While ‘vibe-coding’ an army of AI agents is pretty darn cool, it poses a significant risk for companies globally.
Finish him
In 2023, thousands of small and mid-sized firms quietly began wiring together “duct-tape AI” systems – lightweight automations built with no-code orchestration tools like n8n and Zapier. Many started as experiments, weekend prototypes meant to show that AI could replace routine workflows. But a year later, most of those experiments are gone.
The reason? Finishing turned out to be much harder than starting.
What happens when AI gets too easy to build?
According to one review of user deployments across the n8n ecosystem, 57 percent of automations failed before ever reaching production – not because the tools lacked capability, but because the underlying complexity outpaced what non-experts could maintain. (Source: n8n Community Data Report, 2024).
Now, with OpenAI’s Agent Builder lowering the barrier even further – “eight minutes to production,” according to OpenAI’s own demos – we’re about to see history repeat itself at scale. Tens of thousands of companies will start building autonomous agents without governance, engineering discipline, or lifecycle planning. For mid-sized businesses, that’s both an opportunity and a looming mess.
“Every time we democratize a technical capability, we also democratize the failure modes,” notes Forrester analyst Diego Lo Giudice, referencing similar patterns in the low-code and RPA booms of the past decade.
The analogy isn’t hypothetical. When Microsoft Power Automate rolled out its citizen-development push, adoption soared – yet most enterprises failed to turn that enthusiasm into sustainable results.
In 2021, Gartner estimated that fewer than 20 percent of Power Platform apps built by business users were actively maintained a year after launch, citing “lack of governance, documentation, and ownership” as primary causes. (Source: Gartner, Low-Code Platforms for Enterprise Applications, 2021).
Visual builders make automation look simple. They hide the architecture, error-handling, compliance, and evaluation layers that determine whether an agent can run safely in production. Once deployed, those same agents start encountering messy realities – data drift, conflicting logic, security constraints, or regulatory exceptions. Without oversight, they either fail silently or cause cascading errors that nobody understands how to unwind.
The pattern is almost formulaic:
Lower barriers > More projects start > More projects stall > More experts required.
It happened with AWS (“the cloud will replace infrastructure teams”), with Zapier (“everyone can automate their workflows”), and with Power Automate (“citizens can build their own apps”). Each democratization wave created a new market for experts – not to start projects, but to save them.
But this time it will be different… right…?
Early data suggests the same thing is happening with agents. Industry analysts forecast that up to 80 percent of organizations experimenting with AI agents will hit the ‘complexity wall’ between prototype and production, generating as much as $4 billion in remediation and consulting demand over the next three years. (Source: Emerj Research, The State of Agentic AI Adoption, 2025).
The problem isn’t enthusiasm – it’s architecture. Mid-sized firms often launch agents to speed up customer service or internal reporting, but their projects collapse when integration or compliance needs arise. Pre-built connectors cover only a fraction of real-world use cases. Domain-specific nuances – HIPAA in healthcare, SOX in finance, ISO 27001 in manufacturing – quickly turn prototypes into liabilities.
The lesson from the past decade of “citizen development” still applies: tools can democratize creation, but not accountability.
As one CIO told MIT Sloan Management Review, “The first 10,000 people who build agents this year won’t need help starting – they’ll need help finishing.”
The age of Agent Builders will create as many stranded projects as successful ones. The winners won’t just be those who automate quickly, but those who can recover what others built in haste – turning rogue agents into reliable, governed systems that actually work when it matters.
Tags:
Related content
What Cognitive Offloading Is, and How to Keep Your Team Sharp While Using AI
Cognitive offloading is the delegation of mental work to something outside your own head, and generative AI extends it from arithmetic and directions to analysis and judgment. What the research establishes, what it does not, and five practices that keep your team sharp.
WHITEPAPER: Designing the modern Accounting Firm with Agentic AI
Preparing your firm for the shift to an agentic operating model.
WHITEPAPER: The Map Before The Machine
How mapping the work before choosing the technology enables leaders to build an agentic operating model that delivers measurable returns.
Why Teams Stop Trusting AI: 5 Ways to Earn It Back
McKinsey found that 80% of respondents report personal productivity gains from AI, while 37% can point to any EBIT impact. Part of that distance is trust. Five patterns that break a team’s confidence in AI at work, and the design decision that rebuilds each one.