---
title: "Why Your AI Initiative Is Stalling (And It's Not the Technology)"
url: https://stacklist.com/card/6ed457df-1a6c-4453-8897-9f8a290b67a4
source_url: "https://wildfirelabs.substack.com/p/why-your-ai-initiative-is-stalling"
stack: https://stacklist.com/stack/f3725201-e51d-4977-b61a-61e523b68980
summary: "Why Your AI Initiative Is Stalling explores how leadership failure, not technology, causes AI adoption to stall, and presents a phased approach to building an AI-first culture by setting clear expectations, protecting learning time, and encouraging experimentation. The article emphasizes that executives must participate in AI tool usage themselves and create psychological safety for teams to fail and learn."
tags: "ai-adoption, change-management, leadership, team-transformation, claude, workflow-automation"
key_entities: "Todd Gagne (person), Mike (person), Claude (technology), ChatGPT (technology), LLM (technology), AI-first company (concept), change-management (concept), psychological-safety (concept), team call in March (event)"
classification: "analysis"
content_hash: "sha256:2135db0a714f7f8d87a397c8f6c0a3e88ab1919a09a5dae784c382dd825b865b"
acp_version: "0.2"
token_counts_approximate: 3124
visibility: public
agent_accessible: true
status: "final"
---

# Why Your AI Initiative Is Stalling (And It's Not the Technology)

Why Your AI Initiative Is Stalling (And It's Not the Technology) What happened when we made everyone pick a real project — and gave them permission to fail Todd Gagne May 26, 2026 6 1 Share I was on a team call in March when it hit me. I’d spent the last several months deep inside Claude — built skills that automated my content pipeline, created a chief of staff workflow that prepped me for every meeting, had a Saturday morning sequence that ran analytics, generated content, and scheduled posts across platforms without me touching it. I was living inside the future of how small teams could operate. Then someone on the call asked a basic question about how to use an LLM, and I realized: nobody else on my team had had their moment yet. The gap wasn’t intelligence or willingness. It was exposure. And it was my fault. You can’t build an AI-first company with a team that hasn’t touched the tools. And you can’t blame the team when leadership hasn’t led the change. That call was a mirror. I’d been sprinting ahead, building my own workflows, getting faster every week — and I’d left everyone else standing at the starting line wondering what any of this had to do with their job. So I wrote a memo to my partner Mike. Not a strategy deck. Not a vendor evaluation. A phased plan for getting our entire team from “I’ve heard of ChatGPT” to “I can’t imagine working without this.” What followed taught me more about change management than technology. Phase Zero: Leadership Sets the Tone Before anyone on the team opened a single AI tool, we had a conversation. Not a training session. A conversation about why we were doing this and what it meant for them. The instinct is to skip this part. Buy the licenses, send the Slack message, link a YouTube tutorial, and assume people will figure it out. That’s how you get 15% adoption six months later and a team that quietly resents the mandate they never understood. Three things got said before anyone opened a tool. First: this is not about replacing anyone. This is about eliminating the manual, repetitive tasks that fill your week and elevating the work you actually get to do. The goal is to make your job more interesting, not to make you redundant. Second: this is a learning opportunity. We’re investing in you. Three to five hours a week of protected time to experiment. Not side-of-desk time you’ll never find. Real, carved-out hours that leadership is accountable for protecting. Third — and this was the hardest one to make people believe: failure in this process is not just acceptable. It’s expected. If you try to automate something and it doesn’t work, tell me what you learned and what you’d try next. That’s a win. If you’re not failing, you’re playing it too safe, picking the easy stuff, not pushing into the territory where the real value lives. That last part matters more than any tool selection or workflow design. The moment someone on your team feels like their failed experiment will be held against them, they stop experimenting. And a team that stops experimenting with AI in 2026 is a team that falls behind permanently. Leadership has to participate, not just sponsor. I showed my work alongside everyone else. Mike picked a project too. If the executives announce an AI initiative and then don’t use the tools themselves, the message is clear: this is a mandate for you, not a belief we share. That kills adoption faster than any technical barrier. Phase One: The Art of the Possible We picked Claude as our platform — not because it’s the only option, but because standardizing on one tool meant everyone could help each other. A shared Slack channel became the exchange: tips, frustrations, screenshots of things that worked and things that broke spectacularly. The first week, everyone had one assignment: pick a project from your actual work — not a demo, not a sandbox exercise — and post it in the channel. Something real. Something that, if it worked, would change how you spent your Tuesday. Our operational lead was the person I was most worried about. She’d been slow to adopt any technology change over the years — not just LLMs, any new system. When we asked her to pick a project, she chose something that had been bugging her: our weekly candidate review process. Every week, the team sat in a meeting reviewing Wildfire Labs applicants, treating them all the same — same time, same depth, same discussion — regardless of whether someone was clearly a strong fit or clearly not. She researched it, experimented with it, and built a scoring system that stack-ranked candidates before the meeting. The strong ones got deep discussion time. The weak ones got a quick pass. The meeting went from a grind to something useful. She was proud of it. Genuinely proud. And the rest of the team saw it and thought: if she can do this, I can do this. That’s Phase One working. Not the tool. The moment someone who didn’t think this was for them discovers that it is. We ran weekly meetings to check progress — not status reports, but real conversations about what was working and what wasn’t. Champions emerged naturally. The people who took to it fastest became the ones carrying the message, answering questions in Slack, showing colleagues how they’d solved a problem. You can’t manufacture that. You can only create the conditions for it. After a couple of weeks, we did a demo day. Everyone showed what they’d built. It created accountability but more importantly, it created a shared vocabulary. The team could now talk about what AI was good at, where it broke down, and what