From AI Theater to Unreasonable Agility

Organizations are rushing to deploy AI coding tools, often falling into the trap of “AI Theater”—measuring token usage and lines of code generated instead of real business impact. While it’s true that coding is becoming exponentially faster, simply generating more code without changing the surrounding system only moves the bottleneck. If code review, testing, or feature adoption can’t keep pace, all that AI-generated output is just inventory sitting in the system.

Instead of inflicting AI mandates, leaders need to treat AI as a product that teams genuinely want to use to solve their real friction. By orienting work around outcomes and identifying the new bottlenecks, we can finally stop playing defense in R&D. When engineering is no longer the constraint, we unlock the door to “unreasonable agility”—the ability to jump on customer problems and deliver value overnight.

Over the past decade, we’ve watched many organizations implement Agile by focusing purely on mechanics—tracking velocity, holding rigid ceremonies, and counting story points without actually improving how work flows. Now, we are seeing the exact same pattern repeat with generative AI. This new wave of AI mandates often looks like the “underpants gnomes” strategy from South Park: Step 1 is to use all your tokens, and Step 3 is somehow profit, with no clear idea of what Step 2 actually involves.

When leadership inflicts AI on people by mandating its use and measuring activity, smart people will naturally find ways to comply without creating any real impact. They figure out how to consume tokens and generate output just to check the box. This malicious compliance creates an illusion of progress. We are trading the old Agile theater for a brand new AI theater, celebrating our high token counts while the actual value delivered to customers remains completely flat.

If you look back at how software was built around 2010, the bottleneck often wasn’t writing the code itself. I remember working with a large tech company where development teams were churning out features, but testing was a high-overhead manual activity that couldn’t keep up. The constraint was clearly in the QA phase, meaning all the extra code being written was just piling up as unreleased inventory.

“The testing is a high overhead activity, even if you’re making a small change in a complex system… The same pattern is something we’re seeing today.”

Fast forward to today’s AI-assisted development lifecycles, and we are facing a very similar reality. If you look at AI coding tools, the bottleneck is simply not there anymore. The coding step might be ten times faster, but you don’t necessarily see ten times the throughput at the end of the line. The constraint has just shifted downstream to code review, deciding which features to build in the first place, or getting users to actually adopt those changes. We have to orient our flow around these new bottlenecks instead of continuing to optimize the part of the system that AI has already solved.

To break out of AI theater, we need to stop mandating these tools and start inviting people to use them to solve their actual daily friction. A great diagnostic is to ask your teams a simple product-market fit question: “Would you care if you couldn’t use your AI tools tomorrow?” If they wouldn’t miss them, then the AI deployment isn’t actually solving a real problem for them. Treating AI as an internal product means appealing to their intrinsic motivation and helping them do their jobs better, rather than forcing compliance.

We can also use this transition as an opportunity to build better habits. Instead of just telling an agent to write a script, we can configure our agent preferences to prompt us to think about the leading indicators and outcomes we want to achieve. This kind of habit stacking encourages engineers and product managers to focus on the impact of their work rather than just the output, helping the whole organization gradually shift its mindset away from pure activity tracking.

If we successfully navigate these shifting bottlenecks and get to a point where R&D is no longer the constraint, the possibilities completely change. I was recently talking to a cybersecurity CEO who envisioned this exact scenario: moving from defensive R&D—where processes exist mainly to protect a constrained engineering team—to offensive R&D. In this new world, engineering can actively hunt for ways to delight customers and deliver massive value almost instantly.

“If you look at the AI coding… the bottleneck is not there anymore. Like the coding step is 10x faster, but you don’t necessarily see 10x throughput at the end.”

This opens the door to what we can call “unreasonable agility,” a concept inspired by unreasonable hospitality. It’s the ability to hear a customer mention a strange edge-case problem on a Tuesday afternoon and show them a working solution by Wednesday morning. It means we stop giving excuses, stop hiding behind massive backlogs, and start jumping on valuable opportunities the moment they appear. This is the true promise of agility finally being realized.

If you are ready to stop measuring tokens and start measuring real business impact, the first step is understanding where your constraints actually live today. Let’s talk about mapping your flow, finding your new bottlenecks, and using Organizational AI Coaching to turn all that generative activity into tangible outcomes.