Almost every person and business on the planet is using some form of AI, especially products developed by OpenAI or Anthropic.
Startups are no different. A report by the European Central Bank has found that 56% of young firms, under five years old, were at an advanced stage of AI adoption.
Despite this, research shows that many startups struggle to use AI effectively, with 52% of surveyed startups reporting difficulties with AI integration and data quality, according to General Catalyst’s 2024 report.
Many AI companies have caught on to the struggles that founders are experiencing and have responded by launching initiatives to better support them.
One of these companies is OpenAI. The startup launched DevDay, an annual developer conference to bring together technical founders and software engineers around the world to preview new AI models, developer tools, and platform updates.
DevDay 2026 London saw many discussions and breakouts focusing on updates to OpenAI’s systems, as well as discussions around how founders and engineers can better protect their systems.
At the event, Tricia Troth, Head of Go-to-Market (GTM), Startups EMEA at OpenAI, provides more insight into how OpenAI supports startups and allows founders to build differentiated, trustworthy AI products.
Scaling success with AI
OpenAI has been a foundational platform for many startups, with research from Kruz Consulting reporting that 57% of venture-backed startups were using OpenAI products in September 2023.
From Troth’s perspective, the biggest difference between startups that are successfully scaling with AI and those struggling to gain traction comes down to how well founders understand a real problem and identify the technology needed to solve it.
“You really need to understand the product or the problem you’re trying to address for your end customer. When you have that deep specialism in a field, whether from an industry or vertical perspective, you understand the customer’s challenges and can then bring in technology to help accelerate solutions,” she says.
Troth provides the example of Anima, a UK-founded healthcare technology startup that uses AI to help GP practices and clinics manage administrative work.
Anima integrates OpenAI’s models directly into its products through the OpenAI API, and OpenAI provides the underlying intelligence to the platform.
“[Anima] was founded by an ex-medic who worked for the NHS. He saw some of the challenges within the NHS. As we know, it’s relatively stretched, with limited resources and budgets, so he set out to use technology and AI to address these issues,” she says.
“They work with us and leverage ChatGPT Live to provide what I suppose you could describe as a digital assistant. It’s essentially a chat agent interface,” she says.
Anima handles the initial conversations with patients before escalating cases to clinicians. It works in line with clinically approved workflows and cannot operate outside guardrails.
“What’s really exciting is that, when we did a hackathon with [Anima] two weeks ago, they told me that waiting times used to be around 40 to 50 minutes, but they’ve managed to bring them down to just 10 minutes,” she says.
Troth says that Anima shows that successful startups don’t just use AI because it’s popular; they use it to solve a problem they understand first-hand.
Real-world cases of agentic AI
Troth was keen to share many of the early-stage agentic AI startups and founders that OpenAI is working with.
“One company whose story I particularly love is Parloa, because of how specialised its technology is,” she says.
“What they’ve built is incredibly sophisticated. They’ve taken the time to develop a fleet of specialised AI agents that can look at things like customer account logins, order histories, and inventory,” she says.
Parloa has also built multiple deterministic guardrails and controls to ensure the agents operate as intended, only access the information they’re permitted to, and behave as expected.
Troth says that Parloa’s synthetic conversations, which test its agents repeatedly before they go live, are what make it an impressive startup.
“That’s a very different and challenging development process, but they’ve managed to get it absolutely right, which is why they’re now one of the biggest and most successful startups we see in Germany,” she says.
Parloa reported to Troth that after deploying the technology into a traditional call centre operation in the travel industry, they saw an 80% decrease in the number of human agents required.
Founders’ concerns with AI
When asked about the growing concerns founders have when using AI because of issues around the climate and discrimination, Troth says that these are legitimate concerns but something that OpenAI is working with founders to address.
“I think we all have a responsibility, particularly those of us at OpenAI, to prioritise safety, alignment, and trust. This also includes considering our carbon emissions footprint. That’s our fundamental starting point,” she says.
“However, I work with founders and startups day in, day out, and I see an enormous amount of positivity and good coming from AI,” she adds.
“As a woman in tech, I know that I’ve sometimes felt woefully underrepresented. It’s important that more women get involved, and that’s something OpenAI is trying to support through the work we do,” she says.
OpenAI spends a lot of time working with founders and startups to help them consider issues around AI, Troth says.
“In order to have a great product, trust has to be built into it. Otherwise, it simply doesn’t work.”
How to differentiate yourself from other AI startups
Troth says that as AI is becoming more widely available, startups need to find ways to stand out by combining AI with specialist industry knowledge, as well as an understanding of their customer needs.
Troth uses Model ML, a startup that develops AI-powered financial management and board reporting tools, as an example.
“Model ML currently leverages our Astra model suite, a family of advanced AI models designed to process large volumes of information and generate accurate, high-quality outputs,” she says.
“One of the things they look at very closely is how to produce financial management and board materials. These need to be highly accurate, process large amounts of data quickly and deliver high-quality outputs,” she says.
Troth says that Model ML rigorously tested different AI models to find the best balance between cost, speed, and quality. While it found OpenAI to be at least a third or more cost-effective than Anthropic, it also challenged OpenAI to improve the speed and efficacy of its models.
“What’s important here is that this is a business that really understands its market and the needs of the people using its products. They’re also challenging us, as their AI platform provider, to deliver better latency and higher-quality outputs,” she says.
Troth adds: “They’re willing to say when a model isn’t meeting their expectations and ask us what we’re going to do to improve it.”
Troth concludes by mentioning OpenAI’s commitment to founders. “We’re growing and investing heavily in our teams across our major markets at the moment. Our focus is very much on helping founders and startups succeed, and we genuinely mean that,” she says.
“Reach out to the team, find out what’s available and work with us on co-development opportunities.”
For more startup news, check out the other articles on the website, and subscribe to the magazine for free. Listen to The Cereal Entrepreneur podcast for more interviews with entrepreneurs and big-hitters in the startup ecosystem.
Source link
