GRAIL · Function papers
What does an AI-augmented IT department look like?
By Johan Grönstedt · Last reviewed
The work moves closer to the user. Control stays with IT. This free nine-page position paper sets out what changes, what runs through the agent layer, and what stays with the person.
An AI-augmented IT department becomes the steward of how every company agent reads, reasons and acts across company systems. The business gains reusable capabilities, while IT sets the boundaries and keeps consequential decisions with people.
A controlled flow
A day in The IT function in 2028 as a controlled flow
The flow begins with the employee or agent and the identity, device, entitlements and recent changes attached to the request. An approved capability can then retrieve a permission-aware answer or propose an approved action. Identity, permissions and branch controls are checked before the action reaches an authoritative record, with the accountable person, approval and rollback plan retained.
Employee or agent
Identifies the employee, device, entitlements and recent changes.
Approved capability
Retrieves a permission-aware answer or proposes an approved action.
Identity and permission
Checks identity, permissions and branch controls before action.
System of record
Keeps every interaction as a ticket in authoritative records.
Audit and rollback
Retains each accountable person, approval and rollback plan.
The operating split
What runs, and what stays with the person
IT has spent decades receiving requests, joining systems and protecting access one project at a time. The structural change is that IT becomes the steward of how every company agent reads, reasons and acts across those systems. The business gains reusable capabilities. IT sets the boundaries.
Across support, operations, security, applications and integration, the agent layer structures work, retrieves approved knowledge, correlates records, drafts responses and prepares bounded actions. People retain ambiguous cases, consequential approvals, policy and architecture choices, production judgment, semantics, ownership and exceptions.
The same division continues through software maintenance, business technology projects and governance. For bounded software changes, the agent layer creates tests and documentation, investigates defects, proposes migrations and returns a pull request with provenance and risk. Specification, architecture, review, production acceptance and hard failures stay with the person. For business technology projects, agents turn decisions into requirements, dependencies, actions and risk updates, while people retain scope conflicts, supplier choices, trade-offs and priority. In architecture and governance, agents maintain dependency maps, find policy exceptions and technical-debt patterns, and propose standards from current system evidence. People retain target architecture, acceptable exceptions and technology capital allocation.
| Process | What the agent layer does | What stays with the person |
|---|---|---|
| Service desk and support | Structures the issue, retrieves approved knowledge, drafts replies, verifies the user and device, and proposes standard resets or software access | Ambiguous cases, emotional situations, exceptions and approval of consequential device or access changes |
| Infrastructure and cloud operations | Correlates alerts, configuration, deployments and tickets; prepares an incident narrative, likely causes and a tested remediation or rollback plan | Diagnosis where evidence conflicts; approval of production changes with meaningful blast radius |
| Security, identity, access and compliance | Correlates alerts across identity, endpoint, email, vulnerabilities and the SIEM; explains suspicious activity and drafts response steps | Containment, privilege and exception decisions; policy and detection design |
| Applications, vendors, licenses and cost | Assembles usage, duplication, contract terms, invoices, support history, renewal dates and business dependency | Negotiation strategy, service-risk decisions, acceptable lock-in and approval to remove access |
| Integration and data plumbing | Proposes mappings, generates adapters, tests schemas and exposes versioned business capabilities through a shared gateway | Semantics, permissions, ownership, failure behavior and approval of writes |
The position
Six things GRAIL believes
The connector layer is a product, not a collection of projects.
A one-off interface solves one request and leaves the next team waiting.
Identity comes before agency.
An agent that can act without its own identity, scope and expiry is an uncontrolled shared credential.
The queue should become a sensor.
Tickets matter, but they describe what somebody has already noticed.
More code is not the same as more delivery.
Generated changes still consume specification, review, testing and production judgment.
The system of record should remain authoritative without remaining the main place people work.
Employees can ask for help in Teams.
The role redesign is the investment.
Service specialists become knowledge and exception owners.
Get the paper
The full position paper shows how the company gains more ways to act while boundaries become clearer, records become fuller and consequential decisions remain unmistakably human.