Larridin

AI readiness has become a business asset. Boards are asking questions like: What are we spending on AI? Who is using it? What are they producing with it? Is the investment generating returns we can measure? The challenge is that the data required to answer these questions lives in separate systems. Spend data sits in finance. Adoption data sits in IT. Outcome data sits in business units. Bringing them together into a coherent view requires infrastructure most organizations haven’t built. The result is that executives can report how much was spent. They can report how many people have access. They struggle to report what those people produced with that access, and whether the output justified the cost. The pressure is growing. Investors are factoring AI readiness into valuations. No executive wants to walk into a room and say there is widespread adoption, at significant cost, but no reliable way to measure what it is producing. So surveys go out. Scores come back. The average looks reasonable. The gap between adoption, proficiency, and ROI gets wider.

The consulting industry moved fast on AI, and now it’s recalibrating. KPMG built a dashboard to track AI tool usage across its US advisory division, part of a broader effort to move from basic adoption to more sophisticated use. McKinsey went further. The firm uses roughly 25,000 AI agents alongside its 40,000 human employees, and expects one or more agents to eventually support every employee. The surge in spending has raised a question that consulting firms are now working to answer, both for their clients and for themselves: Are companies investing in AI strategically, or simply spending to avoid being left behind? Boston Consulting Group found that companies expect to more than double AI spending in 2026, from roughly 0.8% of revenue to about 1.7%. For large enterprises, that shift represents billions of dollars flowing into strategies that remain experimental and difficult to measure. The measurement infrastructure that seemed optional a year ago is becoming necessary.

The era of flat-rate AI pricing is ending. GitHub Copilot is moving all plans to usage-based billing, where every interaction consumes tokens. When credits run out, AI-powered work stops. Other AI providers are making similar shifts, moving from seat-based licensing to consumption models tied directly to token usage. Under flat-rate pricing, a junior employee wrestling with prompts for hours costs no more than an expert completing the same task in minutes. Under usage-based pricing, that inefficiency translates directly into higher spend. Uber offers a recent example of how quickly the exposure can surface. After deploying Claude Code to roughly 5,000 engineers in December 2025, the company exhausted its entire 2026 AI tools budget within four months. Is there help on the way? Gartner projects that inference costs could fall by 90% by 2030. Total AI bills are still expected to rise. The reason is that agentic systems require significantly more tokens per task than chat interfaces. Rising consumption can outpace falling unit costs. The implication is that cost management is no longer optional.

Enterprise AI adoption is high. Proficiency is not. Median AI proficiency across enterprise deployments sits at 58.5 out of 100, based on measured output rather than self-assessment. Two-thirds of organizations report using AI across multiple business functions. Yet 85% of workers say they cannot connect their AI training to their actual job. The gap between these numbers tells a story. Adoption measures access. Proficiency measures capability. Self-reported surveys measure perception. None of them measure whether AI is changing how work gets done. Under flat-rate licensing, this gap was invisible. Under consumption-based pricing, it becomes expensive. Every misfired prompt, vague instruction, and iterative exchange to get a usable output costs money.

Six months ago, companies built leaderboards to encourage AI usage. Now they are setting caps. The data: "Tokens" mentioned in 129 earnings calls in Q2 2026, up from 57 the prior quarter (AlphaSense) Tech and media companies spent $66.29 per employee on AI in May, up from $58.84 in April (Ramp AI Index) Pylon approached a $1.4M Anthropic bill before setting caps Coinbase, Walmart, and Amazon have all implemented token limits or shut down usage leaderboards The shift: One Y Combinator CEO said his 6-person company would have needed 20 employees pre-AI. The difference went to tokens. "We have to weigh our token-to-people ratio," he said. Engineers are now asking about token budgets in job interviews: what tier of model, what caps, what partnerships. The pattern: Unlimited spending did not always yield results. Companies are now asking who should get access to which models, and why.

MindFort is a Y Combinator-backed startup with six employees. Pre-AI, the company estimates it would have needed twenty to reach its current scale. The difference went to tokens. CEO Brandon Veiseh on the tradeoff: "We have to weigh our token-to-people ratio. It's not something we think is particularly comfortable or a great feeling to say." The math is shifting across the industry. Headcount decisions and token budgets are now the same conversation. The ratio is: how many people plus how much compute. It has implications for hiring, budgeting, and workforce planning. A company with a higher token-to-people ratio can operate leaner, but becomes more exposed to pricing changes from AI providers. A company with a lower ratio has more predictable costs, but may fall behind on output.

read here: larridin.com/blog/most-toke…

Larridin research shows most of the productivity boost from AI coding assistants comes from just one-third of total spending. The findings: Less than one-third of users are responsible for more than 50% of spend. The first one-third of spending drives 85% of productivity gains. After that point, spending rises much faster than productivity. What happened: AI providers introduced agentic workflows that burn tokens at scale. They moved to per-token pricing. Many companies created token usage leaderboards, promoting heavy use with no tie to productivity. Employee token burn became a proxy for competence. The headlines tell the story: "Uber burned through its entire 2026 AI budget in four months" (Fortune) "Client Accidentally Burns $500 Million on Claude AI in One Month" (Yahoo Finance) "AI sticker shock hits corporate America" (Axios) The problem is not the employees. They were encouraged to maximize usage. It’s certainly not AI. When used properaly, it delivers strong ROI. The issue is adoption without measurement. Move fast and break things…with no framework to optimize. Full article in the comments

