Four requests come up in almost every IT channel in Slack or Microsoft Teams: someone can’t log in, someone needs access to an app, someone’s laptop isn’t working, and someone new starts on Monday. None of them are new. What has changed is how much of each one a person has to do.
Each request below is told three ways: how it gets handled by hand, how it gets handled with the best tools IT had before AI, and how it gets handled with an AI agent sitting in the channel. The pre-AI tools don’t go away in the third version. The agent uses them. What goes away is the part where a technician reads the message, looks things up and types the first reply.
| Request | By hand | Best tools before AI | With an AI agent |
|---|---|---|---|
| “I can’t log in” | A technician resets it in the admin console | Self-service reset in Entra ID or Okta, if the person finds it | The agent answers with your reset steps, and hands identity checks to a technician |
| “Can I get access to X?” | A DM to the manager, then a grant by hand | Access requests in your identity provider, for the apps it covers | The agent asks the named approver and grants through your identity provider on a yes |
| “My laptop isn’t working” | Twenty messages to collect the basics | Device details from Intune or Jamf, read by a technician | The agent pulls the device, drafts the first fix from your article |
| “X starts next Monday, can we set up?” | A checklist in someone’s head | SCIM provisioning and zero-touch laptops, for the standard set | The agent runs your onboarding checklist and chases the non-standard parts |
Request 01 · Identity
“I can’t log in”
Most lockouts are a forgotten or expired password, or a new phone without the authenticator app. The request is simple. The work around it isn’t: it usually arrives five minutes before a meeting, and the only safe way to help involves checking that the person is who they say they are.
By hand
A technician sees the message, asks which system, resets the password in the admin console and sends a temporary one by direct message. Because the person is in a hurry, the identity check is the step under the most pressure to skip, and the technician is the one holding the line. Each reset costs a few minutes of a technician’s time and carries a small risk, and the ticket gets written afterwards, if there is time.
Best tools before AI
Self-service password reset in Microsoft Entra ID or Okta, with two registered methods so people can always recover. A Temporary Access Pass in Entra ID for a lost or replaced phone, so the person can sign in once and register a new method instead of MFA being switched off. A fallback that doesn’t need single sign-on, such as a phone number, because someone locked out of SSO may be locked out of Slack or Teams too.
This removes the reset itself for anyone who knows the link exists. Plenty of people don’t, so they still ask in the channel, a technician still replies with the link, and the MFA cases are still handled one at a time with an identity check over a call.
With an AI agent
The agent answers the message in seconds with the reset link and your documented steps for whichever system it is, taken from your article, with the source attached. Its reply says up front that IT will never ask for a code or a password in the channel. If the request is an MFA reset or anything else that needs an identity check, the agent opens a ticket, notes which system and what the person has tried, and hands it to a technician, who does the check and replies in the same thread.
Request 02 · Access
“Can I get access to X?”
Access requests pile up because they need a decision from someone outside IT. Every version of handling them is really about making that decision easy to ask for and easy to find later.
By hand
A technician asks who owns the app, messages the manager, waits, grants the access in the admin console and writes “done” in the thread. The approval lives in a direct message, so six months later finding out who approved it means scrolling through someone’s DMs. Access tends to stay, because removing it would mean working out why it was given in the first place.
Best tools before AI
Access requests move into your identity provider: access packages in Microsoft Entra ID Governance, or Okta Access Requests. Both route the request to an approver and can set an expiry, so access doesn’t outlive the reason for it. For apps outside them, a Slack Workflow Builder form or Microsoft Forms collects the app, the access level and the reason up front. A short article per app says who owns it, who approves, and what each role gets by default.
This works when people use the form, and when someone keeps it current. Every new app needs an entry, an owner who leaves needs replacing, and an approver who changes teams keeps getting the requests until someone notices. Many still ask in the channel anyway, so someone has to point them to the right flow, and apps that sit outside the identity provider go back to the manual version.
With an AI agent
The person asks in plain language in the channel. The agent reads your access policy, knows who approves that app, asks that approver in Slack or email, and grants the access through your identity provider, Okta for example, when they say yes. If the policy says the person’s role gets the app by default, no approver is bothered at all. The request, the approval and the result land on one ticket. If the app isn’t covered by the policy, the agent hands the request to a technician with the app, the access level and the reason already filled in.
