AI Dependence and the Risks for Businesses
- 5 days ago
- 9 min read
AI doesn’t replace a team. It creates a dependency that must be managed.
AI dependence is a risk that many organizations still underestimate. Over the past few years, the prevailing narrative has been that employees should be given access to AI tools so they can become more productive, or so companies can maintain the same output with fewer people.
On paper, that may seem logical. In practice, however, this view overlooks an important reality: AI isn't a stable, predictable tool. It's a constantly evolving system that depends on external providers and can be restricted, modified, or degraded at any time, often with little control on the user's part.
The Productivity Promised by AI Is More Complex Than It Seems
The productivity gains often associated with AI don't hold up particularly well under intensive, real-world use. I have a strong technical background, I’m highly resourceful, and I’ve been using AI daily for roughly four years. Even so, I'm not twice as productive.
Not even close.
Sure, certain tasks can be completed faster. AI helps with writing, structuring information, coding, analysis, rewriting, and accelerating parts of a workflow. But those gains never come for free.
Outputs still need to be reviewed, validated, corrected, fact-checked, and assessed for subtle errors or flawed assumptions. Productivity only increases when the user retains enough expertise to properly evaluate the results. Otherwise, no real time is saved. The problem is simply shifted elsewhere in the process.
AI Tool Quality Can Degrade Without Warning
AI tools don't maintain a constant level of quality, and documented cases confirm this reality. Anyone who uses these systems extensively eventually notices it: sometimes the tool performs less effectively, understands prompts less accurately, or produces outputs that appear convincing at first glance but require substantial correction afterward.
Anthropic acknowledged this issue in a report published in April 2026. The company investigated user reports claiming that Claude's performance had deteriorated on certain tasks and traced the problems to three separate changes:
a modification to the model's reasoning effort level,
a bug introduced in a cache optimization system, and
a change to the system prompt.
This is where the core issue lies.
AI can be extremely useful, but it also introduces instability, dependency, and an ongoing validation burden that many organizations still underestimate.
In an article published in April, I documented the practical realities of usage limits, token consumption, and the performance of paid AI tools, all of which remain largely opaque to users.
AI Is Closer to a Personal Assistant Than a Traditional Tool
AI isn't a passive tool like a drill or a spreadsheet. Many business leaders misunderstand this distinction. In practice, working with AI often resembles working with another person. You need to provide context, explain the objective, monitor the work being done, correct mistakes, and review the final output.
In construction, a drill increases productivity because it does exactly what it's designed to do. It drills. It doesn't suddenly decide to do something else. Now imagine a drill that occasionally becomes less effective, sometimes stops working, occasionally drills in the wrong place, and requires the worker to inspect every hole afterward.
At that point, it's no longer simply a productivity tool. It has become something that requires supervision, adding an entirely new layer of work to the process.
Replacing Employees with AI Too Quickly Carries Real Risks
AI doesn't fully replace human workers, especially in tasks that require judgment, accountability, and consistent quality. Yet many layoffs have occurred, sometimes openly justified by the arrival of AI and sometimes framed that way after the fact. Some organizations genuinely believed that AI would be capable of replacing a significant portion of human work.
The reality is more nuanced.
AI can generate output, but it does not bear responsibility for the outcome. It can accelerate workflows, but it can also slow them down when mistakes go unnoticed or require extensive correction. What initially appears to be a cost saving can quickly disappear if employees must spend substantial time reviewing, supervising, correcting, or redoing the work.
Human expertise remains essential, not only to perform the work itself, but also to evaluate whether the result is accurate, appropriate, and reliable. That validation layer cannot simply be eliminated. In many cases, it becomes even more important as AI adoption increases.
The Cost of AI in Businesses Is Widely Underestimated
AI isn't free, and its true cost extends far beyond what many organizations initially anticipated. At first, many businesses viewed AI as an almost all-you-can-use service: pay a monthly subscription, purchase a few licences, and enjoy higher productivity at a relatively low cost.
Reality has proven to be far more complex.
The fundamental unit behind AI is the token, and every token carries a computational, energy, and financial cost. The more AI is used, the more tokens are consumed. The longer, more complex, and more frequent the prompts, the higher the cost becomes. AI relies on a massive infrastructure of servers, electricity, computing power, and capital investment.
Bryan Catanzaro, Vice President of Applied Deep Learning at NVIDIA, has acknowledged that, for his team, the computational cost of AI exceeds the cost of employees by a significant margin. In other words, AI is not automatically cheaper than human labour. In some situations, it may actually cost more than the people it is intended to replace.
As a result, the economic argument for replacing employees with AI is far less compelling than it first appears. Organizations that eliminate positions too quickly may simply be exchanging a relatively predictable labour cost for a variable technology expense that is difficult to control and heavily dependent on external vendors.
Quebec newspaper La Presse illustrates the scale of the trend. According to Sundar Pichai, customer token consumption on Gemini is now 300 times higher than it was in the summer of 2024. In total, Gemini reportedly processed 3.2 quadrillion tokens in a single month.
Meanwhile, research firm Gartner predicts that even if token prices fall by 90% by 2030, overall spending on AI will continue to rise because the volume and complexity of AI usage are growing faster than costs are declining. As Montreal-based expert Jean-Luc Sanscartier observed:
“Companies are using AI extensively, but their budgets were built around the monthly subscription cost of web applications, not around actual usage volume.”
Sanscartier also predicts that some organizations may eventually rehire entry-level employees for basic tasks, as their labour costs could ultimately prove lower than the cost of performing the same work through AI.
Dependence on an AI Provider Is Becoming a Major Business Risk
Replacing internal expertise with dependence on an external AI provider is one of the most significant risks emerging from the AI transition.
