Give it a role. Bring it into your team. Put it to work.
An AI agent is software that can work through assignments, use connected business tools, and collaborate with your people. You give it direction, set its boundaries, and review what it delivers.
Written for business owners and employees who have never used AI at work. Every example on this page is illustrative: a picture of how an agent can be set up, not a ready-made employee.
Meet Alex.
An AI agent can become a working member of your team. Like a new hire, it can carry responsibilities, hold a work identity, use accounts your business gives it, write in a style you choose, work alongside your people, meet deadlines, and be measured on what it delivers.
Where a tool supports it, an agent can have its own company email address and its own business accounts, with access set to fit its role and nothing more. Those accounts are ones you create and control, in the same way you would for a person joining the team; Korps does not create them for you.
The difference from a person is simple to keep in mind: an agent is software. It works fast and consistently, it can make mistakes, and it needs someone responsible for checking its work.
Alex AI agent
Sales research and follow-up
Company email
alex@yourcompany.com
Works with
Sales and marketing
Communication style
Friendly, concise, proactive
Connected accounts
Work chat, email, CRM
Requires approval
External outreach
Illustrative setup. The name, portrait and accounts are examples you would choose; nothing here is provisioned automatically.
Onboarding an AI agent takes four steps.
Define its responsibilities.Give it a clear role and useful work. "Research the accounts we are meeting this week" is a job; "help with sales" is not.
Give it its own work identity.A name, an email address, the supported accounts it needs, and access that matches the role, no wider.
Show it how your team works.Instructions, company procedures, examples of good work, the personality you want, and how it should write to colleagues and customers.
Set goals and check the results.Deadlines, quality expectations, which actions need a person's approval, and how you will measure its work.
Individual assignments
One piece of work with a start and an end: "Prepare a brief for tomorrow's meeting with Northwind." You hand it over, the agent does it, you review the result.
Recurring responsibilities
Work that repeats on a schedule: "Every Monday, check last week's timesheets and list anything missing." Once set, it runs without being asked again, and reports each time.
Your team gets another pair of hands.
When an agent is connected to the places your team already works, it can receive assignments, ask coworkers questions, share drafts, send updates, and hand finished work back for review. Workplace chat such as Microsoft Teams, Google Chat or Slack is where that conversation usually happens; email and shared documents carry the rest.
Which tools an agent can use depends on what you connect and what the agent supports. Slack and Microsoft Teams are Korps connections you can sign in to once; other tools are reached through their own integrations or the browser. See how integrations work.
#sales-teamIllustrative conversation
Jamie Sales
Alex, can you prep me for tomorrow's meeting with Northwind Dental? Coordinate with Sarah in marketing on which case study to bring.
Alex AI agent
On it. I've read the Northwind account notes and the last two calls. Sarah, they run three clinics and asked about online rebooking. Which case study fits best: the multi-clinic one or the rebooking one?
Sarah Marketing
Use the multi-clinic case study; it's the approved version in the Sales Enablement folder, dated August.
Alex AI agent
Thanks. Briefing is ready below. Two facts need confirming before the meeting; I've marked them.
Deliverable
Meeting brief: Northwind Dental, Tuesday 10:00
Who they are
Three clinics, one shared front desk, 14 staff. Current scheduling tool renews in November. Source: account notes, call 12 Aug
What they asked for
Online rebooking from SMS reminders; a plan that covers all three sites. Source: call 12 Aug
Case study to bring
Multi-clinic case study, approved August version. Source: Sarah, Sales Enablement folder
Suggested questions
Do all three clinics share one patient list today? Who owns the renewal decision? What would make rebooking a success for the front desk?
Needs confirmation
Confirm renewal month (notes say November; one email says October). Confirm whether the third clinic opened this year.
Jamie Sales
Great. I'll confirm the renewal month on the call. Taking this in.
Different businesses. Everyday work.
