay-as-you-go. Azure subscription requiredFree Azure account with a $200 credit
Gemini Enterprise Business edition (Google)Pre-built Deep Research agent; no-code Workflow Builder for custom agents; connectors to Microsoft 365, Google Workspace, HubSpot and NotionStarting at $21 per seat/month30-day free trial
ChatGPT Business (OpenAI)Workspace agents that run on schedules and act in tools such as Slack, Google Drive and SharePoint (research preview)$20 per user/month billed annually, $25 billed monthly, 2 or more usersSeparate free personal plan
Zapier AgentsAgents that use your company knowledge and act across 9,000+ connected appsPro: $33.33/month billed annually ($400 a year) for 1,500 activitiesFree: 400 activities a month
Make AI AgentsAgents that choose the next step and trigger Make workflows across 3,000+ appsCore: $12/month for 10,000 credits. Pro: $21/month for 10,000 credits (billing period not stated on the page)Free: 1,000 credits a month
LindyAn AI teammate for inbox, meetings, CRM updates, support triage and reportsTeam: from $49.99 per active user/month for 5,000 credits$50 in free credits, lasting 7 days
HubSpot customer agentAnswers support questions, takes action in the HubSpot CRM, hands off to your team$0.50 per resolution (chat and messaging), $0.50 per minute (voice). Needs HubSpot Professional or Enterprise28-day free trial
Intercom FinAnswers and resolves support conversations by chat, email, WhatsApp, SMS and more, on Intercom or another helpdesk$0.99 per outcome. On another helpdesk, a 50-outcome monthly minimum. With Intercom, plus helpdesk seats (from $19 per seat/month billed annually on Intercom's Essential plan)14-day free trial, no credit card
Salesforce AgentforceCustomer-facing and employee-facing agents on Salesforce$2 per conversation, or Flex Credits at $500 per 100,000 credits (20 credits per action)Salesforce Foundations: $0 to get started
Claude (Anthropic)General assistant; the Pro plan lists "Hand off and schedule tasks" and the Team plan lists Claude CoworkTeam: $20 per seat/month billed annually, $25 billed monthly, teams of 2 to 150Separate free personal plan

Two products in this space changed recently. OpenAI's help article on ChatGPT agent now says "ChatGPT agent is no longer available" and points users to ChatGPT Work, which OpenAI's business pricing page lists on both Business and Enterprise. And Microsoft says its $18 Copilot Business price is a discount offer running from July 1 to December 31, 2026, for the first year only, with an annual commitment.

Each option: who it's for and the trade-offs

Agents inside Microsoft 365 or Google

Microsoft 365 Copilot Business suits a business that already runs on Outlook, Word, Excel and Teams. Microsoft lists pre-built Researcher, Analyst and Facilitator agents, and its plan comparison lists agent creation with Agent Builder and Copilot Studio. The trade-off: it is an add-on, available only to existing Microsoft 365 customers with an eligible Business plan, for up to 300 users.

Microsoft Copilot Studio is the step up. Microsoft says Copilot "includes agent-building capabilities for use within your organization", but "to publish or share agents to outside channels, choose a standalone Copilot Studio plan". Pricing is usage-based: credits are used "whenever Copilot or an agent performs a task or generates a response", and the number billed varies by task, so your monthly bill depends on volume. Microsoft says an Azure subscription is required to use agents.

Gemini Enterprise Business edition is Google's version for "individuals, small businesses, and teams up to 300", with no IT setup needed. It includes a pre-built Deep Research agent and Workflow Builder for "building custom, no-code agents using natural language to automate repeatable tasks", and Google says "we don't train on" your data. The trade-off is daily quotas: Google's help page lists 3 Deep Research queries per user per day and 1 agent built per user per day with the no-code tools, pooled across the team. Google also says some capabilities on its Gemini page, including delegating multi-step background tasks, "are currently in early access". Separately, the Google Workspace pricing table lists Workspace Studio, described as "Automate your workday with intelligent AI agents", on its Starter, Standard, Plus and Enterprise plans. The Workspace page did not show plan prices to us, so none appear here.

