AI Transformation in Practice
The difficult AI questions rarely begin with the technology.
They begin when a promising capability enters a real situation, when an executive commitment must become organizational movement, when an AI-supported action crosses a boundary, or when many local successes have to become something the enterprise can responsibly rely upon.
CASES
Six public-source cases. Six different leadership problems.
The cases below use publicly available information to show what those different problems can look like.
These are analytical case vignettes, not client case studies. They do not imply that advice, evaluation, audit, or a diagnostic conclusion about any organization shown here. They illustrate questions that leaders could investigate in their own organizations.
POSSIBILITY
Abridge — When a promising AI future changes the clinical encounter
Abridge began by making ambient clinical documentation less intrusive and has since expanded toward a broader clinician-intelligence platform. In August 2026, Abridge describes a platform connecting clinical conversations with documentation, evidence, decision support, referrals, handoffs, and other clinical workflows; it says the platform will support more than 100 million patient-clinician conversations across more than 300 U.S. health systems this year.
The interesting strategic question is therefore larger than whether ambient AI saves clinicians time.
What future of care is actually becoming possible—and what must remain intact as that future expands?
A leadership team might need to distinguish clinician productivity from changes in attention, patient trust, professional judgment, evidence use, workflow, institutional responsibility, and the quality of the encounter itself.
The KOZCAN lens: Start with the situated reality. Establish how AI materially changes the relationship before deciding whether the possibility deserves broader realization.
Ask yourself: Are we scaling a useful technology—or have we established what future we are actually making real?
Sources: Abridge Unveils Patient-Centered Clinician Intelligence Platform Connecting Care Delivery, Payment, and Evidence-Based Treatment (June 11, 2026), Abridge Makes Context-Aware Clinical Intelligence Available to Every Clinician at Partner Health Systems (August 17, 2026)
MOVEMENT
Shopify — When “AI-first” has to become more than an executive position
Shopify CEO Tobi Lütke publicly made effective AI use a baseline expectation for employees in 2025. Shopify’s product architecture has continued moving in the same direction. Its Spring 2026 release opened end-to-end agentic commerce more broadly to developers, while Sidekick extensions allow outside applications to bring data and staged actions directly into merchant conversations. Shopify reported that weekly active shops using Sidekick had increased fourfold year over year in Q1.
That creates a different leadership question.
When does an AI-first commitment become organizational movement rather than abundant AI activity?
Policies, tools, agents, prototypes, performance expectations, and new interfaces are actions. Transformation requires evidence that those actions are changing roles, routines, decision habits, operating expectations, merchant relationships, product development, and the organization’s trajectory.
The KOZCAN lens: Trace one consequential AI-enabled action through the organization. Determine where it actually changes practice—and where movement stops, distorts, or remains unproven.
Ask yourself: We can show that AI use is increasing. Can we show that the organization itself is moving?
Sources: Agentic commerce for every developer: The Spring ’26 Edition (June 17, 2026), Selling everything, everywhere, all at once: The Spring ’26 Edition (June 17, 2026)
TRANSLATION
Atlassian — When AI starts carrying work across the organization
Atlassian’s Rovo architecture increasingly places AI inside the connective tissue of knowledge work. In May 2026, Atlassian reported millions of agentic automations and described Teamwork Graph as a contextual layer connecting work, people, tools, goals, and decisions. Rovo can search, plan, execute multi-step workflows, and coordinate actions across Jira, Confluence, service portals, chat, and connected applications. Rovo Studio also incorporates roles, approvals, versioning, and governance controls.
That makes the strategic problem one of translation, not merely integration.
What has to travel when AI moves work from one organizational context into another?
Information may travel while meaning changes. A recommendation can arrive without its evidence. Authority can silently shift. Local context can disappear. Responsibility can become ambiguous even when the workflow executes perfectly.
The KOZCAN lens: Follow one consequential handoff. Examine what is transmitted, what actually arrives, what changes in translation, whose agency matters, and which commitments must survive the crossing.
Ask yourself: Our systems are increasingly connected. Are our meanings, evidence, authority, and responsibilities connected too?
Sources: Rovo makes AI-native teamwork real for the enterprise (May 6, 2026), Atlassian Teamwork Graph: The context engine behind your AI—everywhere (May 6, 2026)
ACCOUNTABILITY
Air Canada — When the chatbot’s answer becomes the organization’s problem
In Moffatt v. Air Canada, a customer relied on incorrect bereavement-fare information supplied through Air Canada’s chatbot. The British Columbia Civil Resolution Tribunal rejected the argument that the chatbot could effectively be treated as separate from the company and found that Air Canada had failed to take reasonable care to ensure that the information it provided was accurate.
