Alex Mazer, CEO of Common Wealth, a Flow Capital portfolio company, recently shared a detailed account of the company's AI adoption journey. It's an excellent piece, candid about what worked and what didn't, and a must-read for mid-sized businesses working through this challenge.
His article prompted us to reflect on our own experience at Flow. In doing so, we naturally began to notice where our approaches were similar and where they differed, raising a broader question: is there a one-size-fits-all approach to AI adoption, or does each company need to find what works for its size, culture and operating style?
Our experience suggests the latter.
Common Wealth’s article describes a more formally structured approach to integrating AI, guided by the goal of adopting it quickly and using it with care, judgment, and quality. They built a serious program that is carefully planned and well executed. It is clearly producing the intended results: faster product deliver, better client service, and a collaborate company-wide approach to learning how to deploy and integrate AI, together. Although our approach developed more organically, without a formal objective at the outset, we found ourselves working toward much the same outcome.
We're a small team, and at the beginning, there were no adoption targets, no formal training programs, no restrictions on platforms or experimentation. Instead, we encouraged the team to lean in aggressively and adopt AI where possible. On our weekly standup meeting, we encouraged people to share how they were using AI, and, occasionally, we would organize informal presentations, particularly when someone built something that was proving to be useful. We provided subscriptions and resources for anyone who wanted them, and consistently encouraged people to "do more with AI”.
Early progress came from a few early adopters who, interestingly, were not particularly tech-savvy and had no formal training in programming; they just started building. They experimented, created tools that solved their day-to-day problems, and tested what worked and what didn't. They brought those experiences back to the team and made it all feel approachable rather than intimidating. Their enthusiasm and progress encouraged others, who started experimenting themselves, often with some guidance from these earlier adopters. Seeing colleagues create practical tools prompted even more people to try, and the circle kept widening. As we progressed, we enlisted the help of an outside group to speed our progress and formalize some architectural decisions. My job was to encourage adoption, and make sure people had the time, resources, and freedom to experiment.
What started as individual initiatives became, step by step, a company-wide habit. Today, roughly eighteen months in, AI is a meaningful part of daily work for every member of the team.
The claim “everyone uses AI” is easy to say, but vague in meaning and light on details. There is a pressure to be performative about AI, and to make grand pronouncements designed to signal how far ahead you are. This isn’t my intent here, and you will not see us posting a headline like "we deployed twenty agents in a month." What we have instead is quiet adoption, but a dramatic change in culture and efficiency. We have AI built tools in daily use, people at every level of the business building/doing things they couldn't have built pre-AI, and work that gets done faster and better than before. Our progress is measurable and accelerating.
Rather than cataloguing individual projects, the clearer picture is the pattern.
It watches, gathers, and preps. We use AI to monitor markets, news, and competitor activity continuously, and run structured research on specific niches. Before a call with a company, it assembles what our CRM holds alongside external research, so we arrive prepared. This is work that previously happened sporadically and consumed hours or didn't happen at all.
It researches and produces first drafts. Posts, articles, campaign copy. Every piece starts from something one of us wants to say, and everything is edited post AI, and before it goes anywhere. AI accelerates the drafting; it doesn't originate the thinking.
It handles the repetitive plumbing. We've built extensive workflows to automate recurring processes, keep our data clean, and maintain visibility on where everything stands. Reporting (via ad-hoc dashboards linking disparate information sources) is probably our highest-return example: what used to consume hours across the organization now takes minutes, and materially improves decision-making. It supports one of our key organization wide principles: you can’t manage what you don’t measure.
It surfaces new insights. There was never a shortage of ideas on information that could (if accessible) help us make better decisions. Now, with our AI tools, we can gather insights that were previously impossible, or prohibitively expensive, to obtain.
It spreads knowledge instead of siloing it. We've standardized our AI setup across the team: common plugins, skills, and context files that give every department the same foundation. The practical effect is alignment. When departments collaborate, both sides work in the same environment with the same underlying context, often simultaneously and without needing to be in the same room. It has also shortened onboarding: new hires learn our business, strategy, market, and products at their own pace, with AI answering questions and testing their understanding.
It supports rather than replaces human judgment. AI helps us reach better-informed decisions by flagging what needs attention, gathering the relevant material, and summarizing it so the decision-maker starts further along. The practical effect reaches the companies in our pipeline: internal processes move faster, which means companies get to funding decisions sooner.
Make the investment (in time and tools), even knowing it will age fast. Some/most of what gets built will be obsolete sooner than expected. In one project, we're already on our third architecture. Not only is this OK, but it is also mandatory. The tools were never the real asset, they were building a foundation of learning inside the team, and that compounds even as the tools change.
Don't wait. The exact structure matters less than getting started and learning from practical use. The companies falling behind are those still deciding how to start. Waiting for the dust to settle is the wrong strategy.
Start small. Don't begin by overhauling core processes. Automate mundane, repetitive tasks where the value shows up immediately. Leverage what AI is good at to build simple, new things that were not possible before. These early wins build the confidence and skills to take on bigger things later, without putting anything critical at risk while everyone is still learning.
Encourage and reinforce constantly. The encouragement must keep coming, especially through the stretch where early enthusiasm fades and the harder work of changing actual workflows begins.
Celebrate the successes, and don't dwell on the failures. When someone builds something helpful, make sure it's shared. Examples spread adoption far better than mandates do, and every tool that gets shown off inspires the next builder. The flip side matters too: not everything will work, and that's okay. Some experiments will fail. Treat those as the price of learning, not as reasons to pull back.
Be critical. An idea from Common Wealth's article worth repeating is that informed skepticism isn't an obstacle to AI adoption; it's part of the critical thinking that makes it work. The people who questioned the tools hardest often built the best things with them, because they tested rather than assumed (yes, hallucinations are still a thing).
Think about the people. AI can feel inaccessible (looking at terminal outputs, for example), and for many people it feels threatening. With layoffs in the news every other day, adoption can't be pushed in an environment where people wonder if they're training their replacement. Our message from the start was that AI is here to make our people more capable, not to replace them. The team is doing more and doing it better. Nobody's job was automated away, as that was never the objective. The goal is a small team punching above its weight, not a smaller team.
Adopting AI internally has changed how we evaluate it externally. We lend to growth-stage technology companies, and for every business in our pipeline, the same questions now matter: is this company’s business at risk from AI, or positioned to benefit from it? Could AI tools disrupt its product, compress its margins, or take its customers? Or could they become faster, leaner, and harder to compete with? Those questions are difficult to answer from the outside. Knowing the tools firsthand, having some idea of what they can genuinely do today, what they still can't, and how people use them, gives us a much better read on which businesses are at risk and where the real opportunities might sit. That judgment is now part of how we underwrite.
You can’t “grassroots” forever. As we grow and our capabilities expand, we are moving toward more formalized architectural decisions. But now, we are making those decisions based on insight and experience, not clickbait X posts.
Reflecting on our experience reinforced an important lesson: start now. AI adoption is worth investing in, encouraging consistently, and sticking with it through the unglamorous middle stretch where tools break, experiments fail, and the novelty wears off.
As might be expected, as our use of AI expands our approach is becoming more formal, structured, and rigorous. We are now making more deliberate architectural decisions, and integrating AI more deeply into our core systems. At the same time, the team’s buy-in and enthusiasm remain genuine. The approach we took to arrive at this point worked particularly well with our corporate culture.
The approach to adoption of AI yours to choose, but the one lesson to take away is that commitment isn’t optional.