Pick a role to see what an assignment looks like, what the agent does, how your people stay involved, and what lands on your desk. Each one is a configurable example workflow, not a ready-made employee or a guarantee of what any agent can do.
Change the role above; the example changes with it. Each example can be linked to directly.
Give it expectations. Measure what it delivers.
An agent can be held to the same kinds of expectations as a person: work completed, delivered on time, accepted without corrections, response time, how often it had to escalate, and, where you record it, the cost per accepted task. What you can measure depends on what you keep records of; none of it is automatic.
Illustrative weekly review
Alex, sales research: week of 7 September
20 / 20account briefs delivered by the deadline
18 / 20accepted on first review
2returned for corrections
Outcome: 18 briefs ready to use; 2 returned for revision.
Sample figures to show the shape of a review. They are not measured Korps results, and Korps does not ship a metrics dashboard for them today.
Measured results and estimated savings are different things
"18 of 20 briefs accepted" is a measured result: you can count it. "Saved the team six hours" is an estimate: useful for a decision, but it depends on what the people would otherwise have done. Keep the two apart when you judge an agent's work, and say which one you are quoting.
One idea. Different kinds of agents.
Four words that are easy to mix up:
"Sales assistant" is the assigned role: the job you defined.
The agent is the software carrying out the work, step by step.
The AI model is what the agent thinks with: it helps interpret information and decide what to do next. You choose it, and you can change it. Choosing a model.
Connected tools let the agent take supported actions: read a file, draft an email, update a record.
Several agent programs exist. Korps treats each one as a runtime a worker can run on. Five general-purpose agents are in launch scope, Hermes Agent first and recommended. Descriptions come from each project's own site or README on 9 September 2026; "In Korps" says what Korps can say about it today, which is not the same as a logo being on the page.
Coding agents (Claude Code, Cursor, OpenAI Codex, OpenCode) are built for software work and each needs its own access: Claude Code a Claude subscription or Anthropic Console account, Cursor a Cursor account, OpenAI Codex a ChatGPT plan or an OpenAI API key, OpenCode a model provider account of your own. They can be pointed at business tasks, but the general business examples on this page assume a general-purpose agent such as Hermes Agent with the tools you connect. Every agent thinks with an AI provider account you connect.
Your people. Your agents. One coordinated workforce.
Korps is the layer that turns agents into a team you can run. Today it lets you assign work as a message, organise agents into teams under an AI manager who splits and coordinates the work, see progress as states you can read at a glance, receive results and the record of every system a worker touched, and hold any action you choose behind a human approval. Workers run on a machine you pair; the Korps app for Windows and Android opens at launch. How it works.
How does it work like an employee?
It has a role, a manager, teammates and a record. You give it assignments the way you would brief a person, it reports progress, and its work comes back to a person for review. Unlike a person it is software: fast, consistent, and capable of confident mistakes, so oversight stays with you.
Can it have its own email and accounts?
Yes, where the tool supports it and your business creates the account. An agent can have its own mailbox, chat account and CRM login with access limited to its role. You create and control those accounts; Korps does not provision them. Korps never stores the credentials of your AI model provider.
Can it have a personality?
Yes. Its name, tone and communication style are part of its setup, along with examples of how you want it to write. Keep it recognisably an agent: label it as one in chat and email so colleagues and customers always know.
Which actions require approval?
Whatever you decide. A common starting point is that anything sent outside the company, anything that spends money, and anything that deletes or changes records waits for a person. Reading and drafting rarely need approval. How approvals work.
What about sensitive business information?
Give an agent access the way you would a new hire: only what its role needs, through accounts you control and can revoke. Decide which AI model provider may see which data before connecting it, and read that provider's terms. Regulated data, such as health records, needs its own review before an agent touches it; nothing on this page claims compliance for you.
Who would you add to your team?
A sales researcher? A payroll assistant? A helping hand for operations? Picture the first job you'd give your AI agent.
