Warning: Free AI Could Seriously Damage Your Business

This piece revises Stage 1 of AI in Business: Costs, Benefits, and Phases of Adoption (April 2025). Six months on, the lesson is clear: the real risk lies not in missing out on AI, but in trusting the free version.

Back in April, just six months ago, I outlined four stages of AI adoption in SMEs. Stage 1 was when employees begin using AI spontaneously, without direction from management. I wrote—accurately at the time—that this phase would bring small productivity gains for individual users. Tasks like email drafting, content creation, and data entry might become more efficient, but the organisation as a whole would see no strategic improvement. On the cost side, I expected only mild frustration from employees working without structured support.

I now need to revise that view. Free AI prioritises engagement, not accountability—think conversation over reasoning; guidance without context or memory. The result is a steady accumulation of superficial or ill-judged decisions. To avoid this, businesses will need to invest early and face the more complex implications of using AI as a professional instrument rather than a public toy.

As central bankers start to warn of an imminent correction in tech stocks—pushed to extraordinary heights by the ongoing promises of AI—it’s worth unpacking why the investment remains so intense, and what those directing it expect in return. The answer lies partly in anticipation: if AI can lift global productivity to new levels, everyone wants to secure their claim in advance. “Agentic AI” is the latest expression of that hope, promising systems that can not only complete a task but string together a sequence of them with minimal human input.

The next question is how these platforms will actually make money. What are they selling? The pattern is familiar. On the consumer side, most large-scale investment in “agentic” or “general” AI is flowing into engagement platforms—chatbots, copilots, search assistants, media companions—whose revenue depends on usage hours, clicks, or subscriptions. Their optimisation target is retention, not reliability. . For businesses, the priorities are the reverse. They need reliability rooted in accuracy, continuity, and auditable reasoning. Analytical discipline doesn’t drive daily-active-user metrics, so it has no place in free platforms.

The underlying model for AI remains the same: consumers are seduced to stay engaged so their attention can be sold to suppliers, while businesses are learning they must now pay for discipline. My frustration on the ongoing manipulations of the consumer space by online platforms is well documented. That AI is following the same route is, unfortuately, not a surprise. Lets focus on the urgent need for business to get itself out of this productivity sapping trap by understanding the changes needed to bring AI in-house.

Business Grade AI
What makes an AI model “business-grade.” ? For a company or consultant, usefulness means:

• Context : remembering facts, decisions, and thresholds.
• Judgement: distinguishing between fixed data, assumptions, and hypotheses.
• Audit Trail: maintaining version control and traceable reasoning
• Specialised: training on structured, in-house data rather than generic web content

Achieving all this is not cheap, and management and boards will need to weigh the investment against the risk of watching a startup—or a large competitor—move faster.

The End of the Beginning

There is good news and bad news. The good news is that the path forward is now clearer. The bad news is that there is no longer any “free” or even cheap Phase 1. Yes, employees can and will draft their emails and reports, but when they grow more ambitious and use personal AI accounts to get more done, the organisation risks accumulating suboptimal decisions. The greater danger is standing still while others act.

Serious firms are already building internal layers on top of base models. Finance houses and law firms are developing “governed LLMs” that treat the model as a reasoning engine plugged into verified data stores. Engineering and consulting firms are building knowledge agents that operate under strict rule sets and leave audit trails.

Turning an LLM into a dependable professional instrument requires owning or licensing a base model and retraining it around commercial criteria, with metrics of correctness and continuity rather than engagement. Help is available. Small startups are already mobilising to help SMEs get started—focusing on memory architectures and belief-tracking, the missing pieces of commercial reliability. Money is being invested in AI as a useful business tool. It is smaller, less public, and less spectacular than the agentic hype, but it is where real value will lie.

In summary, free AI is evolving toward entertainment: fluency, warmth, and accessibility. Business AI can no longer rely on this safely. It needs to be brought in-house to deliver precision, continuity, and liability management.