was worth pursuing. Phase Two: If You Had to Automate 50% of Your Job This is the question that changes the conversation. Not “what can AI do?” — that’s too abstract. “If you had to automate half your job, where would you start?” — that’s specific enough to produce action. Everyone writes down every task that’s a candidate for automation. For each one: how much time does it take, how often do you do it, and what’s the priority? Then you start working down the list inside Claude, building workflows that handle the repetitive parts while you focus on the judgment calls. The weekly accountability matters here. Not micromanagement — measurement. How much time did automation save this week? Where is the value showing up? What’s still manual that shouldn’t be? But here’s the part most companies get wrong, and it’s the part that determines whether your team leans into this or quietly sabotages it. When automation starts working, people get nervous. Our operational lead started to feel it. The bucket of weekly tasks that used to fill her time was shrinking. She knew the math. If the tasks that justify her role are now automated, what justifies her role? This is the leadership moment. If you don’t have an answer for that question — a real answer, backed by action — you will lose people. Not because you fired them. Because they’ll stop automating. They’ll slow-walk adoption. They’ll protect the manual tasks that make them feel necessary. And you’ll never know why your AI initiative stalled. We moved her into events. First, building a process for our startup weekend — something she’d never done, something that stretched her. We ran a couple, refined the process together, and now we’re layering AI automation on top of the process she built. She went from nervous about not being busy to owning a new domain. The principle is simple and most companies violate it: when you automate someone’s tasks, you have to fill the space with growth. Not more busywork. Growth. Additional education. New projects. Expanded responsibilities. Things that make the person more valuable, not less. They need to feel that — in the assignment, in the conversation, in how you talk about their contribution. And your actions need to back it up. If you tell people “this is about elevating your work” and then cut headcount the moment automation kicks in, you’ve taught the entire organization to never trust you again. Phase Three: From Personal Hacks to Independent Agents Phase One and Two are personal. The skills and workflows live on individual machines, tied to individual accounts and individual data sources. That’s fine for experimentation. It’s not fine for operations. Phase Three is where personal hacks become organizational infrastructure. The skills that proved their value get hardened — decoupled from any one person’s setup, made reliable enough to run independently, and centralized so the team operates from shared assets instead of individual workarounds. We saw this with our chief of staff workflows. Multiple people had built their own versions — different approaches to meeting prep, different ways of pulling context, different formats. Some were better than others. In Phase Three, we consolidated: one chief of staff pattern, informed by everyone’s experiments, that the whole team could use. Before I send anyone into this phase, I need to tell you what I got wrong three times. I built my chief of staff workflow with Zapier and N8N first — fragile integrations across three platforms. Then with Claude skills — better tools, same mistake. I jumped into building without thinking through edge cases. What happens when the meeting doesn’t have an agenda? When the contact exists in one system but not another? When the data is stale? Rework on top of rework. The version that actually works started with planning. Sixty percent planning, forty percent building. Your instinct will be reversed. You’ll want to start prompting, iterating, seeing output. Resist that. The time mapping the problem — what triggers this, what are the inputs, what are the edge cases — saves rebuilding the thing three times. The infrastructure matters here. We run our centralized agents on OpenClaw in an NVIDIA NeMo environment. That sounds technical, but what it does is simple: it gives us one place to manage cron jobs, error handling, and security across all automated tasks. When a skill breaks at 2 AM, I know where to look. When a new team member joins, they inherit the same tooling as everyone else. Two things become critical at this stage that most teams underestimate. Cost management. Different tasks need different models. Simple work gets a smaller model, judgment-heavy work gets the best available. Segmenting which LLM handles which task is the difference between an AI budget that scales and one that explodes. Consistency. When everyone builds their own workflows, style guides get interpreted differently, code review standards drift, and outputs vary by who built the automation. Phase Three is where you enforce shared assets — one style guide, one review process, one set of standards — across every automated workflow. The output should be indistinguishable regardless of which team member’s skill produced it. What This Is Actually About Each phase earns the next — you can’t plan workflows in Phase Two without the intuition from Phase One, and you can’t justify the infrastructure investment of Phase Three without validated workflows from Phase Two. But if I strip away the phases and the tools and the infrastructure decisions, the thing I actually learned is simpler than any of that. This is a change management problem. It has always been a change management problem. The technology is the easy part. The hard part is getting a team of people — each with their own relationship to technology, their own fears about relevance, their own definition of what their job is — to move together into a fundamentally different way of working. That requires leadership, not mandates. It requires sensitivity to the fear that lives underneath every adoption curve. It requires creating an environment where someone can try something, fail publicly, and have that failure treated as progress rather than evidence. It requires you to answer the question that nobody asks out loud but everyone is thinking: if AI can do my job, what happens to me? If your answer is silence, or worse, if your answer is eventually proven to be a lie, you don’t have an AI adoption problem. You have a trust problem. And no tool fixes that. Our operational lead went from the person I was most worried about to one of the strongest voices for how we use these tools. Not because the technology convinced her. Because she felt safe enough to try, skilled enough to succeed, and valued enough to keep going when the ground shifted. That’s the operating system. Not the AI. The environment you build around it. 6 1 Share