When an organization sheds human expertise too quickly under the assumption that an AI system or external provider will fill the gap, it becomes increasingly vulnerable. It's no longer relying on capabilities it controls internally. Instead, it becomes dependent on a provider that can raise prices, change usage limits, alter output quality, impose new terms, or shift its products and priorities in directions that may not align with the needs of its customers.
This dependency creates a new form of operational risk. The more deeply AI becomes embedded in critical workflows, the greater the potential impact of changes that are entirely outside the organization's control.
Putting all your eggs in one technological basket has never been a sound strategy. That is even more true when the basket belongs to an external company that ultimately controls the rules of the game.
AI Ecosystems Are Likely to Become Increasingly Closed
Big Tech AI providers are deliberately building ecosystems that become increasingly difficult to leave. OpenAI recently launched its deployment-focused business unit, backed by billions of dollars in investment, to help organizations integrate AI into their operations. The company has also expanded its consulting and deployment capabilities through acquisitions and strategic hiring, adding engineers and implementation specialists to support enterprise adoption from the outset.
Anthropic is pursuing a similar path, with initiatives aimed at embedding AI more deeply into business processes and workflows. When a company receives support from an AI provider to redesign processes, connect internal data, build automations, and train employees, it is no longer simply purchasing a software tool. Over time, it becomes dependent on a particular way of working.
The deeper the integration, the more expensive and disruptive it becomes to switch providers.
This dynamic isn't new. We've already seen it with large technology ecosystems such as Microsoft 365, where users are encouraged to operate within a comprehensive environment that becomes increasingly difficult to leave as more systems, data, and workflows become interconnected. AI providers are following a similar playbook. The goal is not merely to sell access to a model, but to become embedded in the daily operations of the organizations that rely on it.
Keeping Human and Technical Control to Preserve Autonomy
AI should remain a lever, not become a crutch. I remain strongly in favour of using AI, but not at the expense of human expertise and not at the cost of becoming completely dependent on a single provider.
Organizations should preserve internal knowledge, continue developing employee skills, and avoid building their entire productivity strategy around one AI platform. A business should be able to benefit from AI while retaining the flexibility to switch tools when necessary. The goal is to leverage AI without losing control over work processes, data, spending, or critical capabilities.
This reality may encourage some organizations to explore self-hosted or internally managed AI solutions. Such an approach comes with its own challenges, including infrastructure costs, hardware maintenance, cybersecurity requirements, and the need for specialized technical expertise.
Even so, many organizations may ultimately view these trade-offs as worthwhile if they provide greater control over critical systems and reduce dependence on external providers.
The companies that benefit most from AI in the long run are unlikely to be those that surrender control to it. They will be the ones that use AI strategically while preserving the human expertise, technical flexibility, and organizational resilience needed to remain in control of their future.
AI in Support of Human Expertise, Not as a Replacement for It
Businesses that will benefit most from AI won't be the ones that replace employees with technology. They will be the ones that recognize AI as a collaborative force rather than a complete substitute for human intelligence. AI can be extremely helpful, but it works best when it supports capable people. The moment an organization loses sight of that principle, it does not truly become more efficient. Instead, it becomes increasingly dependent on a system it doesn't fully control.
Human judgment, experience, accountability, creativity, and domain expertise remain essential. AI can amplify these strengths, but it cannot replace them entirely.
The most successful organizations will be those that combine the speed and scale of AI with the critical thinking and adaptability of their people. Rather than viewing AI as a replacement strategy, they will treat it as a capability-enhancing strategy. To learn more about building a thoughtful and sustainable approach to AI adoption, visit my website or contact me at info@jimmygilbert.com
Jimmy Gilbert, Technology Consultant and AI Trainer . A computer science instructor at Cégep de Sainte-Foy, Jimmy works at the intersection of education, software development, and artificial intelligence. He trains the next generation by combining technical rigour with a practical understanding of the job market, while actively integrating AI tools into his lectures. Jimmy also works as a technology coach, helping individuals and professionals effectively adopt technology, automation, and AI.
FAQ About AI Dependence in Businesses
Can AI really replace employees in a business?
AI can accelerate certain tasks, but it does not fully replace people. It does not assume responsibility, cannot guarantee consistent quality, and still produces outputs that require human review and validation. Organizations that rely on rapid workforce replacement often underestimate the time, expertise, and oversight needed to evaluate, correct, and manage AI-generated work.
What is the true cost of AI for a business?
The real cost of AI extends far beyond monthly software subscriptions. Every prompt consumes tokens, which carry computational, energy, and financial costs. According to Bryan Catanzaro of NVIDIA, the computational cost of AI already exceeds labour costs for some teams. Businesses must also account for the ongoing costs of validation, quality control, training, governance, and workflow integration.
What is vendor lock-in risk with AI?
Vendor lock-in occurs when an organization builds its processes, automations, data connections, and employee training around a single AI provider. If that provider changes its pricing, usage limits, policies, or product quality, the organization may find itself vulnerable and facing significant switching costs. Similar dynamics have long existed in software ecosystems such as Microsoft 365, where deep integration can make migration difficult and expensive.
Is the quality of AI output always reliable?
No. The quality of AI systems can fluctuate without warning. Anthropic itself published a post-mortem acknowledging that Claude's performance had degraded for certain tasks due to a series of technical changes. More broadly, AI systems can produce inaccuracies, inconsistencies, and hallucinations as a result of how large language models operate. Human oversight remains essential.
How can a business use AI without becoming dependent on it?
Organizations can reduce AI dependence by maintaining internal expertise, investing in employee training, diversifying tools and providers, and avoiding reliance on a single AI ecosystem for critical operations. For highly sensitive or business-critical workflows, some organizations may also evaluate self-hosted or privately managed AI solutions. The objective is to benefit from AI while preserving flexibility, resilience, and control.