Request 03 · Devices
“My laptop isn’t working”
“Isn’t working” can mean a full disk, a pending update, a browser with 80 tabs, or failing hardware. Most of what IT needs to know to tell those apart shouldn’t have to be asked.
By hand
Twenty messages back and forth: which model, which OS version, since when, have you restarted. Then a remote session for something a restart would have fixed. The technician loses half an hour, and so does the employee.
Best tools before AI
Device management, Microsoft Intune or Jamf Pro, connected to the help desk so the device, OS version, encryption and compliance status are on hand for every ticket. A written first-steps article: restart, install pending updates, free up storage. A rule for when to remote in, with Intune Remote Help or your remote support tool, so remote sessions go to the cases that need them.
The technician looks up the device, asks only what they can’t see, and follows the article. It’s faster, but it’s still a technician reading the ticket, opening the device record and typing the first reply, for every laptop.
With an AI agent
The agent pulls the device from Intune or Jamf onto the ticket, drafts the first reply from your troubleshooting article, and asks the one or two things it can’t see in a single message: since when, and what they were doing. It asks for a screenshot of Activity Monitor on a Mac or Task Manager on Windows. The simple cases close there. The ones that don’t reach the technician with the device details, what has been tried and the screenshot already on the ticket, so they start at step two.
In Slack a pasted screenshot stays in the thread. In Teams, shared files are stored in SharePoint or OneDrive, so check that your help desk tool can open them.
Request 04 · Onboarding
“Sam starts next Monday, can we get them set up?”
Onboarding is the request with the most moving parts: an account, group memberships, app access, a laptop, and a deadline that doesn’t move. It’s also the one where a miss is most visible, because the new hire’s first message to IT is “I can’t log in”.
By hand
A checklist in someone’s head, or in a document that is a version behind. Create the account, add it to groups, order or image a laptop, request access to each app, chase each approver, and remember to tell the manager. With that many moving parts and a date that doesn’t move, the thing that slips on day one is usually the one tool specific to that role.
Best tools before AI
Your HR system as the source of truth, with SCIM provisioning in Okta or Entra ID creating the account and group memberships the moment HR enters the start date. Birthright access for the role through access packages or Okta groups. Zero-touch laptop setup with Apple Business Manager or Windows Autopilot, so the device configures itself when it’s switched on.
This covers the standard set for standard roles well. Everything non-standard is still a chain of manual tasks: a tool outside single sign-on, a role-specific app that needs an approval, a laptop that has to ship to a home address, a manager who wants to know where things stand.
With an AI agent
The manager posts the message in the channel. The agent opens a ticket, pulls the onboarding checklist for that role from your knowledge base, and asks in one reply for anything missing: start date, team, where the laptop ships. It lists what provisioning already covers and what needs a decision, asks the approvers for the non-standard access, grants it through your identity provider as they say yes, and keeps the manager updated in the same thread. The technician handles the laptop. Everything else is tracked on the one ticket, so Monday morning holds no surprises.
What the AI version needs from you
In each of the four cases the agent isn’t inventing anything. It answers from an article, acts on a policy, and reads from systems you already run. That makes three things worth doing before you add one.
- Write the article for each request. The reset steps, the access policy per app, the first troubleshooting steps, the onboarding checklist per role. If a technician can’t find it written down, the agent can’t either.
- Connect the systems the agent acts through. Your identity provider for resets and access, and your device management for laptops. The agent is only as capable as the integrations behind it.
- Name an approver for every app. The agent can ask for a decision in seconds, but only if your policy says who makes it.
Do the math on your tickets
Scroll back through the last week of your IT channel and sort the messages into the four buckets. Then count how many of them a technician read, looked up and typed a first reply for. That number is the work the AI version takes off the queue, and it is usually higher than it feels from the inside.
If you want to see what that looks like in your own Slack or Teams, here’s how Skillset handles each of the four. Skillset is an agentic service management platform. Skillset agents work in Slack and Microsoft Teams. They answer from your knowledge base, act through your identity provider and device management, and sync tickets with Jira Service Management, Zendesk and other help desks.