AI and IT Teams Often Clash. But They Don't Have To.

Graham Kenny, Kim Oosthuizen, and Ganna Pogrebna · article · 2026-07-31 Reader 🖥️

Agent Governance (Internal)Enterprise AI Adoption & Readiness Legacy & Organizational Forgetting Organizational Debt Organizational Legibility Vendor Research Conflict of Interest

AI and IT teams clash not because of a single technical disagreement but because they optimize for structurally different priorities (control vs. experimentation), and organizations resolve this — without simply picking a side — through deliberate structural interventions: reframing disputes around shared outcomes, separating roles with an explicit risk taxonomy, and mapping AI ambitions to the specific IT foundations they depend on.

practical · mixed — Argumentative framing (unbundling 'technology' into IT vs. AI priorities) delivered through three anonymized narrative case studies, each structured as problem-then-resolution, closing with a practical call to action.

Structure

Part	Role
Opening framing: IT and AI are bundled under the vague label 'technology' but have opposed priorities — IT seeks control, standardization, and risk reduction; AI seeks experimentation, flexibility, and rapid learning.	context / thesis setup
Case 1 — Baringa Group: IT and AI disagree about what counts as legitimate 'data' (structured/quantitative vs. unstructured/qualitative). Resolution: reframe around a shared business outcome, formally extend data-governance standards to unstructured sources, build a toolkit of reusable AI components, and tie IT and AI to joint KPIs.	evidence (case study 1)
Case 2 — Southern Bank: IT and AI disagree about competence and auditability. Resolution: split the Head of AI / Head of IT roles for clear accountability, impose a mandatory taxonomy separating analytical AI (explainable/auditable) from generative AI (different risk profile), and rewrite overlapping role definitions.	evidence (case study 2)
Case 3 — PacificTel: IT and AI disagree about whether the IT foundation is strong enough to support AI ambitions. Resolution: categorize AI initiatives into low-foundation, data-dependent, and foundation-critical tiers; connect the AI roadmap to the IT consolidation roadmap via a single dependency map; replace activity metrics with outcome/readiness metrics.	evidence (case study 3)
Closing call to action: move beyond compromise to integration by redefining shared language around 'AI', aligning KPIs, and embedding governance into shared platforms.	conclusion / prescription

Key terms

technology (as a bundling label)
An umbrella term that organizations use to group IT and AI together, obscuring that the two functions have systematically opposed priorities and operating logics.
foundation-critical AI
PacificTel's top tier: AI use cases (predictive network fault detection, churn intervention, automated customer response) that need stable pipelines, consistent identifiers, and real-time monitoring, and cannot be built on fragmented infrastructure.
low-foundation AI
PacificTel's bottom tier: AI use cases (drafting documents, searching policy archives, internal knowledge tools) that create value on controlled document sets with usage rules and human review, without depending on live operational data.
single dependency map
A shared artifact connecting specific AI ambitions to the specific IT systems/consolidation milestones they rely on, converting an abstract 'IT is blocking us' argument into a concrete, negotiable roadmap both sides can see.
mandatory taxonomy (analytical vs. generative AI)
Southern Bank's classification splitting 'AI' into analytical AI (machine learning, predictive modeling, decisioning — must be explainable and auditable) and generative AI (text, summaries, conversational outputs — a different risk profile), applied to every project intake, risk template, and reporting pack.

