AI Sprawl Survey: When AI Tools Outpace SaaS

TL;DR: A 2025 AI sprawl survey by Zapier of more than 500 enterprise leaders found that the proliferation of AI tools is racing ahead of governance: 70% of enterprises haven’t moved past basic AI integration, over a quarter already run more than 10 AI apps, and only 35% put those tools through proper approval. The pattern matters because disconnected AI tools add a fresh layer of fragmentation on top of existing SaaS sprawl rather than clearing it. The fix is a connective layer, not a longer shopping list.

For a decade the complaint was SaaS sprawl: too many apps, none of them talking to each other. AI was supposed to cut through that. Instead, most organizations are stacking AI tools on top of the mess and calling it progress. The result is a second sprawl problem layered on the first.

The numbers tell the story. In Zapier’s AI sprawl survey, over 1 in 4 enterprises (28%) now use more than 10 different AI applications (Zapier, 2025). Yet 70% of enterprises still haven’t moved beyond basic integration for those tools (Zapier, 2025). More tools, same disconnection.

What is AI sprawl, and why is it different from SaaS sprawl?

AI sprawl is the unmanaged accumulation of AI tools, models, copilots, and agents that teams adopt faster than the company can track or govern them (GoGloby, 2026). SaaS sprawl gave you too many apps. AI sprawl gives you too many apps that also act on your data, often without anyone signing off.

That distinction is the whole problem. A dormant SaaS license is wasted money. An AI tool reads documents, drafts decisions, and increasingly takes actions on its own. When it isn’t connected or governed, the blast radius is wider.

And the proliferation is set to continue. Despite the integration struggles, 66% of enterprises plan to raise their AI tool count over the next year, against just 3% expecting it to fall (Zapier, 2025). The curve points up.

Why does adding more AI tools deepen fragmentation?

Each new tool is one more island. It holds its own context, its own data, its own slice of how the company works, and it shares none of it with the tools next to it. Add a tool and you’ve added an edge case, not a connection.

The survey makes the cost concrete. In total, 3 in 4 enterprises (76%) have experienced at least one negative outcome from disconnected AI (Zapier, 2025). Among the specifics:

• 34% of leaders say tool sprawl makes training employees on AI a major challenge (Zapier, 2025).
• 30% say they’re wasting money on redundant AI software (Zapier, 2025).
• 29% report that manual data transfers between AI tools are eating employees’ time (Zapier, 2025).

That last figure is the punchline. AI that was sold to remove busywork is generating it, because someone has to ferry data from one disconnected tool to the next. The promised time savings get spent on copy-paste.

The governance gap is the real risk

The faster half of this story is adoption. The slower half is control, and the gap between them is where the danger lives.

Only 35% of enterprise leaders say the AI tools used in their organization go through proper approval channels (Zapier, 2025). The other two-thirds are running AI that nobody formally cleared. 36% of leaders say AI sprawl is increasing security and privacy risks for their business (Zapier, 2025).

Some are pushing back. 38% conduct regular audits to catch shadow AI and 44% have started writing detailed AI usage policies (Zapier, 2025). Useful steps. But audits and policies govern tools you already have; they don’t connect them. The leaders themselves seem to know this: 90% say a central AI orchestration platform is critical or important for their business (Zapier, 2025).

The next wave raises the stakes again. Gartner expects 40% of enterprise applications to ship with task-specific AI agents by the end of 2026, up from under 5% in 2025 (GoGloby, 2026). Agents don’t just read data, they act on it. Sprawl that’s merely expensive today becomes operationally risky once the disconnected tools start making moves.

A connective layer beats a longer tool list

If the problem were storage or features, more tools would help. It isn’t. The problem is connection, so the answer has to be a layer that links what already exists.

A knowledge graph connects entities — people, documents, tools, projects — so a single query can traverse relationships across systems instead of stopping at each tool’s wall. Add AI search on top and both employees and AI agents can find and reason over knowledge that used to sit stranded in separate apps. This is the gap a unified semantic layer like SemanticOS is built to close: one queryable brain across the fragmented stack, rather than a fifteenth tool added to it.

Consider Vantage Health, a mid-size health insurer. Over eighteen months its teams adopted a meeting-notes copilot, two writing assistants, an AI helpdesk bot, and a forecasting model, none of them connected. A renewals analyst chasing last year’s exception for a client had to ask the helpdesk bot, dig through the copilot’s notes, and re-key figures from the forecasting tool by hand. Five AI tools, and still an afternoon lost to manual transfer. With a connected knowledge graph underneath, that answer is one query, and the answer carries its source. The tools didn’t shrink; the distance between them did.

Key takeaways

• AI sprawl is proliferation outrunning governance: 28% of enterprises run 10+ AI apps, but 70% are still stuck at basic integration (Zapier, 2025).
• Disconnected AI tools add a second layer of fragmentation on top of SaaS sprawl, and 76% of enterprises report a negative outcome from it (Zapier, 2025).
• The governance gap is real: only 35% of AI tools go through proper approval, and 36% of leaders say sprawl is raising security and privacy risk (Zapier, 2025).
• Embedded AI agents, heading to 40% of enterprise apps by the end of 2026, turn tolerable sprawl into operational risk (GoGloby, 2026).
• The fix is a connective layer, a knowledge graph plus AI search, that links existing tools instead of adding another tool to the pile.