General assistants with agent features

ChatGPT Business fits a team that already uses ChatGPT. OpenAI says workspace agents "take on entire workflows on their own" and can run on schedules for jobs such as reviewing leads. Its page says human approval is required before sensitive actions such as sending messages or updating records. Admins choose who builds agents and which tools they can use, and audit logs track activity. The catch: OpenAI calls workspace agents a research preview, and the page we read lists no usage limits or credit costs for them. OpenAI's business pricing page says "No training on your business data by default". Our guide to ChatGPT for small business covers which plan to pick.

Claude is the comparable option from Anthropic. The Team plan costs $20 per seat a month billed annually and lists Claude Code and Claude Cowork among its features. The pricing page did not describe what agent tasks each plan can run, so we have not compared them line by line.

Agent builders that connect your apps

Zapier Agents suits a business that already automates with Zapier. Zapier says agents "can only take actions in the apps you've connected and use the triggers and actions you've set up". An activity, its billing unit, includes "actions your agent takes in behaviors or in chat, browsing the web, or looking up information from attached knowledge". The free plan's 400 activities a month is enough to try one agent on real work. Zapier also says agents do not currently support the app and action restrictions on a Zapier Enterprise account.

Make AI Agents are available on all Make plans, and Make says you "can set clear rules, add manual approvals, or stop the Agent at specific points" and see every decision step by step in a Reasoning panel. Credits are the billing unit, and Make says some features using its AI provider "may use more" than the usual one credit per action, so test before estimating a monthly cost. If you are choosing between the two platforms, see our zapier vs make comparison.

Lindy sells itself as an AI teammate rather than a workflow tool. Its pricing page says anything "with outside impact waits for approval such as sending an email, updating a ticket, posting to another channel, publishing a doc." Credits are pooled across the team, an everyday task uses 2 to 250 credits and a deep research job 250 to 1,000, and Lindy says it pauses when the pool runs low rather than billing more. Every active user is billed, from $49.99 a month for 5,000 credits each.

Customer service agents

HubSpot customer agent is for businesses already on HubSpot Professional or Enterprise. HubSpot counts a resolution when "the conversation was not handed to a human rep for 72 hours". HubSpot Credits cost $0.01 each, sold in packs of 1,000 for $10 or as pay-as-you-go. HubSpot's agent builder is included in Starter, Professional and Enterprise, but custom agents use credits each time they run. Our directory of AI CRM tools compares HubSpot with other CRMs.

Intercom Fin works with Intercom or with another helpdesk; its pricing page names Salesforce, HubSpot, Freshworks, Front, Zoho and Gorgias among them. You are charged for at most one outcome per conversation. A resolution, a handoff through a Procedure you configured, a disqualified prospect or a self-serve routing costs $0.99, and a qualified lead costs $9.99. Intercom says you are "not charged when a conversation is simply passed to your team without an outcome".

Salesforce Agentforce is for businesses that run on Salesforce. Each Agentforce action costs 20 Flex Credits, which works out to $0.10 at $500 per 100,000 credits. Salesforce's own example: an order-status agent handling 20 requests a day, with 2 actions each, costs about $120 a month. Salesforce's Starter Suite CRM costs $25 per user per month.

For a broader list of support bots, including cheaper ones that only answer questions, see our ranking of the best AI chatbots for business. If most customer contact comes by phone, compare an AI receptionist for small business instead.

What it costs a small team

These are our own sums from the list prices above, as of October 2026, before tax. They leave out the base subscriptions some agents need.

