Sprinting Harder Didn't Make Me Faster

My observations on the fundamentals required to succeed with AI, and the importance of redesigning your whole factory once you get AI working for you.

Soccer players run a lot. A professional soccer player will run 6–8 miles per game. I was not a professional soccer player, but I did grow up playing soccer, and I grew up running a lot. I played soccer throughout my school years and even played some rec soccer into my 20s. When I moved to San Jose, CA I found myself without the time and network to play soccer, but I wanted to stay fit, so I thought: I ran a lot in soccer, why don’t I just take up distance running. Immediately, I found myself signing up for a half marathon.

What I had not realized was that running distances is a completely different way of running than soccer. Soccer is sprinting. In years and years of running, no coach ever said anything about my running form, my pace, my strides. It didn’t matter. After my painful half marathon, and subsequent limp-inducing marathons, I learned that those exact same things matter a lot for distance running, and that you just can’t get away with focusing solely on short-term training speed.

I was charged with driving AI transformation at Zendesk, and quickly realized that many of the things we needed to change were exactly those fundamentals that great businesses already practiced, the pacing, striding, and form that are easy to ignore. Looking around the market, I saw that the companies thriving with AI were already ahead on business hygiene, and are now doubling down on the same practices that separate them from the pack. Those practices add up to a compelling, if not obvious, roadmap for winning with AI, regardless of the technology. These companies are writing everything down (or having everything written), creating the context AI needs to breathe; they are bringing accountability and ownership back to central authorities; and they’re being intentional about how work gets done.

As an immigrant to the US, I spent a large chunk of my childhood trying to decipher those norms and mores that no one had written down but that fueled the interactions around me. This showed up in small ways: we did not know that if it was snowing, you should tune into the news channel to find out if your school is closed; and bigger ways: we didn’t know that tipping less than 15% was considered rude, or that when an American wants to meet at 8:00 am, they probably mean 7:58 am. When I became a globe trotting consultant, I learned that this problem, deciphering the unwritten rules, was not unique. My cubicle came stocked with a 600-page book from that era, Kiss, Bow, or Shake Hands, which was written (and sold) because navigating a different culture can be so difficult without any manual.

Yet, time after time, businesses try to operate across hundreds of people, in multiple locations, over years, without ever documenting key aspects of their own work norms and mores. Does your business care about revenue, profit, or acres of rainforest preserved from deforestation? Was there a decision made about the priorities for the quarter? How do you know? It was announced at the last all-hands? Well, Tammy from accounting can catch you up. This never really worked, but if you’re trying to make AI work for you, these gaps will guarantee mediocre outcomes. The AI will be as lost as I was before classmates knocked on my door to let me know that we didn’t have school that day. It has to guess about the reality you’re operating in, and it will provide answers that assume too much to be useful.

Instead, write it down. Specifically:

• Create decision documents and strategy narratives for all meaningful decisions.
• Record and transcribe your meetings (ideally shared across the organization).
• Save quarterly business reviews, with their talk tracks, in a folder the AI agent can access.
• Your performance review? Make sure it’s written down, and saved somewhere your AI agent can reach it.
• Those norms and mores of the business (how critical you are expected to be with each other, where the avenues for praise and criticism are, how decisions actually get made, not how they’re supposed to get made): write those down too.

Of course, AI can help you create this documentation. But you need to decide to commit to it, and to make it accessible to the agent. As AI gets more and more powerful, it will be able to extend your own capabilities even further. But don’t let the unspoken, unwritten truths get in the way of the right answer.

Once upon a time, you had figures of authority that meant something. Documentation teams created manuals, data teams produced figures, and travel agents booked travel. I wasn’t there, but it seems like most people hated it, given the massive enthusiasm for technology that let you book your own business travel, self-serve your data, and write your own documentation for the new API you launched. Power was pushed to the edge of the network, and it allowed a level of scale and distributed coordination that the central teams could only dream of. But it also meant that over my 20 years in business, nearly every company I’ve worked with has struggled to simply count key business attributes: number of customers, segment definitions, product instrumentation (if you think about it… what really is an active user?).

With AI capabilities coming to the fore, we’re seeing a rebalancing of corporate forces back toward centralized authorities that can ensure the AI is pulling the right data point, referencing the right product context, and using the company-specific definitions appropriately. At Zendesk we’re doing this in a couple of ways: we’ve created a Knowledge Center of Excellence intended to be the orchestrator of product knowledge, driving maximally robust and up-to-date information about how our products work (and don’t work). The data teams, meanwhile, are creating AI-readable context files that guide the AI toward the best data sources for a given purpose.

We’re not unique. Shopify famously has a company agent, River, that everyone interacts with in public Slack channels. They note that everyone can merge changes and updates to River, but what’s often left unstated is that there’s someone on the other side of those merge requests, curating the work, deciding what goes into River and what doesn’t. All of that becomes the source of truth for the company. Anthropic recently wrote up their advice on using Claude Code for self-service data analytics, and while the outcome is self-service data, the input is not self-service at all. It requires a foundation of:

• Canonical data sets
• Enforced standards
• Colocated artifacts
• Metadata as a first-class citizen

Someone is entrusted and expected to be the central source behind those foundational requirements. It’s not just distributed across the business. Again, AI can do much of this, but the human orchestrator is still critical. With agents like River, or the skills Anthropic argues for, you also see a rise of company standards as shared code (written into shared skills, or directly into system prompts), where anyone using the AI has a standard enforced at the production level, rather than relying on each human to choose to follow the pattern or not.