The important organizational lesson is not simply that chatbots can be wrong.
What happens when an AI-mediated interaction crosses from interface into policy, authority, obligation, and remedy?
A customer does not experience the chatbot, policy page, service representative, refund process, and legal department as unrelated systems. The organization may nevertheless govern them that way.
The KOZCAN lens: Map the real decision route. Identify where evidence, policy, authority, obligation, customer standing, escalation, and remedy must remain connected—and where they can separate.
Ask yourself: If our AI tells someone something consequential, can the rest of the organization still stand behind the decision pathway?
Sources: BC Tribunal Confirms Companies Remain Liable for Information Provided by AI Chatbot (February 29, 2024), Moffatt v. Air Canada: A Misrepresentation by an AI Chatbot (February 19, 2024)
DECISION LINEAGE
Morgan Stanley — When AI enters professional judgment
Morgan Stanley’s AI @ Morgan Stanley Debrief illustrates a different class of AI transformation. With client consent, the system can generate meeting notes and action items, summarize key points, prepare an email draft for the financial advisor to edit, and save meeting information into Salesforce. The firm emphasizes that advisor discretion and the human client relationship remain central.
The strategic question is no longer merely whether generative AI can summarize a meeting accurately.
What must remain reconstructable when an AI-supported conversation becomes an organizational record and future action?
A consequential decision pathway may now include client statements, AI-generated interpretation, advisor judgment, compliance constraints, CRM records, follow-up actions, exceptions, and subsequent learning.
The KOZCAN lens: Reconstruct the decision, not just the data. Establish what evidence mattered, what AI contributed, what the professional accepted or changed, who held authority, what obligations applied, and what became part of the official record.
Ask yourself: Could we explain afterward not only what the AI produced, but how the consequential decision actually came to be?
Sources: Morgan Stanley Wealth Management Announces Latest Game-Changing Addition to Suite of GenAI Tools (June 26, 2024), Morgan Stanley Wins Two 2025 Celent Model Wealth Manager Awards for Technology Innovation in Wealth Management (June 30, 2025)
ENTERPRISE COHERENCE
Walmart — When local AI decisions begin accumulating into global effects
At Walmart’s scale, AI is not a single-system question. Public material spans customer experiences, associate tools, inventory, supply chains, forecasting, agentic workflows, commerce, and platform infrastructure. In April 2026, Walmart Global Tech described the problem directly: at organizational scale, agents inevitably interact across teams, local decisions accumulate into global effects, and autonomy becomes difficult to govern unless constraints, provenance, authority, tool use, and decision traces travel with the system.
That produces the enterprise question.
When many AI applications work locally, what makes them compose into an enterprise capability rather than a growing collection of locally sensible decisions?
Uniformity is not the answer. Different contexts may legitimately require different arrangements. The challenge is deciding what must become common, what should remain local, where arrangements conflict, and what enterprise leadership can actually rely upon.
The KOZCAN lens: Preserve the bounded cases first. Then test which findings can responsibly travel across them, which variations are legitimate, where incompatibilities matter, and where broader composition has simply not yet been established.
Ask yourself: Are we scaling AI—or are we also building the capacity to govern what happens when all those local decisions meet?
Sources: Designing Governable Agents (April 20, 2026), Inside Walmart’s Strategy for Building an Agentic Future (May 29, 2025)
Six companies. Six different governing problems.
The cases look different because AI transformation does not have one universal failure mode.
- A possibility may be worth exploring but not yet ready to become real.
- An executive commitment may create activity without organizational movement.
- A connected workflow may transmit information while losing meaning or authority.
- A locally functioning AI interaction may fail when it crosses into organizational responsibility.
- A professionally supervised AI tool may still require reconstructable decision lineage.
- And many local AI successes may still fail to establish something the enterprise can responsibly rely upon.
The point is not to place every organization into one of these six categories.
It is to ask a better first question:
What consequential decision is your organization actually trying to make?
A confidential introductory conversation identifies the smallest responsible place to begin. No diagnostic preparation is required, and no predetermined service sequence is assumed.
About these cases
These vignettes are based solely on publicly available information and are used for educational and analytical illustration. They do not represent client engagements, endorsements, audits, assessments, certifications, or KOZCAN findings concerning the organizations discussed. Company circumstances may extend well beyond the public facts summarized here. KOZCAN findings are produced only within appropriately bounded advisory engagements using relevant evidence and the applicable method.
Public sources last reviewed: August 2026.