Early last year, we discovered our company, Shippo, had organically procured at least 19 GenAI tools across departments. No duplicates flagged. No shared learning. No accountability for outcomes. We had optimized for exploration and self-discovery, resulting in fragmentation and limited adoption.
That story is less unusual than it sounds. A July 2025 MIT study found that 95% of enterprise AI pilots deliver no measurable financial return. The issue is the learning gap inside organizations. OpenAI reached the same conclusion in its 2025 State of Enterprise AI report: “The primary constraints for organizations are no longer model performance or tooling, but organizational readiness and implementation.”
In our codebase, AI-assisted contributions rose from 30% to 60% between Q3 2025 and Q1 2026. The models improved meaningfully, but so did our internal enablement.
The companies that will capture the most value have done the organizational work. The urgency to become AI-native has never been higher.
How leaders can encourage adoption
There’s a gap between how leaders talk about AI and how employees experience it. Deloitte’s 2026 State of AI in the Enterprise found executive confidence rising, while worker sentiment remains mixed. Roughly 21% non-technical employees prefer not to use AI (but will do so if required), while 4% actively distrust it and avoid using AI. You can deploy every tool in the market and still have a trust problem.
In April 2025, Shopify CEO Tobi Lütke made AI usage a baseline expectation. What mattered was the signal: We’re all figuring this out together.
Culture change begins when leaders model learning and share their own AI experiments across the organization, which builds more trust than those who issue top-down policies. When a VP shares a failed prompt or a CTO admits a workflow misstep, it gives others permission to experiment.
One simple ritual to try out: A rotating two-minute “AI win and fail” at all-hands meetings. It normalizes experimentation and makes progress visible.
This applies at every level. Team leads can build AI into project kickoffs. Managers can pair heavy users with light users, track adoption by role, and create visible accountability for progress. Modeling and measurement matter.
Experimentation without direction is chaos
Deloitte found that access to sanctioned AI tools grew 50% in a year, reaching about 60% of those with access who use them regularly in their daily workflows. Meanwhile, 90% of workers are using personal AI tools for job tasks, with or without approval.
The shadow AI economy is real. Our 19-tool discovery was proof. The only tools that reached near-100% adoption had a named sponsor, an enablement plan, and a shared knowledge base.
Freedom to explore needs a frame. In 2025, our approach was permissive: experiment, learn, share. In 2026, we shifted from experimentation to value. New tools now require a champion, a defined use case, and a decision gate.
We’re also integrating AI agents directly into internal systems, so that siloed data becomes accessible in a single session. A product manager can now query our knowledge base, data warehouse, code repositories and support tickets simultaneously. Toil drops as decisions accelerate.
AI proficiency is also becoming part of hiring and performance evaluation. One of my favorite interview questions: “How did you use generative AI this week?” I’m listening for reflexivity.
Adoption requires protected time
Wharton’s 2025 AI Adoption Report identifies lack of training resources and broader human capital challenges as among the top barriers to AI adoption, even as 82% of leaders now use generative AI weekly. Knowing something matters isn’t the same as creating space to learn it.
Not everyone can experiment at night and on weekends. Companies that expect self-education on personal time will get the adoption curve they deserve.
The best investment we’ve made is protected time for experimentation. Our first AI hackathon produced solutions across every department. Marketing, with no engineers, built a custom AI trained on our company’s voice and style guide. Anyone can now draft personal, on-brand LinkedIn posts without waiting for marketing. This solution came from giving people a few days to build.
Two to three AI-focused hackathons per year feels right while tools evolve this quickly. In between: office hours, role-specific sessions, and open channels for sharing wins and failures. Don’t overlook vendors. The good ones run lunch-and-learns, share roadmaps transparently, and even join hackathons. It’s a fast way to distinguish substance from hype.
The bottleneck continues to move
Google’s 2025 DORA report on AI-assisted software development, surveying nearly 5,000 professionals, identified a critical tension: AI increases software development but also increases instability, because faster output exposes downstream weaknesses.
Previously adequate testing, code review, and deployment processes become bottlenecks when engineers write code twice as fast. The 2025 DORA report finds that while AI can improve individual productivity, those gains do not always translate into better overall system performance.
We’re seeing exactly that. As AI-assisted code contribution climbs from 30% to 60%, bottlenecks shift rather than disappear. Today it’s testing and review. Once those improve, constraints will move to requirements and design. That’s how compounding productivity works: Each solved constraint reveals the next.
The pattern extends beyond engineering. Marketing shifts from content-creation bottlenecks to brand governance. Operations move from data gathering to decision quality. Support moves from response speed to resolution depth. Acceleration surfaces weaknesses.
DORA’s other finding is even more important: “AI doesn’t fix a team; it amplifies what’s already there.” Strong teams get stronger. Teams with weak practices, poor documentation, or low trust see those problems accelerate alongside output.
AI adoption offers an organizational change initiative with extraordinarily powerful tools. Becoming truly AI-native means solving the human side first, trust gaps, learning culture and management habits that either enable adoption or quietly suppress it.
The companies that get that right will compound faster. And they’ll be the ones everyone else is trying to catch.
Geoff Smalling is the chief technology officer at Shippo.