Propositions

explicitIT and AI teams, though often grouped together as 'technology,' operate with systematically different priorities: IT emphasizes control, standardization, and risk reduction; AI teams seek experimentation, flexibility, and rapid learning.conf: high
explicitLeft unmanaged, IT's and AI's competing priorities cause the two groups to unintentionally sabotage each other's efforts.conf: high
implicitDisputes that surface as being about 'data,' 'competence,' or 'infrastructure foundations' are, underneath, disputes about differing team priorities and risk tolerances — the technical vocabulary is the costume, not the cause.conf: medium
explicitReframing a definitional dispute (what counts as 'data') around a shared business outcome ('what improves decisions for customers') dissolves the standoff without requiring either side to concede its definition.conf: high
explicitFormally extending existing data-governance standards to unstructured/qualitative sources — rather than treating them as exempt — lets an organization satisfy IT's need for control and AI's need for legitimacy at the same time.conf: high
explicitJoint KPIs tied to shared commercial outcomes, with IT retaining governance/compliance ownership and AI retaining experimentation/delivery ownership, create a structural incentive for cooperation that goodwill alone does not.conf: high
explicitSeparating the Head of AI and Head of IT roles — previously combined under one General Manager — into distinct roles with clear ownership boundaries surfaces and resolves disputes earlier than a blurred, shared-accountability structure does.conf: high
supportsEngineering Leadership Report 2026 · p1 — Southern Bank's split of the Head of AI / Head of IT role is a concrete instance of the team/division restructuring the Postman survey finds to be the most common significant AI-era workforce change (67% of respondents).
explicitCreating a mandatory taxonomy that splits 'AI' into analytical AI (explainable, auditable) and generative AI (a different risk profile) reduces IT-AI hostility because it demonstrates to IT that most disputed AI work already fits within governance categories it recognizes.conf: high
explicitClarifying overlapping role definitions (data engineer, data scientist, ML engineer, AI product owner, IT) makes handover points visible and reduces disputes over who owns model development, deployment, and validation.conf: high
explicitNot all AI use cases depend equally on IT infrastructure maturity; categorizing initiatives by foundation-dependency (low-foundation, data-dependent, foundation-critical) lets an organization proceed on some initiatives immediately while gating others on infrastructure readiness.conf: high
explicitAn AI model that performs well in a controlled test environment but degrades or fails in live production is evidence of fragmented, inconsistent, or late-arriving underlying data — not evidence that IT is simply too slow.conf: high
extendsAI as an Enterprise Operating System · p8 — Extends the general claim that decades of accumulated data-access debt become acute the moment AI is deployed, with a concrete production-failure mechanism (works in test, degrades live because underlying data is fragmented) rather than a security/access framing.
explicitA single dependency map connecting specific AI ambitions to the specific IT consolidation milestones they require converts an abstract 'IT vs. AI speed' argument into a concrete, sequenceable roadmap that both sides can act on.conf: high
explicitAI's competitive-pressure timeline and IT's multi-year legacy-consolidation timeline are both individually valid, which is exactly why the standoff resists resolution through argument alone — it requires structural sequencing of work, not persuading one side that the other is wrong.conf: high
extendsThe Organization Is the Bottleneck · p3 — Extends 'engineering maturity determines whether AI adoption succeeds' by identifying the actual mechanism as a timeline conflict — AI's competitive urgency vs. IT's multi-year consolidation need — that requires explicit structural sequencing, not just higher organizational maturity.
explicitReplacing activity metrics (pilots launched, prototypes running) with outcome and readiness metrics makes IT's otherwise-invisible infrastructure work legible in business terms while making AI's program more credible to skeptical stakeholders.conf: high
implicitThe three cases share a common underlying structure — differing priorities, competence assumptions, or infrastructure timelines between IT and AI — even though each is presented as though it were that company's unique problem.conf: medium
implicitThe article presents all three 'what the company did' resolutions solely through the account of authors who work in an advisory capacity with the (anonymized) companies, treating the consultants' own record of designing these interventions as neutral case evidence rather than as a self-interested account requiring independent corroboration.conf: high
supportsResearch: How AI Agents Broaden the Scope of Knowledge Work · p14 — Same undisclosed-institutional-interest pattern in a different guise: HBR contributors with a direct commercial/advisory stake (Perplexity's own product; the authors' consulting practice) present their own account of outcomes as neutral case evidence without disclosing or examining the conflict.

Arguments

Reframing a definitional dispute around a shared business outcome resolves IT-AI conflict better than adjudicating whose definition of 'data' is correct. strength: medium

Explicit premises:

Baringa's leadership reframed the data question at the enterprise level rather than resolving the IT/AI definitional debate directly.
The company formally expanded its enterprise data inventory to include unstructured sources, classifying them with the same standing as structured databases.
A joint accountability model with shared governance structures and KPIs tied to customer/business outcomes was established.

