Goodbye Tokenmaxxing: From AI Usage to Agentic AI Outcomes Transcript, AI Summary & Key Points

00:00 Have you ever been told to use more AI at work as if pure usage is what's being rewarded? Teams often rely on activity metrics to measure how AI is being used. But the challenge comes when those metrics become proxies for success rather than simply indicators of activity. For example, if I'm on a sales team and I become focused on call volume, I might hit my activity targets every single week.

00:30 But call volume alone doesn't tell me whether I'm on track to hit my quarterly minimum. Today, we're going to explore how teams are facing a similar challenge with AI. Over the past year, many teams measured AI usage through adoption metrics. How many people are using AI? How many tokens are being consumed? And a token is the basic unit AI models use to process text.

00:59 Making token consumption a common way to measure AI usage and AI cost. Token consumption became a proxy for engagement. Some teams encouraged broader AI usage, believing heavy consumption would lead to better outcomes. This is tokenmaxxing, maximizing AI usage in the belief that more activity equals more value. As an adoption signal, it worked, but as AI scaled to everyday infrastructure, a problem emerged.

01:34 As Neil Dhar, SVP for IBM Consulting pointed out, in the absence of true metrics, teams created usage dashboards, which people quickly learn to game. Usage soon became a proxy for value. As AI evolved into systems that could complete work, that approach became increasingly limited. Agentic AI systems were doing far more than generating code. They were planning workflows, analyzing repositories, calling tools, testing solutions, and coordinating work across multiple systems.

02:14 More capable agents naturally consumed more tokens, but token consumption alone couldn't distinguish between activity and measurable outcomes. Teams could see AI usage increasing without a clear understanding of the value being produced. So teams started asking a different, difficult question. What happens when we have to pay for all of this and what did we actually get out of it?

02:45 The dashboards showed growing usage and the token counts kept climbing, yet the connection between that activity and system-level outcomes remained unclear. That gap between consumption and value triggered the next phase of the conversation. As AI costs increased, teams responded by focusing on the most visible metric, token consumption. New policies emerged around limiting usage, restricting newer models, shrinking context windows, and reducing prompts.

03:20 But token minimization falls into the same trap as token maxing. Both assume token consumption is the primary metric that matters. Neither measures operational outcomes. Here's what often happens. Teams remove obvious inefficiencies like oversized tool catalogs or stale context. And that makes sense, but then they keep cutting and start removing critical information like task descriptions.

03:49 Domain constraints, and architectural context. For example, strip out architectural context to save tokens, and the AI might generate code that works in isolation, but breaks the broader system. You save 500 tokens on input, but now you're spending 5,000 tokens on debugging and rework. Costs don't disappear, they just move. Teams celebrate lower input tokens while paying for the same complexity elsewhere in the workflow.

04:24 So what should we optimize for? Valuemaxxing, a term coined by Mark Boroditsky, CRO of Nebus, shifts the conversation from consumption to outcomes. Instead of asking, how many tokens did we use, teams are asking different questions. How many deployments were completed? How much developer time was saved? How much rework was avoided? How many vulnerabilities were resolved?

04:57 These are some of the metrics that determine whether AI is creating those measurable outcomes and whether incremental token usage is justified. Token consumption is rooted in higher quality software and successful outcomes across the SDLC. Now let's take that one step further. This distinction becomes important as models become infrastructure. Access to great models is no longer the main differentiator.

05:28 System effectiveness is shifting more to the system built around the model. This includes things like context management, workflow orchestration, governance, as well as optimization. If outcomes matter more than tokens then systems matter more than models. That's why model orchestration is also becoming more important than model selection. IDC predicts that by 2028, 70% of leading large-scale AI deployments will work across multiple models rather than rely on a single model.

06:07 Developers and platform leaders are both accountable for value-maxing. For developers, AI efficiency is becoming a new engineering skill. Developers should think about AI usage the same way that they think about cloud resources, databases, or application performance. The objective is not to use less AI, but to use AI more effectively. Good context hygiene, planning before execution, and understanding trade-offs between cost and outcome.

06:40 All contribute to higher quality results. For platform leaders, the responsibility is different. Platform leaders need visibility into both costs and outcomes. They need to measure value instead of volume, reward efficiency, and give teams the tools to understand the impact to track outcomes. Building a culture of AI efficiency requires accountability, visibility, and shared metrics across engineering workflows.

07:10 Platforms are part of the equation as well. Developers and leaders cannot shoulder responsibility for AI efficiency alone. The platforms they use should help connect AI consumption to outcomes. Administrative controls provide budgets, governance, and visibility into how AI is being used. Analytics help teams connect AI consumption to operational outcomes.

07:38 Rather than token counts alone, while workflow capabilities, skill, and tool integrations reduce unnecessary work and preserve the context needed to deliver high quality results. The economics of AI will continue to change as model efficiency may improve and costs may shift. The first chapter was about adoption and the next chapter is about value. Token consumption helped teams understand whether AI was being used. Valuemaxxing helps teams understand how AI is creating value.