As of
Is the L1/L2/L3 IT support model outdated? The tiers that remain are approvals
IT support tiers were built to gather context and route to skill one escalation at a time, and once an AI agent has that context before the request, the L1/L2/L3 tiers that remain are about who can approve.
Most service desks still run on a ladder. A request lands with level 1, which handles what it can and passes the rest to level 2. Level 2 investigates and passes the hard cases to level 3, the engineers who own the systems. Contracts, team structures and career paths are built on those three rungs.
The ladder has real virtues. Specialists aren't interrupted by password resets, and new hires learn the environment on requests that forgive a mistake.
The case against the ladder has been made before, and its core is right: the tiers were never mostly about skill. A level 2 analyst isn't smarter than a level 1 analyst so much as further into the investigation. That's where most of the argument stops. The more useful question is what's left of the ladder once nobody has to climb it for facts.
Is the L1/L2/L3 support model outdated?
Yes, as a way of routing work by skill. The tiers mostly existed to gather context step by step, and once that context is ready, the main reason left to hand a request up is that someone has to say yes.
Look at what each rung does. Level 1 learns who the requester is and what they're trying to do. Level 2 learns what they hold: groups, devices, licenses. Level 3, or the owner of the system, learns whether they should hold it and what changed recently. Each tier is a stage of investigation, and escalation is how the investigation reaches whoever can see further.
A finance analyst moves teams on a Monday, and HR updates the record that morning. On Wednesday the analyst can't open the finance reporting workspace. Level 1 sees an active account and resets the password. Level 2 opens the identity provider, which still shows the old team and its groups. The systems owner emails the manager to find out which record is right. Access arrives on Friday.
Nobody on that ticket lacked skill. Two of the three handovers were for facts. Only the last one was for a decision, and it was sent to whoever happened to be holding the ticket, by email, with no record of the rule behind it.
Why shift-left, swarming and knowledge bases kept the ladder
Service management has spent years pushing context toward the first responder, and each approach helped. Shift-left moved runbooks and permissions down a rung. Knowledge-centered support captured each fix as an article. Swarming dropped the queue between tiers and pulled the right people into one conversation.
None of them changed when the context is gathered. A person still assembles the picture after the request arrives, opening the HR system, then the identity provider, then device management, and reconciling what they find by eye. A knowledge article tells you the fix for a role mismatch; it can't tell you this person has one. And the disagreements are where the ladder earns its keep: HR says one role and the identity provider another, or a contractor converts to full-time and the agency account keeps its seats because nothing linked the two. The ticket climbs until it reaches someone who knows which system to believe.
What's left for the tiers when an AI agent has the context
Two things change.
The first is that the context exists before anyone asks. Role and manager come from the HR system, accounts and groups from the identity provider, devices from device management, each read from the system that owns it as it changes. A person's accounts are joined into one identity, so the corporate account and the old agency account resolve to the same person. Every fact carries the system it came from and when it was observed, so a stale group membership doesn't outweigh last week's HR change.
The second is that the AI agent can act. A fix your team has reviewed and approved runs through the systems you've connected, so the agent doesn't hand a known fix to someone with admin rights.
With both in place, a request doesn't go up the ladder because a fact is missing or because the fix needs another rung's permissions. It goes up because the request needs a yes from someone with the authority to give it.
So the tiers that remain are approval tiers, defined by authority rather than know-how. The requester's manager approves what their team uses. The owner of an app or a data set approves who gets into it. Security approves what touches sensitive systems. Finance approves new spend. These aren't levels anyone climbs. They're a map of who can say yes to what.
Approval tiers belong in policy, not in the queue
On a ladder, approval authority lives in people's heads. Level 1 knows to ask the manager for some requests, level 2 knows the data owner for others, and the rest is learned by getting it wrong. Escalating for authority looks the same as escalating for skill, so nobody can say how much of the ladder is really waiting for a yes.
Written down, it becomes a policy. For each kind of request, policy says whether it runs on its own, is blocked, or needs approval, and who approves. A request for a paid design tool from someone already on the design team might run on its own; the same request from another team waits for the budget owner. The approver gets the request directly, with the requester's role and current access attached, and makes one decision. The agent carries it out.
