News

Arthur Azizov is the Founder of B2BROKER Group and B2BINPAY.

getty​The past couple of years I’ve spent introducing AI into my own company have taught me that buying the technology is just a small first step in a much bigger process—one I seriously underestimated at the beginning.

We tried several approaches. Some worked, some didn’t, and team size mattered much more than I had expected. We started with no clear plan, tested early tools, bought business subscriptions and later told the entire company that AI was now part of the workflow. On the one hand, people had finally got access to the tools. Yet, time passed, and the company itself hadn’t really changed.

This forced me to distinguish individual use of AI from genuine companywide adoption. We had to stop treating AI adoption as yet another software rollout and start changing teams, handoffs and responsibilities, as well as working habits around it.

If I were advising a tech founder beginning this process today, here are some of the lessons I’d share first.

1. Don’t roll AI out everywhere at once.The first mistake I can highlight now, looking back, was trying to make adoption happen everywhere too early. I was trying to push things too quickly and got ahead of myself. The problem is, when everyone receives access to AI at once, it becomes difficult to understand what’s actually working and what’s not. Because teams use different tools and employees have different levels of interest, productivity gains are hard to separate from enthusiasm around a new product.​

Once we narrowed the adoption process down to one team, the progress became more tangible. Suddenly, we could compare output, turnaround time, quality and cost, as well as see who was genuinely interested in AI and who was simply following a company instruction. ​

This is how I figured out that the best people to lead an AI rollout were those who were eager to experiment with the technology. A founder sets the direction and determines business goals, while the people closest to the tools should retain considerable influence over the entire workflow.​

I’d also keep the planning horizon as short as possible. AI is a fast-evolving technology, so a three-year road map can become outdated long before those three years are over, which is why, for us, a quarter is more useful. Define the objective, metrics and budget for the next few months, then when a major model release materially changes, reassess what your team is capable of doing. Planning matters; overcommitting to today’s implementation does not.

2. If your roles haven’t changed, you’ve only bought software.Giving an employee a better tool doesn’t automatically change the company around that employee, and this insight ended up shaping how we approached development.​

We used to organize work around relatively limited roles and handoffs: engineers, testers, analysts, front-end specialists, mobile specialists and so on. As AI tools grew more capable, some of those boundaries started to make less sense. One person could cover a wider part of the process, provided that person understood how to work with AI and remained responsible for the final result. ​

So we started changing the roles themselves. In development, the concept of an AI engineer became broader and much closer to a full-stack role than the jobs we had before. It helped us reduce handoffs and made ownership clearer. By August 2026, our development speed was roughly five times higher than at the end of 2025.​

And what did that fivefold increase show us? For me, it was a strong indicator of whether AI adoption was changing the business from within. If, a year after a serious rollout, every job description, approval step and reporting line still looks exactly the same, then AI has most likely been added on top of existing processes rather than been embedded into the company’s internal processes.

3. AI gets more expensive as it gets more useful.Here comes one of the most interesting parts: the cost of adoption.

As usage expanded, our AI spending grew by roughly 15 times. There were more users, workflows became deeper, and naturally, employees started to prefer the most capable models available on the market. Of course, better models cost more, and once people become dependent on high-quality models, maintaining that level of quality turns into a tangible expenditure item.

This is why model routing matters. Companies need rules for deciding which requests go to which models, while some requests should not consume AI resources at all. Define those rules before the bill defines them for you. Plan budget in terms of consumption, because the usage will continue to grow as adoption proves successful.​

AI can also raise the cognitive load on employees. They may do less manual work, yet they have to define tasks, break them into pieces, review more output and make more decisions at a faster pace. So while productivity may rise, so does the amount of mental processing required from strong employees.

The real job is to redesign continuously.Overall, what I came to after years spent on testing is that AI adoption at a company level requires a continuing redesign of how the entire company works.

Models will change, prices will change, and roles that make sense today may look unnecessarily narrow a year from now. After all, the real advantage is to build a company that can keep changing its processes, responsibilities and spending as the technology changes around it.​

Forbes Business Council is the foremost growth and networking organization for business owners and leaders. Do I qualify?