AcademyLead

The AI-Native IT Organization

What changes in the IT operating model when agents perform routine work: decision rights, roles, controls and metrics. The course includes a delegation framework, the DORA and METR evidence on AI productivity, and a 180-day plan.

L101Foundation14 slidesAbout 13 minutes

Start the course Download the PDF
1 / 14
PDF
AxiareteAcademy

L101Lead · Foundation

The AI-Native IT Organization

How the IT function changes when software agents carry a growing share of the work, and how to lead that change.

FormatKnowledge pack · 14 slides
ForIT leadership teams
LevelFoundation
Axiarete Academy · L10102 / 14

The premise

An AI-native IT function redesigns work, decisions and controls

Adding a copilot to each workflow can speed up individual tasks. Redesigning the workflow around what agents can now do reliably, and around what still requires human judgment, changes what the IT function can deliver.

WorkTasks agents execute are separated from decisions people own.
KnowledgeAgents work from a current model of the estate, maintained from live evidence.
ControlPolicy is checked inside the workflow, before an agent acts.
The AI-Native IT OrganizationLead
Axiarete Academy · L10103 / 14

What the evidence says

AI use is widespread, while measured gains vary

DORA finds that AI acts as an amplifier, magnifying an organization’s existing strengths and weaknesses.

90%

of technology professionals surveyed by DORA reported using AI at work, and more than 80% believed it had increased their productivity.

DORA, State of AI-assisted Software Development, 2025

19%

longer to complete tasks where AI was allowed, for 16 experienced open-source developers. Afterwards they estimated AI had cut their time by 20%.

METR randomized controlled trial, July 2025

30%

of DORA respondents reported little or no trust in AI-generated code. DORA linked AI adoption to higher delivery throughput but lower stability.

DORA, State of AI-assisted Software Development, 2025

In February 2026, METR called its newer data “an unreliable signal”, as developers unwilling to work without AI opted out, but believed developers are likely more sped up than in early 2025. Measure gains in your own workflows.

The AI-Native IT OrganizationLead
Axiarete Academy · L10104 / 14

The shift

Six shifts define the AI-native IT function

From
To
Projects staffed by people
Outcomes delivered by teams of people and agents
Documentation written after the fact
Context maintained continuously from code, telemetry and configuration
Periodic assessments and audits
Continuous evaluation that supplies audit evidence
Tickets routed to queues
Issues resolved where they are detected, escalated by exception
Controls applied at review gates
Controls enforced in the workflow, with a complete audit trail
Headcount as the limit on capacity
Review capacity, context quality and trust as the limits
The AI-Native IT OrganizationLead
Axiarete Academy · L10105 / 14

Delegation framework

Decide what to delegate by consequence and reversibility

HighConsequence of errorLow
Person approves, agent actsConfiguration changes and patches with tested rollback, through change control.
Person decides, agent informsRetiring a system of record, exiting a vendor, committing to an architecture.
Agent acts, people sampleTicket triage, test generation, documentation updates and routine queries.
Agent prepares, person executesDeleting data past retention after a legal-hold check, ending unused licenses, removing dormant accounts.
Easy to reverseReversibilityHard to reverse
The AI-Native IT OrganizationLead
Axiarete Academy · L10106 / 14

Architecture of the work

Work moves across four layers

IntentOutcomes, priorities and risk appetite, set by accountable leaders.
DecisionsArchitecture, portfolio and risk acceptance, owned by people and informed by agents.
ExecutionChanges, remediation, migrations and routine operations, increasingly performed by agents under policy.
ContextA current, evidence-based model of applications, infrastructure, code, data flows, identities and cost.
The assetEvery workflow delegated later reuses this layer. DORA names AI-accessible internal data as one of seven capabilities that increase the benefits of AI.
The AI-Native IT OrganizationLead
Axiarete Academy · L10107 / 14

People

Roles shift toward specification, review and exceptions

New responsibilities appear in every team: owning each agent, evaluating it, managing its identity and curating the context it uses.

