One of you sent me a job posting last week that listed AI tool access under “Benefits & Perks,” right between unlimited PTO and a cold brew tap. As if compute were a flavour of kombucha.
Your company offers free sushi. Your competitor offers ten million tokens a month. One of these attracts senior engineers.
That framing treats a work tool as an optional benefit. When an organisation files AI access under perks, it is telling the market it hasn’t yet realised this is a tools conversation — like a 2004 job posting bragging that employees receive a laptop. A candidate who relies on AI needs to know whether the job supplies the tools to use it. Daily AI users report 92 percent productivity gains and earn salaries 52 percent higher than their infrequent-user peers — a gap that reflects both the value of AI fluency and the seniority of early adopters. And the access gap is not random: 78 percent of leaders use AI regularly, while only 51 percent of frontline employees do.
That is not a perk distribution. It is a ceiling.
The Executive Luxury Playbook
Work tools have made this journey before.
In 1983, the Motorola DynaTAC cost $3,995 and sat on Gordon Gekko’s desk. By 1999 the BlackBerry meant you were important enough for mobile email. By 2007 the iPhone made the whole question irrelevant. Laptops traced the same arc — executive privilege, then road-warrior necessity, then standard issue. In the late 1990s, many companies metered internet access by seniority. An analyst who wanted to check a competitor’s website needed a manager’s approval.
Every wave had its own risk profile. The internet brought malware, phishing, and the productivity panic of employees browsing ESPN at lunch. Mobile phones created always-on work culture and security nightmares that took a decade to govern. Nobody argued that the solution was to keep the internet away from junior staff. They built firewalls, acceptable-use policies, and training — governance, not denial.
AI adds a new dimension: it generates confident wrong answers, which no prior workplace technology did. That demands new governance — review workflows, output validation, domain-specific guardrails — but it does not change the trajectory. Different risks require different controls, not permanent restriction.
AI is mid-arc. The only variable is how much talent and productivity you burn during the rationing phase.
The Silicon Ceiling
Access matters most when it reaches the person trying to finish the work. The usage figures suggest it still follows seniority instead.
BCG surveyed more than ten thousand workers across eleven countries and found a clean split: 78 percent of leaders and managers use AI regularly, compared with 51 percent of frontline employees. The people closest to the work are the last ones to receive the tools — the same pattern as the laptop era, played back at higher speed.
Leadership’s estimates can miss the usage gap. McKinsey found the C-suite underestimated it by a factor of three: executives estimate only 4 percent of employees use AI for a meaningful share of daily work, when the actual number is 13 percent. They are setting policy based on a picture of their workforce that hasn’t been accurate for two years.
Meanwhile, adoption moves on its own. U.S. workplace AI usage grew from 40 percent to 45 percent in a single quarter, with daily users climbing from 8 to 10 percent. Employees are not waiting for permission.
Effective use also requires training: right now, only 5 percent leverage AI in advanced ways. But the same was true of spreadsheets in 1988. Companies didn’t withhold calculators because most employees couldn’t use Excel. BCG’s own data shows that employee positivity about AI jumps from 15 percent to 55 percent with just five hours of coached training. The argument is not for access alone — it is for access plus training.
Shadow AI Is the New Shadow IT
Seventy-eight percent of employees admit to using AI tools not sanctioned by their employer. Nearly half hide their usage to avoid judgment. Forty-five percent pretend to understand AI tools in meetings. Much of that use takes place outside approved systems.
Samsung learned this the expensive way. In 2023, engineers in the semiconductor division uploaded proprietary source code into ChatGPT. Samsung banned all external AI tools, threatening termination for violations. Seven months later, the company unveiled Samsung Gauss — an internal AI platform that cost hundreds of millions to build. Engineers had discovered AI-assisted coding for twenty dollars a month. Samsung spent the next two years building, at enterprise scale, the capability it had banned.
The ban-to-build arc is not an outlier. It is the default trajectory for every organisation that chooses restriction over enablement — the mobile phone story replayed at double speed. Ban personal devices, fail, then build BYOD and MDM frameworks to govern what you couldn’t prevent.
The security argument for restricting AI access is real. But at 78 percent shadow usage, restriction is not mitigating risk — it is creating it. Your employees are already using AI. The question is whether they are using yours, or the one in the incognito tab. Fifty-four percent say they will use unauthorised tools if official ones fall short; among younger workers, 62 percent. Only 7.5 percent have received extensive AI training, yet 78 percent are using the tools anyway.
Those usage figures suggest that restrictions alone are not controlling access.
What candidates need to know
AI access and training also affect recruitment and retention.
EY found that employees with extensive AI training are 55 percent more likely to leave their organisations. The argument is that their market value has increased, and the companies on the other side of the table are offering an AI stack that their current employer won’t match.
