What It Actually Takes to Put AI to Work in Your Company

Two months after its launch, ChatGPT became the fastest-growing consumer application ever, reaching 100 million monthly active users in record time. That kind of adoption signals something more than a trend — it signals a shift in what's expected of software teams.

Rudolfs Heimanis is an X-Team software engineer who works for a large media organization, where his job is to help the partner implement AI in ways that genuinely fit their business. He came to the field not through a graduate program in machine learning but through curiosity and a willingness to experiment — and he has strong opinions about where companies go wrong before they even write a single line of AI code.

In this story, Heimanis explains how companies can use AI to become more efficient without eliminating jobs, what the real risks look like and why he believes a proof of concept is always the right first move.

AI Is a Tool, Not a Replacement

For most of his career, Heimanis assumed that working seriously with AI required a deep understanding of the math underneath it. That assumption kept him at arm's length. Then he started experimenting with OpenAI's general-purpose offerings, and his thinking changed fast.

"When I experimented with the general purpose offerings from OpenAI, I couldn't believe how easy they were to use," he says. "ChatGPT showed me that you don't need to know complex math. There are so many layers of abstraction above the math that, sometimes, you only need ten lines of code to load a basic open-source model that solves your specific problem."

That hands-on approach — learning by doing rather than watching — led him to a broader point about what AI can do for companies. He sees three distinct use cases worth pursuing. The first is assistance: a salesperson who currently spends hours manually researching prospects could instead use an AI tool that surfaces all available information about a prospect along with the product features most relevant to them. The second is automation: a machine learning model trained on real employee behavior can speed up time-consuming compliance flows, like Know Your Customer verification, by generating a customer trust rating that reduces manual review. The third is content creation — not full delegation to a large language model, but using one for ideation, first drafts or finding better ways to phrase things for a specific audience.

The common thread, Heimanis argues, is that none of these scenarios eliminate people. "AI shouldn't be thought of as something that will eliminate jobs," he says. "Instead, it's a tool to enhance people's existing capabilities, to make them more productive, and to have them spend less time on tedious and repetitive tasks."

The Risks Companies Underestimate

Heimanis is quick to acknowledge that enthusiasm for AI needs to come with clear-eyed risk management. The most underestimated risk, in his view, is also the most mundane: data privacy.

Companies considering third-party AI providers need to carefully review their privacy policies. Some providers use company information to train their models; others retain data for a period of time for debugging purposes. For any organization operating under frameworks like GDPR, those details matter. "As a general rule, if you're working with sensitive data, don't use the general purpose AI models that are available on the internet," Heimanis says. "Train your own model that's based on an existing and trustworthy open-source model." He points to most solutions provided by Google's Vertex AI as examples of providers that keep data in the client's control — though even then, he recommends verifying any claims independently.

The broader concern about AI displacing workers doesn't worry him in the same way. His argument is structural: every generative AI system has been trained on content that humans created. Replacing human-generated content with AI-generated content would produce factual inaccuracies and a sharply limited worldview. "When civilization evolves, so do our problems," he says. "We can't rely on AI exclusively because we'd be solving new problems with old information."

He also pushes back on complacency. Individuals still need to invest in their own skills. AI will create new job categories — and the people best positioned for them will be the ones who kept learning.

How to Start Without Wasting Money

The practical framework Heimanis recommends for companies begins well before any technology decision. It starts with a clear understanding of what the business actually sells — not the product itself, but the value it delivers to clients — and then moves into genuine conversations with employees across departments about the friction in their daily work.

Once a workflow worth improving has been identified, he recommends imagining a solution and benchmarking the current process before building anything. How long does the task really take? How often does it happen? Those numbers don't need to be perfect — they just need to exist as a baseline.

The next step is finding an AI partner with proven experience on similar problems and defining the smallest possible MVP together. "My colleagues and I always start with a pilot project," Heimanis says. "We ask ourselves, what's the most amount of value we can provide our partner for the least amount of work? What's the best and most cost-effective solution to solve the partner's problem? That's where we start."

After the pilot is running, the validation questions are simple: how does the solution compare to the benchmark, is the team satisfied with it and was it worth the cost? If the answer is yes, the path forward is iterative — scale the solution, add useful functionality and let it grow with the business. "There's tremendous value in AI," he says, "and I encourage you to at least experiment with it." A proof of concept, he notes, costs almost nothing. The risk of not trying is higher.

Ready to build work you're proud of? Apply for an open role at X-Team.