Canva cofounder and chief product officer Cameron Adams disclosed in a July 15 interview with Rapid Response that the design platform gives its more than 5,300 employees the freedom to choose their own AI tools, supported by individual budgets and dedicated time away from regular duties to experiment. Instead of anointing a single assistant like ChatGPT, Claude, or Gemini, the company tells staff: “You can figure it out.”
The announcement, covered by Business Insider, lands just as enterprises grapple with AI adoption rifts—some mandating tools, others discovering rampant shadow IT. But Canva’s model isn’t anarchy. It offers a governed, funded framework that lets teams tailor AI to their workflows while IT maintains control over identity, data, and procurement. For Windows shops weighing Microsoft 365 Copilot as a standard, it’s a concrete case study in moving from “thou shalt use this” to “here’s a budget and a week to learn what works.”
What Canva Actually Did
Adams made clear the policy is deliberate: “We’re not mandating Claude or ChatGPT or Gemini or whatever tool you want to use in your part of the company,” he told Rapid Response. “You can figure it out.” Each employee or team receives a budget to test services, build workflows, and solve real problems. The goal is to foster an experimental mindset rather than begrudging compliance that kills curiosity.
“If you force a tool on them, they’re just going to do that very begrudgingly, and they’re not going to enter into this very experimental mindset that we need them to,” Adams said. That philosophy shapes every aspect of Canva’s AI adoption strategy.
The flagship event illustrating this approach is AI Discovery Week. During this company-wide pause, Canva tells its 5,300 employees to stop normal work entirely. Instead, they attend sessions, try tools they’ve heard about, and tackle genuine pain points. The company’s second such week included a hackathon, surfacing practical applications across departments. It’s a stark departure from the “learn this between meetings” approach that often dooms enterprise AI rollouts.
But this isn’t a free-for-all. Canva has a supported set of enterprise-grade AI services—including offerings from Anthropic, Google, and OpenAI—ensuring that experimentation happens inside vetted, secure environments. Employees aren’t creating random accounts on consumer-grade chatbots and feeding them sensitive data. The company has described guardrails that keep usage within approved boundaries while still granting wide latitude for tool choice.
Why This Matters for Windows and Microsoft 365 Shops
The rise of Microsoft 365 Copilot has created a default path for many Windows-centric organizations: adopt the assistant built into Office apps, expect universal adoption, and measure success by license counts and prompt tallies. But Canva’s experience suggests that adoption metrics alone don’t signal whether AI is genuinely making people better at their jobs. A single approved assistant, no matter how deeply integrated, may not fit every workload.
For IT leaders, the practical challenge isn’t Copilot versus a zoo of consumer AI. It’s how to build a policy that lets teams evaluate alternatives—whether that’s a coding-specific assistant, a contract-analysis tool, or a design copilot—while keeping the organization’s security and compliance posture intact. Managed freedom, not unilateral mandate. Shadow AI already runs rampant: a 2024 Cyberhaven report found that 11% of data employees paste into ChatGPT is confidential. Microsoft’s own research indicates over 78% of AI users bring their own tools to work. Ignoring that reality in favor of a single sanctioned tool just pushes usage underground.
Several specific practices emerge from the Canva playbook:
- Funded exploration: Give teams real budgets to test AI tools, not just a verbal permission. This moves evaluation from a side project to a sanctioned activity.
- Protected learning time: Dedicate whole days or weeks to AI experimentation, free from regular task pressure. Canva’s week-long pause cost productivity in the short term but aimed to spark lasting gains.
- Approved ecosystem, not a single product: Define a core set of vendors and enterprise plans that meet security standards. Allow choice within that circle.
- Outcome-based metrics: Steer clear of tracking mere prompt volume or token consumption. Ask: did the tool help a team produce higher-quality work faster?
- Identity, data, and endpoint controls: Loop in Microsoft Entra, data loss prevention, and device management to enforce boundaries across all AI services, not just one.