Shopify made this explicit in 2025. CEO Tobi Lutke issued a memo declaring AI usage a “fundamental expectation” for every employee — embedded in performance reviews and peer evaluations. Before requesting new headcount, teams must demonstrate that AI cannot do the work. Compute access is no longer a benefit at Shopify. It is a measured competency.
Goldman Sachs rolled out its AI assistant to the entire firm in the same year, treating it the way it treats Bloomberg terminals — expensive per seat, but non-negotiable for competitive performance. JPMorgan embedded AI into an eighteen-billion-dollar technology budget. Candidates on Wall Street now ask about AI infrastructure in interviews the way they once asked about bonus structures.
The cited 66 percent productivity gain makes tool access a more consequential benefit than office amenities.
The cost objection is reasonable on the surface: token compute has a recurring marginal cost that laptops didn’t. But GPT-4-class models have dropped by orders of magnitude in cost since 2023, and the deflation curve shows no sign of flattening. The answer is not unlimited access; it is right-sized access at a cost that token deflation is making trivial. EY’s data says companies forfeit up to 40 percent of potential productivity gains when AI sits on weak talent foundations. The invisible denominator in every token budget is the talent and output you lose by not having one.
The mechanism is not “people quit because you didn’t give them ChatGPT.” It is that AI-skilled employees become more valuable in the market, and organisations that restrict AI skill-building create a competence gap that pushes top performers toward competitors who closed the debate a year ago.
The Token Line on Your Offer Letter
The organisations leading this shift didn’t start with unlimited access. They started with a diagnosis — mapping where they sat on the access spectrum and measuring the gap between what leadership assumed and what employees actually experienced. They quantified the cost of inaction: not just tokens, but the productivity delta of a workforce operating at 66 percent below its potential, the attrition premium on AI-trained employees who can see better offers, and the compliance exposure of ungoverned shadow usage.
Then they built tiers. Universal baseline access for every employee — the equivalent of a laptop, non-negotiable from day one. In regulated industries, “universal” means within the governance constraints of the sector; even classified environments are building internal AI tools, because the trajectory holds regardless of the compliance wrapper. Role-enhanced tools for functions where domain-specific AI drives competitive performance. Power-user access tied to demonstrated fluency, the way Shopify links expanded compute to performance reviews. Governance and access shipped together, rather than one perfected while employees bypassed the other.
The last step is the one most organisations haven’t taken: embedding compute fluency into hiring, onboarding, and performance management. Thirty-four percent of employees already expect AI for a significant share of their work within the next year. That number only moves in one direction.
The next job posting you write will either list AI tool access as infrastructure — alongside the laptop, the email account, and the security badge — or it will list it as a perk, alongside the kombucha and the ping pong table.
Tell candidates which tools they can use, what training they will receive, and where the usage limits sit. That gives them something concrete to assess alongside the work you are asking them to do.
References
- BCG, “AI at Work 2025: Momentum Builds, but Gaps Remain,” 2025. https://www.bcg.com/publications/2025/ai-at-work-momentum-builds-but-gaps-remain
- McKinsey & Company, “Superagency in the Workplace,” 2025. https://www.mckinsey.com/capabilities/tech-and-ai/our-insights/superagency-in-the-workplace-empowering-people-to-unlock-ais-full-potential-at-work
- PwC, “2025 Global Workforce Hopes and Fears Survey,” 2025. https://www.pwc.com/gx/en/news-room/press-releases/2025/pwc-2025-global-workforce-survey.html
- WalkMe/SAP, “Shadow AI Survey,” 2025. https://news.sap.com/2025/08/new-walkme-survey-shadow-ai-rampant-training-gaps-undermine-roi/
- EY, “AI Productivity and Talent Strategy Survey,” 2025. https://www.ey.com/en_gl/newsroom/2025/11/ey-survey-reveals-companies-are-missing-out-on-up-to-40-percent-of-ai-productivity-gains-due-to-gaps-in-talent-strategy
- Gallup, “AI Use at Work Rises,” Q3 2025. https://www.gallup.com/workplace/699689/ai-use-at-work-rises.aspx
- Nielsen Norman Group, “AI Improves Employee Productivity by 66%,” 2023. https://www.nngroup.com/articles/ai-tools-productivity-gains/
- Various, “From Luxury to Necessity: The Evolution of Mobile Phones,” 2024. https://vocal.media/earth/the-evolution-of-the-mobile-phone-from-luxury-to-necessity
- Shopify — CNBC, TechCrunch, Fast Company reporting, April 2025.
- Samsung — TechCrunch, CNBC reporting, 2023-2025.
- Goldman Sachs / JPMorgan — CNBC, PYMNTS, AI Magazine reporting, 2024-2025.