Lastly, using AI is requiring companies to be intentional about how they do business. Until recently, “do we need humans to do that?” would have been a silly question. Of course you need humans; they just might be PhD humans or recent-grad humans, American humans, Indian humans, or Ukrainian humans. Now there’s an emerging choice: manage some aspects of your business (the coding, the customer service, the accounting) with more direct human involvement, and others with less of a human imprint.

At Zendesk, we’ve made deliberate choices here. We’ve built our Community with a very human feel; we want it to be a refuge from AI, a place to foster peer-to-peer connection, which requires human judgment, warmth, and coordination. In contrast, we’ve chosen to lean on AI in spaces where time is of the essence, or where the best experience already looks a lot like a mechanical one. These are documented choices that we’re building against, and that we can reconsider as we learn more about the capabilities of AI and the needs and expectations of our customers.

In the same mode, leaders now need to be intentional about what the desired result of a given journey or touchpoint with a customer actually is. It used to be that you’d have a fleet of trained human SDRs, ultimately asked to drive new business through a mix of helpfulness and persistence. When AI steps into that role, the business has to decide: When do handoffs happen? Should everyone who engages the AI SDR get human follow-up? Can the AI create a deal from an interested lead on its own, and do we want it to?

Do all three well (write it down, centralize the authority, choose where humans belong) and you’ve earned something real: a start line. But fixing your form was never the point of the race. The bigger prize isn’t doing your old work with AI; it’s rethinking the shape of the work itself. And that’s the hardest thing to see, precisely because it requires understanding all the unwritten decisions, tradeoffs, and defaults baked into the system (see the write-everything-down section above). As David Foster Wallace shared in his 2005 Kenyon commencement address:

There are these two young fish swimming along and they happen to meet an older fish swimming the other way, who nods at them and says “Morning, boys. How’s the water?” And the two young fish swim on for a bit, and then eventually one of them looks over at the other and goes “What the hell is water?”

This is a standard requirement of US commencement speeches, the deployment of didactic little parable-ish stories. The story thing turns out to be one of the better, less bullshitty conventions of the genre […]. The point of the fish story is merely that the most obvious, important realities are often the ones that are hardest to see and talk about.

You can’t redesign a system whose shape you can’t see, and the assumptions that run deepest are the ones you’ve stopped noticing entirely. So before you can change the water you’re swimming in, you have to be able to name it.

In the early 1900s, American factories were slow to capture the advantages of electrification, because the steam engine and shaft-and-pulley systems were baked into the very shape of the factory. The biggest advantage of electricity, it turned out, was not that it could turn turbines faster or cheaper than steam. It was that it let you redesign the entire factory. As the economic historian Paul David documented:

The advantages of the unit drive for factory design turned out to extend well beyond the savings in inputs of fuel derived from eliminating the need to keep all the line shafts turning […]. Factory structures could be radically redesigned […]: savings in fixed capital through lighter factory construction […], further capital savings from the shift to building single-story factories […]. Single-story, linear factory layouts, in turn, permitted closer attention to optimizing materials handling, and flexible reconfiguration of machine placement […]. [T]he modularity of the unit drive system and the flexibility of wiring curtailed losses of production incurred during maintenance, rearrangement of production lines, and plant retrofitting; the entire power system no longer had to be shut down in order to make changes in one department or section of the mill.

That is to say: electrification didn’t just change how factories made power. It reduced their overall cost structure, optimized materials handling and production, and improved their agility overall, but only for the companies willing to rebuild around it.

We are currently at the stage of AI implementation where we’re swapping out humans doing certain tasks (coding, writing, customer service) with AI doing those same tasks. But what is the equivalent of the full factory redesign? What are the assumptions about how work gets done that are baked into the very “AI transformation” efforts everyone is taking on? I promise there’s something there, but finding it may require asking about the very purpose of your business, examining what your customers actually expect, and reimagining how you deliver it in a world powered by AI.

Despite all that sincere effort, I never did get faster by sprinting harder. I’m training for another half marathon now, and the thing that’s actually working is the boring stuff I used to skip: form, pace, stride. The fundamentals I was sure didn’t matter.

It turns out most of the things that matter are like that. The unwritten rules I spent my childhood decoding. The water the fish couldn’t see. The factory floor nobody thought to redraw. They’re invisible right up until they’re the only thing standing between you and where you’re trying to go.

AI doesn’t change that. It just makes it impossible to ignore. The companies that win this era won’t be the ones that sprint hardest at it. They’ll be the ones willing to do the unglamorous work of writing down what they actually believe, deciding who owns the truth, and asking, honestly, what their business is even for.

So: what’s the water you’re swimming in? And are you willing to look at it?