AI Uplugged Podcast Episode 15
The AI conversation is changing.
For a while, the AI conversation was about possibility.
Could it write code? Could it summarize documents? Could it answer customer questions? Could it automate the work nobody wanted to do?
But in Episode 15 of AI Unplugged: ROI, Tokenomics & the AI Bubble, the conversation moves into a more mature phase. The sharper question is no longer whether AI can create value, but how that value is measured, managed, and repeated inside the business.
That shift matters. The next wave of AI adoption will not be won by the companies using the most tokens, launching the most pilots, or adding the most chat interfaces. It will be won by the teams that can connect AI to real workflows, control the cost of running it, and prove that it changes business outcomes.
AI ROI is getting harder to ignore
The episode opens with a question that cuts through the hype: who is actually seeing returns on AI investment?
The team discusses a Financial Times report on hyperscaler AI investment, where Microsoft, Alphabet, Meta, and Oracle were cited as showing negative implied ROI, while Amazon was the only one showing positive return. That contrast creates a useful tension. Maybe the winners are not always the companies trying to build the biggest model. Maybe they are the companies enabling everyone else to build, run, and scale AI.
For customers, the lesson is practical. AI ROI does not come from chasing the flashiest model. It comes from understanding where the business actually creates value.
The same logic applies to the bigger “AI bubble” conversation. With names like OpenAI, xAI, and SpaceX moving into the IPO discussion, it is fair to ask whether valuations are getting ahead of reality. But a valuation reset does not mean the underlying technology stops mattering. The dot-com era proved that. The market corrected, but the internet still changed business.
AI may follow a similar path. Some bets will not hold up. But the companies with real use cases, real infrastructure, real workflows, and real economics will be the ones that last.
Usage is not the same as value
More usage does not automatically mean better outcomes. More AI-generated output does not automatically mean a faster team. More automation does not automatically mean more profit.
That tension shows up clearly in the discussion around tokenomics. For a while, heavy AI usage was treated like a signal of progress. More tokens meant more adoption. More adoption meant more productivity. But that logic is starting to break down.
Token consumption can become its own kind of vanity metric. Teams can use more AI, spend more on models, and still miss the business outcome they were trying to change.
The real question is simpler and harder: What changed?
Did the workflow move faster? Did the team reduce manual effort? Did the business improve consistency, accuracy, capacity, or customer experience? Did AI become part of how work gets done, or did it remain a separate experiment?
That is the difference between AI adoption and AI management.
Productivity may be a lagging indicator
AI can create measurable lift when it is close to the work.
Innovative can point to its own experience as a system integrator, where a large share of work output is code. In that context, the team discusses a conservative 15–20% productivity gain from AI-assisted work.
But the episode also challenges the idea that productivity will always show up cleanly or immediately. For customers, the bigger return may come later, when agentic workflows repeat across monthly, quarterly, or annual processes. The value is not just in completing one task faster. It is in changing the rhythm of the work.
That is also why the AWS tweet discussed in the episode lands so well: “More AI-generated code does not make your team faster. It might actually slow you down.”
It is a useful reminder that AI productivity is not just about output volume. When AI generates more code, teams also inherit more to review, test, secure, refactor, document, and maintain. Speed only matters if the output improves the workflow, reduces risk, and helps the business move with more confidence.
Cost and compliance are becoming the next AI frontier
AI is entering its cost-control era.
As usage grows, so does the need for visibility, governance, and control. Leaders need to understand what AI is doing, how often it is running, what it costs, and whether that spend is tied to a business result.
That becomes even more important with agentic workflows. When an agent goes off to complete a task, it may not be obvious up front how much the work will cost. By the time the task is done, the bill may already be real. That makes AI cost management less of a reporting issue and more of a design issue.
This is where AI starts to look a lot like cloud.
Early cloud conversations were often framed around cost savings and security concerns. But the deeper value of cloud was not simply replacing servers. It changed how businesses built, deployed, and scaled applications. Over time, organizations learned that cloud value depended on operating discipline: architecture, monitoring, governance, optimization, and smart scaling.
AI is reaching a similar point. The opportunity is still real, but the management layer matters more now.
The chatbot is not the strategy
Many companies still begin with “we want AI” and mean some version of a chat assistant. But chat is only one interface. It may be useful, but it is not the full strategy.
The more durable value comes from embedding AI into products, processes, systems, and decisions. That is the practical frontier: AI that does the work inside the business, not just talks about the work from the outside.
A chatbot can help someone ask a better question. A workflow-connected AI system can help the business move faster, reduce friction, improve consistency, and unlock new capacity.
That distinction matters. Because “AI in the business” is not the same as “AI beside the business.”
The enabling layer may be where the value lives
Foundation models may continue to consolidate around a handful of major players. But for most businesses, the model itself is not always the most important differentiator.
The value increasingly lives in the enabling layer: the systems, workflows, integrations, controls, and context that turn model capability into business impact.
That layer is what connects AI to the way the business actually runs. It is what turns a model from a powerful tool into a practical operating advantage.
The path forward
For growth-stage companies trying to move from AI experimentation to AI outcomes, the path is becoming clearer.
Start with a workflow worth improving. Define what better means. Understand the cost to run it. Build the right controls. Measure what changes. Then repeat.
AI ROI is not dead. It is becoming more disciplined.
And for teams willing to manage AI with the same seriousness they brought to cloud, that discipline may be exactly what turns experimentation into advantage.