How the AI Governance Conundrum Evolved
The pressure to adopt AI has pushed many companies into extreme positions. Duolingo tied employee AI usage to performance reviews. JPMorgan and Disney created internal leaderboards ranking staff by AI token consumption—a strategy that can incentivize volume over value. Other firms, like Coinbase, instead advise using different models for different tasks to control costs, with CEO Brian Armstrong even recommending cheaper Chinese models as defaults for simple queries.
Meanwhile, shadow AI has become the new shadow IT. Employees, frustrated by slow corporate rollouts, sign up for personal ChatGPT or Claude accounts and paste in sensitive data. Traditional blocking tactics fail because determined users find ways around them. Canva’s approach is a direct response to that tension. By giving employees sanctioned flexibility, the company channels experimentation into monitored channels.
The company has skin in the game. Canva launched Canva AI 2.0 in April, a conversational design platform that turns prompts into visuals. Business Insider found its output comparable to Claude Design in a head-to-head slide-deck test. That Adams openly endorses non-Canva tools even as his company pushes its own AI signals a genuine belief in the multi-tool future. It also builds credibility with staff who might otherwise suspect a vendor lock-in agenda.
What IT Leaders Should Do Now
Adopting Canva’s model doesn’t mean ripping out your Copilot deployment. Instead, use these steps to layer governed flexibility on top of your existing stack:
-
Audit current AI usage
Use your identity provider and network logs to see which AI services employees are already accessing. Microsoft 365 admins can leverage Defender for Cloud Apps to spot shadow AI. Understand the footprint before you try to reshape it. -
Define an approved AI catalog
Select a handful of enterprise-grade services that meet your security, privacy, and residency requirements. Offer choices that map to common job functions: a general assistant, a code helper, a design tool, a data analysis platform. Publish clear documentation on which tool suits which task and why. -
Allocate dedicated funds and time
Assign a small budget per team or individual for AI tool evaluation. More importantly, block calendar time—a day, a week, or recurring Friday afternoons—where normal work stops and people experiment. One European manufacturer that trialed a quarterly “AI Day” found that 70% of participants identified at least one process to automate. -
Wrap everything in identity and data controls
Require sign-in through Microsoft Entra ID across all approved AI tools. Enforce conditional access policies that block untrusted devices. Use Microsoft Purview or equivalent to classify and protect data before it enters an AI service. Prohibit sensitive data in prompt fields, period. -
Measure what matters
Replace token counts with before-and-after assessments of task speed, quality, and employee satisfaction. Ask teams to report concrete use cases where AI made a difference, not just activity metrics. Use that evidence to decide which tools earn a permanent budget. -
Involve the business, not just IT
Canva’s program was driven from the product chief and affected every department. Avoid the trap of IT unilaterally selecting an AI stack and then bemoaning low adoption. Form a council with representatives from each major business unit to review tool requests and share success stories.
What to Watch Next
Canva’s experiment is a live demonstration that the “one assistant to rule them all” era may be short-lived. As models specialize and price points diverge, the pragmatic path is likely a curated portfolio, not a monolith. Microsoft will almost certainly respond by making Copilot more extensible—offering plugins, custom GPTs, and connectors that let it serve as a front-end to multiple models, thereby keeping the identity and data plane unified. Already, Microsoft 365 Copilot supports extensibility with first- and third-party plugins, but the AI discovery week concept remains rare.
For Windows IT pros, the immediate watchpoint is how your preferred endpoint management and security tools evolve to handle multi-AI environments. Expect more granular policies that allow an admin to say: “Only allow enterprise ChatGPT and Claude accounts, block consumer ones, and require DLP inspection on prompts.” The future isn’t about stopping AI; it’s about steering it toward productive and safe channels.
Canva’s bet is that giving employees agency fosters genuine, self-sustaining AI fluency. Whether that pans out will be measured not in token counts but in whether the company’s next product innovations emerge from those hackathon projects. It’s a wager other enterprises might be wise to replicate—on a smaller scale, at first.