EngineersSpecify, review and test agent output, and own the design decisions that require judgment.
ArchitectsMaintain architecture decision records and the guardrails within which agents operate.
OperationsTurn runbooks into executable policy and handle the exceptions agents escalate.
LeadersOwn outcomes across human and agent work, and decide what is delegated.
The AI-Native IT OrganizationLead
Axiarete Academy · L10108 / 14

Decision rights

Write decision rights down before agents act

DecisionAgent mayApprovalEvidence retained
Classify and route an incidentActNone; outcomes sampledClassification, confidence, outcome
Change an agent’s model, prompt or toolsPropose onlyAgent owner, after evaluation passesVersions and evaluation results
Patch a production systemAct after approvalService owner or standard changeChange record, test results, rollback plan
Retire an applicationPrepare evidence and the runbookBusiness owner and architectureUsage, dependencies, retention decision
Grant or elevate accessProvision approved requests only; never approveIdentity governanceRequest and approval chain
The AI-Native IT OrganizationLead
Axiarete Academy · L10109 / 14

Control plane

One control plane for every agent

Decision rights hold only when they are enforced the same way for every agent, whichever team built it.

RegisterEach agent listed with its purpose, accountable owner, model, tools and decision rights.
IdentityIts own non-human identity, with short-lived credentials scoped to the task.
PolicyDecision rights checked at every tool call, outside the model.
EvaluationA versioned test set, rerun before release and after any model, prompt or tool change.
Trace and costEvery action logged with its requester, agent, tool, result and cost.
StopA tested way to halt any agent and revoke its access at once.
The AI-Native IT OrganizationLead
Axiarete Academy · L10110 / 14

Maturity

Five stages from assisted to native

Moving to Delegated requires current context, controls enforced in the workflow and shadow results that meet thresholds.

AssistedIndividuals use copilots. Gains are personal and unmeasured.
AugmentedTeams standardize tools and review practices, and measure each workflow.
DelegatedAgents complete bounded tasks end to end under policy. People approve by exception.
OrchestratedAgents coordinate across workflows through shared context and one control plane.
NativeOperating model, funding and metrics assume teams of people and agents.
The AI-Native IT OrganizationLead
Axiarete Academy · L10111 / 14

Measurement

Six measures for delegated work

Cycle timeFrom request or detection to a verified change in production.
Change fail rateShare of deployments that need immediate intervention, such as a rollback or hotfix.
Autonomous completionShare of work agents finish without human intervention, with sampled quality checks.
Correction rateShare of agent output that people correct or revert.
Cost per outcomeFully loaded, including model, compute and review effort.
Context coverageShare of the estate whose records were verified against live data within a set period.

AvoidLicenses activated, prompts per user and lines of code generated count activity; report them only beside outcomes.

The AI-Native IT OrganizationLead
Axiarete Academy · L10112 / 14

Pitfalls

Six common failure patterns in IT’s AI programs

Buying tools before redesigning the work
Redesign three workflows end to end, then choose tools
Reporting adoption as success
Report cycle time, quality and cost per outcome
Granting agents broad standing access
Scoped, short-lived permissions for each task
Letting every team build its own stack
One control plane for identity, logging, evaluation and policy
Ignoring the quality of context
Invest first in a current model of the estate
Delegating without a shadow period
Compare agent and human results before removing approvals
The AI-Native IT OrganizationLead
Axiarete Academy · L10113 / 14

Getting started

A 180-day plan

Days 1–30Choose and baseline
  • Select three high-volume, reversible, measurable workflows
  • Baseline cycle time, quality and cost
  • Write decision rights and thresholds for each workflow
Days 31–90Build the foundations
  • Stand up agent identities, logs, evaluation and cost tracking
  • Build current context for the systems in scope
  • Run agents in shadow mode against human results
Days 91–180Delegate and scale
  • Approve by exception where shadow results meet thresholds
  • Publish results against the baseline
  • Extend to the next workflows; adjust roles and funding
The AI-Native IT OrganizationLead
Axiarete Academy · L10114 / 14

Summary

Key takeaways

  1. AI-native IT redesigns work around what agents do reliably and what still needs human judgment.
  2. Delegate by consequence and reversibility, and write decision rights down before agents act.
  3. Current, evidence-based context and enforced controls make delegation safer.
  4. Measure cycle time, stability, quality and cost per outcome; treat usage as context only.
  5. Start with three workflows, one control plane and a published baseline.
Next course · L201IT Economics: Cost, Value and AI SpendLead · Practitioner
The AI-Native IT OrganizationLead

Further reading

Continue learning

L201 · LeadIT Economics: Cost, Value and AI SpendAxiarete AcademyAll courses and tracks

Bring this course to your leadership team

We run private working sessions that apply these frameworks to your own estate, data and pending decisions.

Request a Private BriefingExplore the Academy