Implicit premises:

IT will accept unstructured data as legitimate only if it arrives bundled with governance guarantees (control frameworks), not from an executive mandate alone.
A 'shared business outcome' can be made specific and measurable enough to actually adjudicate contested technical/data decisions, rather than functioning as vague rhetoric that both sides nod along to without changing behavior.

Inferences:

If a definitional dispute is really a dispute about whose standards govern the enterprise, other definitional disputes (e.g., what counts as a 'production-ready model,' what counts as a 'supported system') might be resolvable the same way — by re-anchoring to a shared outcome metric rather than settling the definition.

Structurally separating AI and IT leadership while imposing a mandatory AI taxonomy resolves a competence standoff better than either merging the functions or leaving AI subordinate to IT. strength: medium

Explicit premises:

Southern Bank split the Head of AI and Head of IT roles, which had previously blurred accountability under one General Manager.
A mandatory taxonomy separated analytical AI (explainable, auditable) from generative AI (different risk profile), applied to project intake, risk templates, and reporting.
Overlapping role descriptions were rewritten to make handover points visible.

Implicit premises:

IT's hostility toward AI stemmed specifically from generative AI's unfamiliarity and inauditability, not from AI capability in general — which is why showing that most disputed work is actually analytical AI (already governance-familiar) defuses the conflict.
Clear role boundaries reduce interpersonal/organizational conflict independent of whether the underlying technical capability gap between the two functions has actually closed.

Inferences:

A nascent capability gains legitimacy faster inside a skeptical mature function when it is shown to already comply with that function's existing governance categories, rather than by demanding the mature function adopt new categories on the nascent capability's terms.

The IT-AI standoff over infrastructure 'foundations' is resolved not by siding with IT's caution or AI's urgency, but by categorizing AI initiatives by foundation-dependency and sequencing investment through a shared dependency map.

Explicit premises:

PacificTel divided its AI portfolio into three foundation-dependency tiers (low-foundation, data-dependent, foundation-critical).
A single dependency map connected specific AI ambitions to the specific IT consolidation milestones they required.
Activity metrics were replaced with outcome and readiness metrics.

Implicit premises:

AI and IT can actually agree on which tier a given initiative belongs to — the categorization scheme itself does not eliminate disagreement, it just relocates it to a classification question.
Giving AI a subset of initiatives (the low-foundation tier) to pursue immediately can partially decouple AI's competitive urgency from IT's multi-year consolidation timeline, defusing pressure without actually resolving the underlying resource/timeline conflict.

Inferences:

If categorizing by infrastructure-dependency works for IT/AI conflict, the same technique likely generalizes to other 'we need to move fast but the platform isn't ready' conflicts between product/feature teams and platform-engineering or infrastructure teams.

Narratives

Baringa Group (diversified conglomerate): Head of AI Helen describes IT's quantitative, table-and-column view of 'data' clashing with AI's need for qualitative, conversational data (call transcripts, service notes); leadership resolves it by reframing around a shared customer-outcome goal, formally extending data-governance standards to unstructured sources, and creating joint IT/AI KPIs.
→ Supports propositions 4-6 (shared-outcome reframing, governance extension, joint KPIs) — shows structurally, not just asserts, how a definitional standoff was dissolved.
Southern Bank (36,000-employee bank): IT manager Gerry and AI lead Anjali describe a standoff over competence and auditability; the bank splits the combined Head of AI/IT role, imposes a mandatory analytical-vs-generative AI taxonomy on intake/risk/reporting, and rewrites overlapping role definitions.
→ Supports propositions 7-9 (role separation, taxonomy-driven de-escalation, role-architecture clarity) — illustrates that reorganizing categories, not adjudicating who is 'right' about competence, resolved the dispute.
PacificTel (9,200-employee telecom): IT head Ian and AI head Jennifer describe a standoff where an AI model that worked in test degraded in production because of fragmented legacy data; the company categorizes AI initiatives by foundation-dependency, builds a single dependency map linking AI ambitions to IT consolidation milestones, and shifts to outcome/readiness metrics.
→ Supports propositions 10-14 (foundation-tiering, model-degradation-as-evidence, dependency mapping, dual-timeline validity, outcome metrics as legibility mechanism) — the case study the authors treat as showing that neither side's account was simply wrong, only incomplete without the other's.

