QBS Fall Series – Article 2 of 4
Somewhere in your organization this month, a 2027 plan is being written. It almost certainly includes a list of Artificial Intelligence (AI) initiatives and a line for AI licenses. Now look for what it leaves out: a line for governance and a line for training. Those two lines decide whether the rest of the plan survives contact with real work.
The first article in this series stressed that the reinvention of the knowledge worker depends on purposeful AI access and training, because employee capabilities do not arrive with the license. This article creates a process for ensuring successful AI implementation at the organizational, team, and individual levels.
Quantum Bridge Solutions (QBS) recommends five areas of focus and three implementation moves to make before advancing your AI plan for 2027.
1. The Ferrari problem
Handing powerful AI applications to employees who have never been taught to use it is like handing a Ferrari to a sixteen-year-old driver and skipping the driving lessons. The car is not the problem. And a Ferrari burns fuel at a rate a new driver never notices.
QBS has seen various challenges occurring this fall. Organizations that launched ambitious AI work this year are now shutting projects down, rolling them back, and/or facing AI operating expenses exceeding budgets. Employees have rarely been taught which model fits which job, so they reached for the most powerful one every time. Further, employees often did not know how to correctly access or build with the AI applications. Then organizational leadership expresses concern with AI work because they are not recognizing the return on the AI investment.
In a survey of 48,340 people across 47 countries, only two in five employees said their organization had any policy on generative AI, and 70% were working in free public tools rather than sanctioned ones1. In one large test, 45% of AI generated code samples introduced well known security vulnerabilities2. Documented AI incidents rose from 233 to 362 in a single year, even as more organizations adopted responsible AI policies3. AI use policies are a must for every organization, and their effectiveness is limited when it does not define employee guidance and training.
2. Every question runs a meter
The license is only the entry fee. Most AI use is billed by the token, the small unit of text a model reads and writes. Every question and every revision adds to the bill. The most powerful models charge the most per token4. A clear question sent to the right model costs little. A vague one sent to the biggest model and revised a dozen times costs more for the same result.
The token bill is where training meets the budget. H-AI-H, the constant in this series, runs human inquiry, AI processing, human empowerment. Its first H, a structured human question, is also the cheapest step in the work. Employees trained to frame the question and choose the correct AI model get more from the same dollars, so the organization’s talent and budget can go further together. Untrained use spends both.
3. The deadlock inside the plan
This is where most plans stall. You cannot train employees who have no access; you cannot give access to employees who are not trained. So training waits on IT, IT waits on training, and the initiative launches anyway.
Governance breaks the deadlock when its policies and procedures include ‘targeted AI use training’. At QBS, we make the training level the access credential. As employees demonstrate competency, their access to tools, systems and data widens through the identity and access. The research agrees that recognizing AI literacy has been the strongest single predictor of whether people use AI and whether they use it with critical judgment, stronger than policy or governance structures5.
4. What is worth building?
When anyone can build anything, the plan has to ask a harder question: what is worth building at all, who will maintain it, and who will use it? Skip those questions and you pay for the answers anyway, in abandoned tools, code nobody owns, and tokens spent on both.
Among U.S. manufacturers adopting AI, roughly a third of the short run losses came from abandoning structured management practices under the pressure of the change6. Being disciplined in the H-AI-H framework using structured Human inquiry, AI processing and Human empowerment is a must for all AI-assisted work.
5. The last ‘H’ belongs in the plan
In a 2027 plan, the final ‘H’ needs to include a named employee who reviews usage, cost and output before anything ships, and who can say whether it was worth it. Naming that employee costs almost nothing. Not naming them is how the surprise bill arrives.
Three moves before the plan is approved
None of these require new money.
- Add the two missing lines. Put governance and training beside the initiatives they protect, with one owner for both.
- Tie access to competency. Use the identity system you already run, ensuring demonstrated skill, not job title, opens the next tools and data. The federal AI risk framework already expects employees to be trained for their AI responsibilities7.
- Set a cost threshold that starts a conversation, not a cutoff. Teach employees to choose the right model for the job and read the token bill as a training report.
What comes next
A plan can fund individual capability and still produce a team that builds nothing together. The next article takes up why capable individuals do not add up to a capable team.
References
- Gillespie, N., Lockey, S., Ward, T., Macdade, A., & Hassed, G. (2025). Trust, attitudes and use of artificial intelligence: A global study 2025. The University of Melbourne & KPMG International. https://doi.org/10.26188/28822919 ↩︎
- Veracode. (2025). 2025 GenAI code security report. ↩︎
- Stanford Institute for Human-Centered Artificial Intelligence. (2026). Artificial intelligence index report 2026, Chapter 3: Responsible AI. ↩︎
- Anthropic. (2026). Pricing [Claude Platform documentation]. Retrieved October 5, 2026, from https://platform.claude.com/docs/en/about-claude/pricing ↩︎
- Gillespie, N., Lockey, S., Ward, T., Macdade, A., & Hassed, G. (2025). Trust, attitudes and use of artificial intelligence: A global study 2025. The University of Melbourne & KPMG International. https://doi.org/10.26188/28822919 ↩︎
- McElheran, K., Yang, M.-J., Kroff, Z., & Brynjolfsson, E. (2025). The rise of industrial AI in America: Microfoundations of the productivity J-curve(s) (CES Working Paper No. 25-27). U.S. Census Bureau, Center for Economic Studies. ↩︎
- National Institute of Standards and Technology. (2023). Artificial intelligence risk management framework (AI RMF 1.0) (NIST AI 100-1). U.S. Department of Commerce. ↩︎
QBS Sources behind this article
- QBS Governance Foundation (2026)
- Governance People Actually Use (QBS, 2026)
- QBS Vibe Coding Foundation (v2, 2026)
- From Lead Doer to Lead Designer: Research Foundation (QBS, 2026)
- QBS H-AI-H Decision-Making Framework (v4, May 2026)
