Early in his talk, Georgios Chatzimilioudis, a Solutions Engineer at Oracle, asked the audience to raise a hand if they expected their business to use more AI in three years than today. Hands went up. He asked who expected to use more than one model, and the hands stayed up. Then he asked who was confident they knew which model, chip, platform or framework would matter most to their business in three years. Every hand came down.
This set the tone for the inaugural Tech Race Summit, organised by SOFTSWISS, which brought together over 1,500 participants from fintechs, e-commerce companies, iGaming operators and other digital players facing the same pressures in high-load industries.
Organisational gap in AI adoption
Businesses worldwide face significant uncertainty, given the breakneck speed of technological innovation. Beyond just keeping up with the newest AI-based tech, they are also facing challenging organisational and strategic decisions about how and where to implement AI to ensure the biggest positive impact on the bottom line.
Buying access to a model is straightforward, but giving it reliable company data, deciding what it can do and proving its work justifies the expense is harder.
Sergey Khurson, Principal Solutions Architect at AWS, said 88 per cent of organisations use AI in at least one part of their business, but only 8 per cent have reached advanced maturity, where AI runs in production across the business rather than in isolated pilots.
The 2027 iGaming Trends Report, based on input from more than 500 industry professionals and 65 experts, points to the same gap, with AI moving deeper into engineering, fraud prevention, customer support, compliance and infrastructure. The challenge is shifting from whether companies can deploy AI to whether they can integrate it reliably, govern it and demonstrate measurable business value.
One solution discussed was to define work specifically around AI tools with strict boundaries that guard against project creep and protect value for money.
Pilots without a business case
There is also another curious issue facing businesses, mainly that AI experiments can survive long after their purpose has become unclear. Viktor Vedmich, Senior Solutions Architect at AWS, described “zombie projects” that continue indefinitely, consuming resources without a decision on whether to expand or stop them.
Vedmich argued for shared platforms between teams and fewer projects, chosen by potential value and the effort required. Before work begins, a company should define what success means and when management will decide on a wider rollout.
Putting programmes under the CEO or COO, with targets tied to revenue and margins, makes those decisions harder to evade. It also forces the business to confront whether it is automating work that is worth doing in the first place.
Managing data in the AI age
Another problem is that a capable model has limited value if it cannot find or interpret the information it needs. Oracle’s Chatzimilioudis cited a Gartner report, suggesting that 63 per cent of organisations lack the data management practices AI requires. His remedy was to give data clear owners, record its origins and set rules for access. For companies holding sensitive financial or customer data, he argued for bringing AI to the data, rather than repeatedly copying information into new stores.
Cloudflare put that approach into practice, starting with finding out what employees needed, according to Field CTO Michael Tremante. The company set up an email address where staff could ask an agent team for help. This helped create a shared context layer, built around a GitHub repository, which supplied knowledge and reused across tasks.
The result was a central source of information and data, which drove adoption and laid the foundation for more complex tools.
Ensuring usability for all
The next step was giving staff more powerful tools, so Cloudflare preconfigured the open-source tool OpenCode with enterprise model accounts, which helped engineers. However, non-technical employees struggled with setup and authentication, so the company then built Cloudflare OS, an internal tool intended to be usable from an employee’s first day.
Tremante said that rolling out Cloudflare OS meant that a new marketing hire, for example, could write a blog post in the company’s voice on day one, because the tool already held all corporate guidelines, key facts and approval process.
Cost of access and responsibility
Beyond the usability puzzle, the cost is the big elephant in the room. Choosing the right model for each task is key when costs are hard to predict.
Cloudfare’s Tremante questioned the logic of sending every request to the most powerful, and expensive, option when simpler models can often do the job. To address this, the company routes requests between models and runs open-source alternatives on its own infrastructure to keep costs down.
At the Tech Race Summit, the CTO panel was divided over whether productivity gains justified current spending. Some warned that modest improvements would eventually be difficult to finance; while others saw the cost as small compared to what their best engineers could now produce.
In addition to the cost issue, understanding who is held accountable when AI takes on more workstreams, was another theme that loomed large at the conference. The issue of responsibility is especially key, given that the reliability and veracity of AI output remains difficult to verify.
Andrey Doronichev, the former Google executive now running AI-native pharma company Bioptic, shared one approach to improve quality of AI-based output. At Bioptic, human pharma experts maintain a bank of questions and answers used to test agents and improve their performance, effectively defining what a good result looks like as the system is built.
This is one way of addressing the quality issue to enable AI to produce more reliable results. However, the industry consensus is that ultimate responsibility for the work performed by AI will remain with the business.
Tech Race Summit delegates agreed that for now, there needs to be a human in the loop, ensuring that an output is good enough to deliver, and understanding the process of how it was achieved, and how much it cost well enough to make that call.
To become a Tech Race Summit partner, contact the team under info@techracesummit.com. To share your technology experience or a case study on stage, apply to become a speaker here. To join Tech Race Summit 2027, get an early-bird ticket on the website.