No Room for Tourists
Wade Foster doesn't send memos about AI. He runs hackathons and show-and-tells. That distinction matters more than most CEOs realize—and it's the same thing I've been telling my own teams.
Nino Chavez
Principal Consultant & Enterprise Architect
I’ve been saying something to my teams lately that probably sounds harsh.
If you want to get in this boat, you need to learn to row. There’s no room for tourists.
It’s not meant to be gatekeeping. It’s an observation about what’s actually happening.
The gap between people using AI and people building systems around it is widening so fast that the middle ground is disappearing.
Then I watched Wade Foster’s interview about how Zapier’s CEO actually uses AI—not the marketing version, the real version—and I felt like someone finally said the quiet part out loud.
The Delegation Trap
Here’s the thing Foster calls out that I’ve been circling for months: CEOs fall into a “delegation trap.”
They send memos about AI. They sponsor initiatives. They approve budgets for tools.
But they don’t use the tools themselves.
This creates a specific kind of organizational paralysis. The people at the top are signaling that AI is important, but they’re not demonstrating what “good” looks like. They’re not making mistakes in public. They’re not showing the messy middle.
I’ve seen the difference this makes. When leadership actually builds something with AI—even something small, even something imperfect—it changes the entire conversation.
Suddenly it’s not abstract strategy. It’s craft.
The Economically Unviable Becomes Viable
One of Foster’s points hit particularly hard: AI agents make tasks that were previously “economically unviable” suddenly possible.
Think about it.
There’s a category of work that’s too tedious or too expensive for humans to do consistently. Not because it’s not valuable—because the cost-to-value ratio was upside down.
I see this in my own work constantly. Things I would never have spent time on—because the ROI didn’t justify it—are now table stakes:
- Documenting every decision
- Running structured reviews on every output
- Building linter-bots to police my coding agents
None of this would have been worth the effort before.
Now? It’s just… what you do.
AI Fluency Rubrics
The part of Foster’s approach that made me stop and take notes: Zapier has created specific rubrics to measure “AI fluency” for different roles.
Not vague “embrace AI” mandates. Specific, measurable expectations for what AI skills look like at different levels.
This is the “learn to row” principle made operational.
I’ve been guilty of this. Telling people to “use AI more” without giving them a clear picture of what “more” means. Without showing them the difference between prompting and governing. Between vibe coding and intent engineering.
The rubric forces the conversation. What does “good” look like for a product manager? For an engineer? For a designer?
The answers are different—and the specificity is the point.
Mining the Unspoken Culture
This one might be the most interesting.
Foster uses a meeting recorder (Granola) to transcribe meetings, then feeds the transcripts to AI to extract “unspoken culture.” He compares what the data shows against Zapier’s stated values.
What behaviors are actually rewarded? Not what’s written in the handbook. What happens in the room.
It’s auditing culture through signal processing.
I’ve been thinking about this in the context of AI governance. We write instructions for our agents—CLAUDE.md files, system prompts, constitutional frameworks. But the agents’ actual behavior often drifts from what we intended.
Foster’s doing the same thing, but for humans. Surfacing the gap between what we say we value and what we actually do.
The Hiring Evaluator Agent
Foster built a custom agent that acts as an “expert hiring evaluator.”
It reviews interview transcripts against the job description and company values, provides a yes/no recommendation with reasoning, and serves as a bias check for hiring managers.
This isn’t replacing the hiring decision. It’s adding a structured second opinion that:
- Doesn’t get tired
- Doesn’t have unconscious biases toward candidates who remind them of themselves
- Doesn’t forget to check for specific criteria
It’s the same pattern I’ve been writing about: AI as governance layer. Not making the decision—providing oversight on the decision-making process.
Systems, Not Prompts
The reason Foster’s approach resonates is that it’s not aspirational. It’s operational.
He’s not talking about what AI will do someday. He’s showing what he does with it today:
- The meeting recorder that extracts culture
- The hiring evaluator agent
- The Grok searches for niche talent
- The fluency rubrics
Each of these is a system, not a prompt. Each of these required someone to build it, test it, iterate on it. Each of these required learning to row.
