Why the 'wildfire' of AI agent sprawl in the enterprise is a feature, not a bug

In the wake of watershed moments like the leaked Shopify memo, a singular directive has echoed through boardrooms: become an "AI-first" company, and do it yesterday. But this ambition has collided with a security-conscious fear of maintaining governance as tools stack up and a new generation of technical debt accumulates. Some leaders balk at visions of uncontrolled "agent sprawl," where employees create siloed, duplicative, and ungoverned automations, creating a crisis of orchestration.

But what if this fear is fundamentally misplaced? What if the sprawl that leaders are trying to prevent is actually the very sign of success they should be striving for? SaaS-sprawl from the previous tech hype-cycle did not become any companies' undoing. They simply cut back once the tech-overload became too costly and reliable software bundling solutions emerged.

According to Philip Lakin, Head of Enterprise Innovation at Zapier, the key to unlocking AI's true potential lies in embracing this controlled chaos. A practitioner with deep experience in operations and systems architecture at high-growth companies like Compass, Lakin joined Zapier after it acquired his automation management company, NoCodeOps. He argues that in the new AI-driven landscape, the old playbooks for technology adoption are obsolete.

• Blurred product lines: "The question for most companies used to be: 'Do we buy a point solution that's super highly opinionated and somewhat customizable, but very expensive?', 'Do we build from scratch with code?', or 'Do we build it from scratch with no-code/low-code?'. Now with agentic AI solutions, that whole conversation is just getting totally blurred."

The primary obstacle holding back enterprise AI adoption is the fear of what happens when it succeeds too well. Leaders worry that empowering every employee to build will create an unmanageable mess. But Lakin believes this perspective is the single biggest mistake a company can make.

• A luxury problem: "Everyone thinks they're going to have to deal with the problem of AI sprawl, but they never get to because they fail to achieve adoption in the first place. If you have the luxury of that problem, congratulations. You're in the top 1% of companies. The other 99% are so busy fearing that problem that they don't do anything at all."

Lakin's advice is provocative: actively pursue the chaos of widespread adoption, because the alternative of widespread organizational apathy is a far more dangerous problem to have.

• Lighting the wildfire: "You want that to grow like wildfire across your company because you can always wrangle it in, but the harder part is getting people to care."

• Hands-on: "At Zapier, we get as many people hands-on as possible, not because we think they'll all become builders, but because once they understand the fundamentals, they become a sensor in their department. They're the ones who can say, 'That process is no longer acceptable. I might not be the person to build the solution, but I know we can do better.'"

This "wildfire" of adoption doesn't have to be reckless, however. The key to successful orchestration, Lakin says, is enabling this freedom within a framework of strong, centralized governance. It’s about balancing empowerment with control.

• Freedom with guardrails: "We advocate heavily for tools that have centralized governance controls," using the example of restricting writing data to Salesforce without explicit permission at the user level. These rules are not unique to agentic AI solutions, and have been in place since the dawn of cloud. With fundamental guardrails in place to manage risk, Lakin argues that companies can then focus on a far more strategic and transformational metric for success.

• The new ROI: "The real ROI is that CEOs want everyone at their company thinking AI-first. They want every employee asking, 'Can this whole process be reimagined with AI?' Because what does it cost your business to not have everybody thinking that way? That's a big cost."

This shift is already taking hold, even in historically slow-moving industries. Lakin notes that large hospital systems are now seeking automation, driven by a new generation of employees asking why they aren't using the same modern tools available elsewhere. It's a sign that the grassroots pressure is working, finally making good on a promise that was once a tough sell. "The promise of citizen development before the AI generation was a tough sell because the buy-in wasn't there from the top. I think now it is, and now it's more possible than it ever was before. True democratization is now here."