SetupMonthly costHow we got there
Microsoft 365 Copilot Business, 5 users$90, paid yearly (first-year sale price)5 x $18, on top of a Microsoft 365 Business plan
ChatGPT Business, 5 users$100 billed annually, or $125 billed monthly5 x $20, or 5 x $25
Gemini Enterprise Business edition, 5 usersFrom $1055 x $21
Zapier Agents Pro$33.33, billed annually as $4001,500 activities a month
Support agent resolving 300 conversations a month: HubSpot customer agent$150 in credits, plus a HubSpot Professional subscription300 x $0.50
Same 300 conversations: Intercom Fin$297, plus helpdesk seats if you use Intercom300 x $0.99
Same 300 conversations: Agentforce on per-conversation pricing$600300 x $2

The support figures are not like for like. HubSpot charges per resolution, Fin per outcome and Agentforce per conversation, and each defines its unit differently, so read each vendor's definition before comparing.

Where AI agents fit in a small business

  • Customer questions. The most mature use. HubSpot, Intercom and Salesforce all sell agents that answer from your content and hand off to a person.
  • Research and reports. Microsoft's Researcher and Analyst agents and Google's Deep Research agent are built for multi-step research. OpenAI describes agents that research vendors and produce a structured report for review.
  • Sales follow-up. Zapier lists a lead qualification agent and a meeting prep agent among its templates. HubSpot's announcement of its prospecting agent prices it at $1 per lead recommended for outreach, paid in credits, on Professional and Enterprise.
  • Inbox and admin. Lindy lists inbox management, meeting prep and follow-up, and drafting replies.
  • Recurring internal jobs. ChatGPT workspace agents and Lindy both run on schedules, for example a weekly summary of support requests.

Agents fit work that changes shape each time. A task that runs the same way every time is usually cheaper and more reliable as a fixed workflow.

Risks and the controls to set first

Agents can do the wrong thing confidently. Anthropic says "the autonomous nature of agents means higher costs, and the potential for compounding errors" and recommends "extensive testing in sandboxed environments, along with the appropriate guardrails."

Too much access is the main security risk. The OWASP GenAI Security Project calls it "Excessive Agency": an agent performs damaging actions "in response to unexpected, ambiguous or manipulated outputs", with root causes of "excessive functionality; excessive permissions; excessive autonomy." Its example: an agent that only needs to summarize emails should not also be able to delete or send them.

Content the agent reads can try to steer it. OpenAI's ChatGPT agent help article names "prompt injection" attacks as a privacy risk and says its safeguards "don't eliminate all risks."

Many projects fail on cost and value, not technology. Gartner predicted in June 2025 that "over 40% of agentic AI projects will be canceled by the end of 2027", citing "escalating costs, unclear business value or inadequate risk controls." It also estimated that only about 130 of the thousands of vendors calling their products agentic are real, and called the rebranding of chatbots and older tools "agent washing".

The controls to set up before an agent touches customers or money:

  1. Give it the least access that works. Read-only where possible, and only the apps the task needs (OWASP's first mitigations).
  2. Require approval for anything with outside impact. Lindy, Make, OpenAI's workspace agents and OWASP all describe human approval for sending messages, updating records or publishing.
  3. Test before it goes live. HubSpot's agent builder lets you "test your agent before you turn it live". Use made-up or old data first.
  4. Set spending limits. Zapier, Lindy, Intercom, HubSpot and Microsoft all bill by usage. Intercom and HubSpot describe usage alerts and limits; use them.
  5. Read the logs weekly. Make's Reasoning panel, OpenAI's audit logs and the transcript views in support tools show what the agent actually did.

How to start

  1. Pick one job. Choose a task that eats hours every week and changes shape each time, such as triaging support email or preparing meeting notes on new leads.
  2. Check what you already pay for. If you run Microsoft 365, Google, HubSpot, Salesforce, Zapier or ChatGPT Business, look at the agent features on your current plan before adding a new subscription.
  3. Use the free tier or trial. Zapier Agents has a free plan, Make and Lindy offer free credits, Gemini Enterprise and Intercom Fin have trials, and HubSpot offers a 28-day trial of its customer and prospecting agents.
  4. Start with approval on. Let the agent draft and propose for the first few weeks, and approve every outside action yourself.
  5. Measure it against the old way. Count the hours saved and the errors caught for a month, then decide whether to widen its access or switch it off.