What Rowing Actually Looks Like
It’s the same thing I’ve been finding in my own work. The linter-bot I built to police my coding agents. The constitutional framework I use to govern AI behavior. The replay prompts I save so I can regenerate features cleanly.
None of it is glamorous. All of it compounds.
Let me be specific. Here’s what “learning to row” looks like in my projects:
Aegis Framework
I built an entire governance system for AI agent development. Not because I wanted to. Because after burning two prototypes to the ground from “vibe coding,” I realized that without constitutional constraints, AI agents drift.
The framework has:
- Execution modes (lean, strict, generative)
- Agent behavior profiles
- Drift detection tools
- Snapshot testing for pattern fidelity
It’s unglamorous infrastructure work. It’s also the only reason my AI-assisted projects don’t collapse under their own weight.
CLAUDE.md Files
Every project I work on now has a detailed instruction file that defines the project’s soul.
Tech stack constraints. Diff budget policies. Auto-rejection triggers. Test contract requirements.
The photography gallery project lists every Svelte 5 rune pattern the agent must follow and defines which files require “thorough mode” vs “direct mode.” The ACP tools project locks the entire dependency stack—“DO NOT substitute these dependencies”—and defines a strict implementation order that agents cannot violate.
Intent-Driven Engineering
I codified a methodology for AI-human collaborative development that prevents tool concurrency explosions.
Sequential feature execution. Phase-based task decomposition. Human approval gates between major features.
It sounds obvious. It took three weeks of failure to articulate.
That’s what rowing feels like.
The Dissolution of Syntax
Here’s the connection I keep making to the broader shift.
I wrote recently about the dissolution of syntax—the idea that we’re transitioning from an era where code is the artifact to one where intent is the artifact. The syntax becomes something AI handles. What matters is whether you can articulate what you want and govern how it gets built.
Foster’s approach is this principle made operational.
He’s not delegating “use AI” to his organization. He’s building the governance layer—the fluency rubrics, the culture audits, the structured agent workflows—that makes AI use actually productive.
Most organizations are still sending memos. They’re tourists in the AI landscape. They visit, take photos, and go home unchanged.
The practitioners are building infrastructure.
The Tourist Problem
Here’s the uncomfortable truth I keep landing on:
The people who are “waiting for AI to mature” before really engaging with it are making a bet that the learning curve will flatten. That they’ll be able to catch up later when it’s more stable, more predictable, more like a traditional software tool.
I don’t think that bet pays off.
The gap between tourists and practitioners is widening precisely because the practitioners are building muscle memory that compounds. Every system I build teaches me something I couldn’t learn from reading about AI. Every failure reveals constraints I wouldn’t have anticipated.
You can’t delegate that. You can’t memo your way to it. You have to row.
What This Crystallized
Foster’s interview put words to something I’ve been circling.
The CEO has to be a practitioner. Not the best practitioner. Not the most sophisticated user. But someone who uses the tools, makes the mistakes, and shows the work. I’ve seen what happens when leadership stays abstract—the whole organization stays abstract.
Fluency needs rubrics. I’m done saying “embrace AI” to my teams without defining what that actually means. What does good look like for a PM? For an engineer? For a designer? If I can’t answer that specifically, I’m just hoping people figure it out.
Culture audits work on AI too. The gap between what you tell your agents to do and what they actually do is the same gap between stated values and actual behavior. Both require signal processing. Both require someone paying attention.
The economically unviable is now viable. The governance work, the documentation, the meta-layer stuff I would have skipped before—that’s where the compound interest lives now.
I don’t know if “no room for tourists” is the right way to say it. Maybe it’s too harsh. Maybe it sounds like gatekeeping when it’s meant as an invitation.
But the boat is moving. The people rowing are getting better at rowing. And the distance is growing.
That’s the thing about this moment. You don’t have to be the fastest. You just have to be in motion.