That changes three things about escalation. It's explicit: the rule that sent the request for approval is named. It's routed to the right person the first time, not passed along until it finds them. And it's on the record, so an auditor can see which rule applied, who approved and when.
Where "only approvals" overstates it
Approval isn't the only reason a request still reaches a person. Some work needs hands: a laptop that won't boot needs someone to look at it. Some needs judgment nobody has written down: a major incident, a problem your team hasn't seen before, a request for an exception to the policy itself.
Those still go to people, and they should. What changes is how they arrive. The ticket comes with the requester's identity, device, access and recent changes already attached, and the reasoning the agent followed. When the agent stops because HR and the identity provider disagree about a role, whoever picks it up sees both values, which system reported each and when, and decides which is current.
So escalation by skill doesn't vanish. It becomes rare, and it stops being the default route for everything the first responder couldn't see. Escalation by authority becomes the common case, and it's the one you can write down.
What your team's work becomes with an AI agent
The work moves from gathering context to deciding with it. Your team makes the judgment calls the policy can't: exceptions, incidents, the problems nobody has seen. It also writes and owns the policy, which is where the real control sits. Deciding who approves finance reporting access teaches more about the environment than granting it by hand a hundred times.
Level 2's backlog gets worked: the access review, the device refresh, the identity cleanup. Accounts nobody linked stop being something you discover one escalation at a time and become a list. The skills path moves too. People early in their careers learn by reading why the agent decided what it did and which facts it rested on, and by turning a fix the team keeps repeating into an approved workflow.
How STLabs routes requests by authority
STLabs starts from Axiom™, a live model of your organization's people, devices, apps and access, built from the systems that own those facts and kept current as someone joins, changes roles or leaves. When your tools know the same person by different IDs, Axiom™ keeps one record with every ID attached. Every fact records which system it came from and when it was observed, and Axiom™ tracks who reports to whom, so a manager approval goes to the current manager, not the one on last quarter's org chart.
Your approval tiers live in Policies & Approvals. Your rules allow a request, block it or require approval, and name who approves. The workflow waits for the decision before it acts, and every decision is logged with the rule applied, the approver and the time. Approvers see only the requests that need a person. Fixes your team approves become versioned Workflows the agent runs through your connected systems.
When a request needs hands or judgment, the Service Desk delivers it with context attached, and Automated Triage sets the queue, SLA and owner. For what this looks like at the level of a single request, see how to reduce Tier 1 service requests; for the practices around it, ITIL without the paperwork. This is what AI-native service management changes about support: the investigation is done before the request, and the authority is written down.
How to test this on your own systems
Your ticket history shows what your ladder is really for. Export the last 90 days of tickets and pick every one reassigned at least once. For each handover, write down why it moved: to get a fact from another system, to reach someone with the skill or admin rights to fix it, or to get a yes from someone with authority.
Then look at the yes column. For each approval, note whether a written rule said who should give it, and whether it reached that person first. If most handovers were for facts or for a yes, your tiers are a context problem and an approval map nobody wrote down. To see it with real requests, connect STLabs alongside the service desk you already run and compare what reaches your team for one request type. Teams go from integration to results in five days; the plan is in Moving to STLabs from your current service desk.
Keep the judgment, write down the authority
Start with the request type that climbs most often. Write the policy for who approves it, let the agent take the first pass, and watch what still reaches your team.
The ladder sorted people by what they could find out; what's left of it sorts them by what they can say yes to.
Frequently asked questions
Is the tiered ITSM support model outdated?
Largely, yes. Tiers gathered context one escalation at a time. When an AI agent has that context before the request, most handovers left are for approval, so the remaining tiers are defined by who can approve.
What is the difference between L1, L2 and L3 support?
Level 1 handles requests with a known fix, like password resets and how-to questions. Level 2 investigates issues that need specialist knowledge. Level 3 is the engineers who own the systems and handle the hardest cases.
Do ITSM approval tiers replace L1, L2 and L3?
For most requests, yes. Policy says which requests run on their own, which need approval and who approves: the manager, the app or data owner, security, or finance. Hands-on fixes and new problems still go to people.
How do IT staff build skills without tier 1 work?
By handling exceptions and incidents with the full picture attached, reading which facts each agent decision rested on, writing the policies that decide who approves what, and turning repeated fixes into approved workflows.