Most AI visibility tools track human activity or agent activity. Not both. That leaves a gap. In a typical enterprise deployment: 27% of sessions are agents 72% of variable spend is agent-driven Agents run continuously. They chain requests. They do not pause. A human runs ten queries. An agent runs ten thousand. If you only track humans, you miss most of the spend. If you only track agents, you cannot tie spend to outcomes. You have to have both views, in one place, with ownership attached.

Find out more: ai-spend.larridin.com/ai-spend-intel…

Larridin launches Token & Spend Insights. Sources tracked: Cloud model providers, subscriptions, browser plugins, desktop OTel agents, custom connectors, API gateway integrations Attribution: Human and agent token usage in a unified view. Spend tied to teams, use cases, and business outcomes. Alerts: Total spend, projected overages, orphaned agents, dormant seats Larridin internal data: less than half of unmanaged AI spending is productive. Larridin customers reduce shadow AI by 90%+, identify tools and practices worth scaling, and reach 50%+ power users company-wide. Learn more in the comments.

Read the full report here: accenture.com/us-en/insights…

New research from Accenture and Wharton quantifies what the "Human+" workforce looks like in practice. Using biopharma as a test case, researchers analyzed 300 tasks across 90 roles and mapped them to 50 AI agents. The findings: Approximately 55% of total workforce hours will be impacted by digital and physical agents. The top digital agents: Assistant, Analytics, and Knowledge Base. The top physical agents: Robotic Equipment Handling, Vision Inspection, Automated Sample Preparation. The functions most affected: Finance, HR, Procurement, Insights & Analytics, Marketing. The estimated annual opportunity in US biopharma alone: $180-240B. But the research also surfaces a critical nuance from a meta-analysis of 106 experiments: When humans outperform AI alone, augmentation improves outcomes. When AI outperforms humans alone, adding humans back in actually decreases performance. The implication: companies need visibility into which tasks belong to which workforce component, and the ability to measure the economics of each combination. The full report covers individual, economic, organizational, and societal implications of the blended workforce.

Most CIOs can name every licensed AI tool in their stack. Almost none can tell you what those tools are actually returning. "It's not about impeding freedom of the employees. It's helping to find out the right tooling that gets the measurable outcomes." The CIOs getting this right are not building better “approved” lists. They are building evidence that connects specific tools to specific outcomes, and using that data to make portfolio decisions. A control frame drives you toward governance policies. A measurement frame drives you toward proof. That distinction changes everything.

Find out more: ai-spend.larridin.com/ai-spend-intel…

Today Larridin launches Token & Spend Insights. Sources tracked: Cloud model providers, subscriptions, browser plugins, desktop OTel agents, custom connectors, API gateway integrations Attribution: Human and agent token usage in a unified view. Spend tied to teams, use cases, and business outcomes. Alerts: Total spend, projected overages, orphaned agents, dormant seats Larridin internal data: less than half of unmanaged AI spending is productive. Larridin customers reduce shadow AI by 90%+, identify tools and practices worth scaling, and reach 50%+ power users company-wide.

How do you tell the difference between AI proficiency and AI theater? Mike Giresi, CIO at Vertiv, has a simple test: can you show a before and after that anyone can see with their own eyes? His team deploys AI against activities that used to take 25 hours or 25 days. When the output that took a week now takes 15 minutes, you do not need a slide deck to make the case. "There's no emotion involved. Just data." Most AI initiatives inside large companies are theater. Not because anyone is lying. People want to look good in front of their boss. In the absence of real measurement, stories fill the gap. The difference between proficiency and theater: the latter needs a story. The former speaks for itself.

Larridin CEO Russell Fradin on @CNN yesterday to talk about Anthropic's IPO. He didn't mince words about AI. 2026 might go down as a turning point for most of the F500, and everyone's pretty much adopting AI on the fly. Larridin's State of Enterprise AI report uncovered the truth: less than half of organizations are aware of their AI adoption rate. Anthropic's IPO shows the demand. What shows the impact? That's what we're building. Great segment, and thanks to CNN International and @BeckyCNN for having us.

listen to more here: podcasts.apple.com/vc/podcast/ai-…

Dan Zhang, CFO at ClickUp, uses three messages to communicate AI impact, each designed for a different audience. For frontline employees: Keep it immediate and tangible. No jargon. No strategic framing. Ask one question weekly: what did your work look like before, and what does it look like now? A concrete before-and-after tied to their actual job. For business leaders: AI productivity that floats free of objectives does not register. The impact has to connect directly to the metrics they are accountable for. If the connection is not explicit, the enthusiasm fades. For the board: Quick wins are not enough. They want evidence of sustainable, long-term improvement to the business model. Not stories about individual productivity - structural changes to how the organization operates.