Questions addressed

✅ Why do AI and IT teams clash inside organizations that lump them together under the single label 'technology'?
Because IT and AI optimize for structurally opposed priorities — control/standardization/risk-reduction versus experimentation/flexibility/rapid learning — and the shared label obscures this until it surfaces as friction.
✅ How can an organization resolve IT-AI conflict without simply forcing one side's approach to win?
Through three structural interventions demonstrated across the cases: reframing disputes around a shared business outcome, separating leadership roles alongside an explicit risk taxonomy, and mapping AI ambitions to the specific IT foundations they depend on.
⭕ Do these structural fixes (role separation, taxonomies, dependency maps, joint KPIs) actually hold up over time, and were they causally responsible for the reported improvements rather than coincidental with other changes or simply the passage of time?
Not addressed — the article presents each resolution in the past tense as an accomplished success, with no follow-up timeframe, no metric showing durability, and no consideration of alternative explanations for the reported improvement.

Gaps

incompleteAll three case studies are the authors' own advisory engagements, with company names anonymized and altered; the sole evidentiary source for whether each resolution 'worked' is the consultants who also participated in designing the intervention. No independently checkable detail (real company name, date, financial or retention outcome, third-party account) is offered for any of the three resolutions, and the authors' own commercial position (Kenny is CEO of an advisory firm; Oosthuizen and Pogrebna hold advisory/consulting roles) is never flagged as a potential source of favorable framing.
Read next: Look for independent, longitudinal accounts of IT/AI organizational restructuring — written by parties without an advisory stake in the outcome — to check whether structural fixes like role separation, mandatory taxonomies, or dependency mapping hold up beyond the initial engagement.
incompleteThe article treats classification into taxonomy tiers (Southern Bank's analytical/generative split; PacificTel's low-foundation/data-dependent/foundation-critical split) as though the categorization itself is unambiguous, but never addresses who arbitrates a disputed classification or what happens when AI and IT disagree about which tier a specific initiative belongs to — precisely the kind of disagreement the taxonomy was introduced to eliminate could simply recur one level up, as a fight over labels instead of foundations.
Read next: Compare against more developed AI-governance frameworks that specify an arbitration mechanism for contested classifications (e.g. Entur's ki-løype structured governance process) to see whether any resolve the 'who classifies the classification' regress.
incompletePacificTel's central trade-off is never quantified: the article states that IT's consolidation work 'will take years' that AI's competitive pressure 'does not allow,' but never explains how much runway the low-foundation tier actually buys before competitive pressure forces a foundation-critical bet regardless of readiness, nor what happens to Jennifer's competitive position if consolidation continues to slip.
Read next: Read alongside sources on legacy-system consolidation timelines and technical-debt paydown rates in telecom/regulated industries to assess whether a 'years' estimate for foundation work is realistic against typical AI-driven competitive cycles.
incompleteThe article assumes a clean binary of a centralized 'AI team' standing opposite a centralized 'IT team' in every organization, but does not address organizations where AI work is distributed across many existing product/engineering teams rather than owned by one function — a structure that is increasingly common as AI tooling diffuses. In that structure, the proposed remedies (splitting leadership roles, imposing one enterprise-wide taxonomy) may not map cleanly onto how the conflict actually manifests.
Read next: Compare against accounts of embedded/distributed AI ownership models (AI capability spread across product teams rather than centralized) to test whether the 'two tribes' framing and its structural remedies still apply.

AI and IT Teams Often Clash. But They Don't Have To.

Structure

Key terms

Propositions

Arguments

Narratives

Questions addressed

Gaps

Related documents