AI Agents at Work: Vendor Reports by Business Function

An empty open-plan office at dusk with every desk lamp on and one desk glowing warmer than the rest

Ten scored vendor reports on AI agents for business work: agent builders, integration platforms, AI coworkers, and agents for IT, HR, voice, go-to-market and research, plus insurance for agents.

AI agents for business are sold by function now. This hub groups our scored vendor reports by the kind of work the agent does, so you can go straight to the one that matches your problem.

How is the agent market organized?

A year ago most agent products were general platforms. Today the market has sorted into layers. At the bottom are integration platforms that give agents safe access to other software. Above them are builders and coworkers. At the top are agents made for one department, where the vendor has already encoded the workflows.

The reports below follow those layers. Each scores its vendors on four criteria using what the vendors publish about product, price and security.

What does each layer of the market do?

Integration platforms hold each user's credentials, expose app actions as tools an agent can call and keep those connectors working when an app's API changes. Pricing has moved to the tool call: Arcade lists $0.01 a call, Composio $0.005 beyond its allowance and Scalekit $0.50 per 1,000 calls. The deciding question is who the agent acts for. Agents that work for your own employees fit seat-priced tools such as Composio and StackOne. Agents inside a product you sell fit usage-priced tools such as Arcade, Scalekit, Nango and Unified.to.

Agent builders come from three directions. Workflow automation tools added agents (n8n, Zapier, Gumloop). Knowledge assistants grew into builders (Dust, Lindy). Enterprise platforms lead with governance (StackAI, Relevance AI, Microsoft Copilot Studio). Entry plans sit under $40 a month almost everywhere, from n8n at 20 euros to Gumloop at $37.

AI coworkers are agents you message like a colleague in Slack, Teams, email or text. They work out the steps themselves with the tools they are connected to. Tasklet starts at $25 a month for 10,000 credits with no seat fee, while Town and Dust charge per user. Audit status differs: Tasklet, Town and Dust state SOC 2 Type II, and Viktor states Type 1 with Type 2 in progress.

Work platforms that teams already pay for now sell agents on top of the seat. Notion, Asana, monday.com, ClickUp and Airtable all bill AI in a second currency, such as Notion's $10 per 1,000 credits for custom agents or Asana's $0.50 per prepaid AI request. For many teams this is the cheapest first agent, because the data and the permissions are already in place.

Department agents arrive with a function's workflows built in. IT agents carry out the request, such as resetting an MFA device, instead of replying with instructions. HR agents split into employee service (Kinfolk, Leena AI, MeBeBot), hiring (Paradox, Humanly) and compliance (G-P Gia). Voice agents chain speech recognition, a language model and speech synthesis over a phone line. Go-to-market platforms join account data, buying signals and agents that act on them. Research platforms either interview real people with an AI moderator or simulate respondents.

Underneath all of it sits liability. Standard business policies were written before generative AI, and insurers such as Armilla AI now write cover for AI errors and agent actions, up to $25 million aggregate per insured, underwritten at Lloyd's.

The reports in this cluster

How is agent work priced?

Almost every vendor in this cluster charges in two currencies: a seat or platform fee, and a usage meter. The meter goes by different names, such as executions at n8n, credits at Tasklet and Notion, tool calls at Arcade, minutes for voice agents and records enriched for go-to-market platforms. The monthly fee is usually the smaller number.

Before comparing plans, run one real task end to end on a trial and note what it consumed. Multiply by the volume you expect. That cost per finished task is the only figure that compares across vendors. Voice needs extra care, because an advertised per-minute rate often leaves out the model, the voice and the phone carrier. Retell AI's published range of $0.07 to $0.31 a minute shows how wide the spread gets once those parts are added.

What should a buyer check before signing?

  • Permissions. The agent should act only with the rights of the person it works for, and an admin should be able to limit which actions it may take.
  • Audit trail. Every action should be logged with its input, the tool call and the result, and the log should be exportable.
  • Human approval. You should be able to require sign-off for steps that move money, change access or contact customers.
  • Data use. Ask whether your data trains any model and where it is stored.
  • Exit. Check that agent definitions and connector settings can be exported if you leave.

Vendors that answer these points in public documentation scored higher on control in our reports. Vendors that answer only on a sales call scored lower on buying ease, because a buyer cannot check the claim before the conversation.

What is changing next?

Three shifts show up in what vendors published this autumn. Department agents are absorbing the builder layer: IT and HR vendors now let customers add their own workflows, which narrows the case for a separate builder. Usage pricing is spreading to the work platforms, so seat budgets no longer predict AI spend. Assurance is becoming a buying criterion: AIUC pairs insurance with its AIUC-1 standard, and buyers in regulated industries are starting to ask for a certificate alongside a SOC 2 report.

We update each report when vendors change prices or plans. This hub always lists every report in the cluster.

How to use these reports

  1. Start from your problem. Pick the report closest to the work you want done, not the vendor you have heard of.
  2. Read the criteria first. Each report scores four things from 1 to 5. Check they are the four you care about.
  3. Treat scores as a shortlist. They reflect what vendors publish. Run your own trial before you buy.
  4. Check the date. Prices and plans change monthly. Each report says when its sources were read.

What is the difference between an agent builder, an AI coworker and a department agent?

A builder is a tool for assembling your own agents from steps and integrations. A coworker is a ready agent you message in chat or email that works out the steps itself. A department agent is built for one function, such as IT help desk or HR, and arrives with that function's workflows.

Where should a company start with AI agents?

Start with one repeated task that has a clear owner and a measurable result, such as resetting passwords or answering HR policy questions. Choose a department agent if one exists for that task, and a builder or coworker if it does not.

How much does an AI agent cost?

Entry plans for agent builders and coworkers start between $20 and $40 a month, but usage decides the bill. Vendors meter executions, credits, tasks, tool calls or minutes. Run one real task on a trial, note what it consumed and multiply by your expected volume to get a cost per finished task.

What should an AI agent pilot measure?

Measure the share of requests the agent completes without a person, the error rate on the ones it completes, the time saved per request and the cost per finished task. Agree the targets before the pilot starts, and review a sample of completed work by hand every week.

Explore More

Ready to Find the Right AI Tools?

Browse our data-driven rankings to find the best AI tools for your team.