OpenAI’s Astra Could Change How Startups Are Built, Says OpenAI’s Tricia Troth

If there’s one thing everybody knows, it’s that time and money are scarce commodities in the startup world, closely followed by sleep, a social life and just a general sense of calm.

More often than not, small teams have to build products, test ideas, understand their markets and find customers all at once, and often without the resources available to larger competitors. But, what if all that work could be done by just a few people (or even one person), without compromising quality and efficacy? What if AI could help founders do more of that work before they have to expand their teams or commit significant funding?

OpenAI believes its new GPT-6 Astra model could help make that possible. In an exclusive interview with TechRound, Tricia Troth, Head of GTM Startups for EMEA at OpenAI, discussed how the model could change the way founders approach everything from early-stage research to product development, and why its implications could extend beyond simply writing code faster.

“I love the fact that AI is changing who gets to build a company and how quickly they can turn that idea into a viable business,” Troth says. “It gives some of those smaller founding teams real capacity to build more, be more ambitious [and] do more with their time.”

It’s an interesting proposition at a time when founders are under pressure to demonstrate progress, manage costs and get products into customers’ hands as quickly as possible. But the opportunity, according to Troth, is about changing what smaller teams can realistically achieve, not just giving them a more powerful model.

From Idea To Product, But a Whole Lot Faster

Astra brings together complex reasoning, software engineering and computer use to help teams tackle demanding, multi-step tasks. For startups, that could mean getting help with technically challenging development work while maintaining momentum across the wider product-building process.

“It’s really powerful as a combination for some of those smaller scaling teams when they’re trying to tackle difficult technical problems,” Troth says. “Astra can help them move faster, be much more ambitious with their product and help them scale more quickly, essentially.”

Essentially, it can give them access to a tool and skills that would normally only be achievable via a far larger, more experienced and way more expensive team. With Astra, the aim is to shorten the distance between having an idea (the initial lightbulb moment, if you will) and finding out whether it actually works. Rather than spending months and months developing a product before gathering meaningful feedback, founders could simply use AI to move towards a working prototype, test it with users and then make changes based on what they learn.

It’s plain to see how much of a gamechanger this could be for startups with limited funding (so just startups in general, I guess). The sooner a team can test its assumptions, the sooner it can make more informed decisions about where to invest its time and resources.

Choosing the Right AI Model Matters

However, Troth is not suggesting that founders should automatically use the most powerful model for every task. In fact, she argues that choosing the right tool for the job is an important part of building a sustainable business. “You need to be thinking about cost, about quality, about speed, about latency,” she says.

Troth points to Kodo, a Tel Aviv-based startup building an AI code-review tool, as an example. In its published evaluations, the company used ChatGPT 5.6 to review code quality rather than a frontier model such as Astra. According to Troth, the model generated a third as many tokens on pull requests and halved average response times compared to its previous model. Indeed, this example highlights a practical issue that needs to be considered by founders, which is that as tempting as it may be, more capability doesn’t necessarily mean a better fit for every workflow. A startup needs to weigh up what a task actually requires, how much it costs to run and how quickly the result needs to be delivered.

As AI models of all shapes and sizes, so to speak, become more and more embedded in operations and day-to-day life, it’s going to be increasingly important that AI-based solutions are implemented intentionally and thoughtfully.

Testing a Startup Idea Before You Hire a Team

AI could also help founders make better decisions before they really dig their heels in and begin building in earnest. Market research, customer discovery and product planning all take time, and early assumptions can be expensive if they turn out to be wrong. Somewhat necessary, many would argue, but risky. So, what if at least some of that risk could be removed from the equation?

Troth suggests starting with a real problem that a founder understands, then using Astra to challenge their thinking rather than simply asking it to confirm an existing idea. “I think they need to start with one real-world customer problem that they understand super well, give Astra the context, give it your assumptions. Give it your constraints, and then ask it to really challenge your thinking and help design probably a small useful experiment.”

The next step is to put something in front of a potential customer, ask a clear question and observe the response. Then, founders can use what they learn to decide what to do next, rather than relying entirely on their own assumptions about what the market wants.

This is an approach that could help early-stage teams assess potential opportunities before having to full-on commit to expensive and complicated new hires or investing heavily in product development.

However, having said this, AI-generated analysis isn’t a substitute for speaking to customers or finding out how people actually behave. As Troth explains, founders should also actively use these tools to identify gaps in their thinking, including potential blind spots, the partners that they may need going forward and the other technologies that are likely to be required to build and scale a the product in question. In this sense, the AI would be used as an additional tool rather than a complete replacement for previous methods or a fix-all solution.

AI-Native Startups Need More Than Just a Clever Model

As we know, there’s also a very clear and important difference between using AI to complete a few (or many) individual tasks as opposed to designing an entire business around AI from the outset. For founders, the lesson is to think carefully about the problem they are solving and how AI fits into the experience they want to create. The technology might help make a product more useful, but its presence alone does not guarantee that customers will want it.

Naturally, there are risks in getting carried away by what a model can do in a demonstration, and it’s incredibly easy to be tempted to fall into that trap. “I think what we can see working really well across the industry…[is to] stay absolutely obsessed with your customer and your user, the problem that they’re trying to solve for and the experience of using that product, and whether it genuinely has the ability to make somebody’s life better.”

Sometimes, asking those simple questions can cause a founder or a team that’s gotten carried away with the potential of  AI technology to take a moment to remember what the point is of what they’re doing and who the product is really meant to serve. These models’ capabilities are incredible, and there are so many fantastic, mind-blowing things that they can be used to do, but the truth is, not every single startup needs to take advantage of all these little things.

Troth’s message to startup founders here that they should focus on getting products into users’ hands early, seek feedback from them and keep improving the experience. Another quick pro tip, if you will, that she added was to avoid the temptation of assuming that a successful demonstration will automatically translate into a successful product. It’s a great start, but there’s still a lot more to be done from that point forth and on a consistent basis too. And as always, quality checks and human judgement should always be imposed, without question.

Could Astra Change Who Gets To Build a Startup?

The broader possibility is that more ambitious products could become achievable for smaller teams with the assistance of advanced tools like Astra. If AI can help founders research opportunities, tackle complex engineering problems and produce working prototypes more quickly and before having to spend loads of money, the resources required to get a business off the ground could begin to look very different.

That doesn’t mean we’ve suddenly eliminated all friction and barriers to entry involved in launching a successful startup. It certainly doesn’t mean that the founder journey becomes easy. Founders still need to understand their customers, develop a viable business model and make decisions about where to spend their limited resources. It simply means that now, AI can accelerate parts of that process and help remove a few of the growing pains that used to be a real thorn in many founders’ sides.

Indeed, Troth is undeniably optimistic about what GPT-6 Astra could make possible, particularly for people who might previously have struggled to turn an idea into a product. “People are able to spin up companies from idea to actually getting a product into the hands of a potential customer within a matter of weeks,” she says.

For startups, Astra’s significance seems to be centred on expanding what a small group of people can do. The challenge this introduces (or at least one of the challenges) is that this extra capacity will need to be used wisely in order for it to be truly beneficial.