Go Slow To Go Fast: A Parallel Between Scaling Organizational AI And Marathon Training
In marathon training, the hardest part for many runners, myself included, isn't the early morning wakeup calls, the long runs, or the seemingly never ending soreness in your legs. It's the taper before the race.
Let me set the scene: You’ve been consistent for months, and built the endurance, discipline, and confidence to know that you can actually finish the thing. But then, leading up to race day, you do something that feels completely backwards... you slow down.
Cut the mileage. Ease off the intensity. You stop trying to prove how hard you can push. Not because you’re losing momentum, but because you are enabling your body with the rest and recovery to perform at 100% efficiency on race day. That’s called tapering. Sounds easy enough right? But ask any endurance athlete you know, and they will share this sentiment: tapering feels frustrating, and counterintuitive (we all want to keep moving, go faster). But your taper can make or break your race.
With many of the companies I work with, this is the step they skip when they start implementing organizational AI native systems and processes. What I see instead feels a lot like a runner who decides the week before a marathon is the perfect time to double their mileage (good luck with that!). There’s a rush of energy at the beginning of AI adoption: new tools, new possibilities, a sense that if you move fast enough you might unlock something transformative. So teams go wide! They launch multiple use cases, roll out different tools, encourage everyone to start experimenting at once.
The data isn’t ready. The use cases aren’t clearly defined. Workflows feel clunky. This is the moment where "tapering" should kick in across the org (not necessarily as a slowdown, but a shift).
In scaling internal AI tools, that usually means stepping back from the urge to launch more and instead asking a slightly harder question: what’s working, and how do we make it stick? This is where things get real. It’s one thing to introduce a capability. It’s an entirely different ball game to make it useful in someone’s day-to-day work. A lot of AI efforts stall because they stay in what I’d call “launch mode.” There’s an announcement, some excitement, maybe some early adoption. But they never quite make the transition into habit. And without that, there’s no real impact.
Tapering is what creates that transition. It’s where you narrow your focus to a small number of use cases that are practical and tangible. It’s where you clean up the unglamorous stuff, like data quality, access, basic standardized workflow design, because that’s what determines whether AI tools feel helpful or frustrating.
Just like in running, this part of the process is hard. It can feel like you’ve lost a bit of momentum. But this is the work that makes everything else possible. Because if you skip this phase, you risk inefficiency, underutilization of the investment allocated towards these tools, and above all, you risk losing trust among your people. Once someone decides a tool is unreliable or more trouble than it’s worth, they'll stop using it altogether. Getting engagement back is much harder than getting it started.
That’s why the “go slow to go fast” idea, as overused as it is, matters here. Resisting the urge to spread your efforts too thin and instead building systems that can hold up in the day to day routine, and taking the time to train teams through tangible, repeatable use cases.
The companies that are doing this right are focusing their energy, tightening their feedback loops, and treating enablement as part of the product deployment rather than an afterthought once the utilization data is down.
This is where your speed kicks in: systems become standardized, curiosity turns to confidence, and efficiency is at an all time high.
The irony in all of this is that tapering feels uncomfortable because it’s working, just not in the fast paced way we're used to. That discipline, and intentionality, is what sets up the performance. If you’re in the middle of rolling out an internal AI tool, and things feel a little slower, it’s worth asking whether that’s a problem, or a sign you’re finally in the right phase. Give your team the space to actually get good at this!
When it’s time to really push, you’ll be ready to run.
Comments