The gap between the AI adoption story companies tell publicly and what is actually happening inside their teams is significant. Most enterprises are not AI-first. They are AI-patched. A few workflows got automated, most of the organization kept working the same way. Here is what genuine AI adoption looks like at the team level.
When a major enterprise puts out a press release about "becoming an AI-first company," the gap between that statement and what is happening at the team level is usually wide.
The leadership team has committed to AI. The procurement team has signed licenses. A few teams are running pilots.
The rest of the organization is mostly doing the same work they were doing eighteen months ago, with some people using AI tools informally and some not using them at all.
This is the normal state of enterprise AI adoption in 2026. Not because organizations are resistant to AI - most are genuinely trying. But because the distance between an executive AI commitment and actual team-level behavior change is longer than the announcement suggests.
What does genuine AI adoption look like at the team level? I have enough visibility into what is happening at enterprise clients in my experience, and what worked and did not at Zendesk and Adobe through previous transitions, to give a grounded answer.
The Honest State of Enterprise AI Adoption
Per McKinsey's 2025 State of AI report, 65% of organizations are using AI in at least one business function. That is a real number. But "using AI in at least one function" is a low bar.
The more useful question is what percentage of work at the team level is meaningfully changed by AI. That number is much smaller. Most enterprise AI adoption sits in a pattern:
A handful of use cases are fully operationalized. Document processing, meeting summarization, first-draft generation, support ticket categorization. These are working and producing real efficiency gains.
A larger set of use cases is in pilot. Teams are experimenting, results vary, operationalization has not happened.
The majority of work is unchanged. The people, the processes, the outputs are roughly the same as pre-AI, with individual employees using AI tools informally and inconsistently.
The organizations that are actually ahead are not the ones that have signed the biggest AI contracts. They are the ones that have operationalized the right use cases at the team level and built the workflows to make AI adoption consistent rather than individual.
What the Best Enterprise AI Implementations Actually Look Like
The teams doing this well share some consistent characteristics that are worth examining.
They started with workflow, not with tool. The question was not "what can we do with this AI tool?" It was "what are the highest-value, highest-friction parts of our team's workflow, and which of those is AI well-suited to help with?" (For a deeper dive into this, see AI workflow automation for business: what is real, what is hype).
That inversion matters. Tool-first adoption produces a collection of individual use cases that do not compound. Workflow-first adoption produces system changes that affect how the team works as a unit.
They built shared norms around AI use, not individual ones. The teams with the highest AI adoption rates have explicit team-level agreements about when AI is used, for what, with what quality checks. These are not corporate policies.
They are working norms developed by the team based on what they have found actually works.
At Zendesk, when our team started using AI-assisted drafting for client communications, the adoption was initially inconsistent. Different people used it at different points, with different levels of review, with different results. The improvement came when we established a shared norm: AI for first draft, human for tone calibration and strategic framing, peer review before anything client-facing.
Not a policy - a team practice that we developed and refined over about six weeks.
They measured the right things from the start. The teams that are ahead are not just measuring efficiency (time saved, volume processed). They are measuring quality outcomes - whether the AI-assisted output is as good as the human-only output across the metrics that actually matter.
When quality declines, they know it before it becomes a problem.
Where Enterprise AI Is Producing Real Value by Function
Being specific here, because the generic "AI is transforming business" frame is not actionable.
Marketing and content teams: AI is producing real value in first-draft generation, content repurposing, SEO analysis, and campaign performance reporting. The quality of the output requires human editorial judgment - AI content without human refinement is visible and underperforms. But the speed gains in the ideation and first-draft phase are genuine and significant.
Customer service and support: AI value is concentrated in tier 1 resolution for well-defined issue types and in agent assistance tools that help human agents respond faster and more consistently. (I wrote more extensively on this in AI for customer service: the honest version in 2026). The highest-value implementations use AI to speed human agents rather than to replace them at complex interactions.
Sales and revenue operations: AI is producing value in call analysis and coaching, in CRM data hygiene and enrichment, in first-draft proposal sections for well-established content types, and in pipeline reporting and forecasting. The judgment layer on deals and client relationships remains human.
Finance and operations: AI is doing well in invoice processing, expense categorization, financial report summarization, and data reconciliation for structured datasets. Complex financial analysis and exception handling remain human.
Legal and compliance: AI is being used for contract review (first pass, flagging non-standard clauses), regulatory monitoring and summarization, and policy document drafting. Final review and judgment on complex matters remains human. This is an area where the risk of AI error is high enough that most organizations are being appropriately cautious.
The Adoption Barriers That Are Not Getting Enough Attention
Most discussion of enterprise AI adoption focuses on technology barriers. The real barriers tend to be organizational.
The quality assurance gap. Organizations have not built systematic quality review processes for AI output. Individual users are reviewing their own AI output with varying levels of rigor.
The result is inconsistent quality that is hard to detect before it creates a problem.
The skill gap in output judgment. Knowing how to use an AI tool is not the same as knowing how to assess whether the AI's output is good. Most employees have not been trained to be effective editors and judges of AI output.
They accept it when it sounds plausible, even when it is subtly wrong.
The integration gap. AI tools that do not integrate with the team's existing workflow get used inconsistently. The tools that stick are the ones where using AI is the easiest path through a task, not the ones that require a context switch.
The accountability gap. When AI output causes a problem - a client communication goes out with an error, a report contains a wrong figure - it is often unclear who is accountable. The organization tends to treat the AI as the source of error and the employee as a victim of it.
This distributes accountability away from the quality review process, which is where the failure actually occurred.
What Leaders Should Actually Do About Enterprise AI Adoption
Stop measuring AI adoption by license usage. Most enterprise AI platforms report on license activation and feature engagement. These measure whether people have the tool, not whether they are using it in ways that produce value.
Measure workflow change. Which workflows have actually changed at the team level? What is the before-and-after picture on quality, speed, and team capacity?
That is the adoption metric that matters.
Invest in output quality as much as tool access. AI tool access without quality review training produces the risks without the benefits. The most underinvested capability in most enterprise AI programs is teaching employees to be effective editors and judges of AI output.
Build team-level norms, not just corporate policies. Corporate AI policies tell employees what they are allowed to do. Team-level norms tell them how to actually work with AI on the specific tasks their team does.
The policies can be set centrally. The norms need to develop within each team, with enough structure to be consistent and enough flexibility to fit the actual work.
Identify the ten to twenty use cases where AI produces the highest value in your specific organization and operationalize those completely before expanding. Broad shallow adoption produces less value than narrow deep adoption in the early stages.