If you are still deciding where AI fits at all, our list of the best AI tools for small business covers the wider stack.

What we don't know

  • Real-world accuracy. We have not run these agents. Resolution rates and time savings on vendor pages are the vendors' own figures, so we left them out.
  • Monthly bills for usage-based plans. Copilot Studio, Make, Lindy, HubSpot custom agents and Agentforce vary credits by task, so the cost of a given job is only clear after testing.
  • Workspace agent limits. OpenAI's workspace agents page lists no usage limits or credit costs, and the feature is in research preview.
  • Make's billing period. Make's pricing page shows Core at $12 and Pro at $21 a month for 10,000 credits next to a monthly or annual toggle, but does not say which billing period those figures are for.
  • Google Workspace prices. The Workspace pricing page listed Workspace Studio on every plan but did not show plan prices to us.
  • Agentforce on the Starter Suite. Salesforce's pages do not say which Agentforce features come with the $25 Starter Suite.
  • Claude's agent features by plan. The pricing page lists features but does not describe what each can automate.

FAQ

What is an AI agent for business?

It is software that takes a goal, decides the steps on its own and acts in your business apps, such as updating a CRM record, answering a customer or producing a report. A fixed automation follows steps you set in advance, and a chatbot mainly answers questions. Most business agents work within the apps and permissions you connect, and many can be set to ask for approval before acting.

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

A chatbot answers questions. An agent can also take actions, such as updating a ticket, booking a meeting or sending a message, and decides which steps to take. Many customer service products now do both: HubSpot's customer agent and Intercom Fin answer questions and can update records or hand off to a person.

How much do AI agents cost?

As of October 2026, prices on the makers' own pages range from free tiers (Zapier Agents, 400 activities a month) to per-seat plans such as Microsoft 365 Copilot Business at $18 per user a month paid yearly (a first-year sale price) and Gemini Enterprise Business edition from $21 per seat a month. Customer service agents charge per result: $0.50 per resolution for HubSpot, $0.99 per outcome for Intercom Fin and $2 per conversation for Agentforce.

Is there a free AI agent for small business?

Yes, with limits. Zapier Agents has a free plan with 400 activities a month, Make's free plan includes 1,000 credits a month and covers its AI agents, and Lindy gives $50 in free credits for 7 days. Gemini Enterprise Business edition has a 30-day free trial and Intercom Fin a 14-day trial.

Which companies make the main AI agents for business?

The large platforms are Microsoft (Copilot and Copilot Studio), Google (Gemini Enterprise), OpenAI (ChatGPT workspace agents), Salesforce (Agentforce) and HubSpot (its customer, prospecting and data agents). Zapier, Make and Lindy sell agent builders that connect to many apps, and Intercom sells Fin for customer service.

Are AI agents safe to use with customer data?

They can be, with controls. Give the agent the least access it needs, require human approval for actions such as sending messages or updating records, and check the vendor's data terms. Several vendors state they do not train on business data by default, including OpenAI for ChatGPT Business and Google for Gemini Enterprise Business edition. Security groups such as OWASP warn that agents with too many permissions can take damaging actions, so start narrow.

How this page was put together

We read each vendor's own product and pricing pages on October 12, 2026, along with Anthropic's guide to building agents, the OWASP entry on Excessive Agency and Gartner's June 2025 press release. Some vendor pages would not load in our page reader (OpenAI, Make and Gartner), so we read them through a second fetch tool on the same day. We did not run these products. Tool mentions rest on published prices and each vendor's own description, and the cost table is our own arithmetic from list prices. Prices change often, so check each pricing page before you buy.

Sources