Thoughtworks: The inside line on AI software development from the company that created Agile

Securing an interview with an executive from the company that quite literally invented Agile is a bit of a thrill, far more so when technical issues meant your humble editor showed up late and unprepared. Ever the gentleman, Thoughtworks' APAC managing director Steven Yurisich is completely unfazed and entirely accommodating. Especially when learning that authentication, or the lack thereof, was the root of the problem. "Not so much fun at all," he laughs.

While there were plans for a 'topic du jour' revolving around recent Thoughtworks research, our discussion turned instead to matters more organic. Perhaps out of necessity, given restricted access to essential assets like, oh you know, email.

A reminder of what and who Thoughtworks is, is instructive. As Yurisich relates, it is a software engineering, design, and strategic services provider with around 10,000 people in 47 countries and "The world falls into three categories. Those who know Thoughtworks extremely well, those who have heard of us but don't quite know what we do but reckon it has something to do with thought leadership like Gartner or IDC, and those who really don't know us at all."

The first group, he says, tend to know the company because it keeps, ahem, good company: "Our chief scientist Martin Fowler co-authored the Agile Manifesto, of course," he explains, "And was responsible for inventing a number of other things like CI/CD, and also microservices with James Lewis."

"We have almost 9,000 engineers and developers and the likes of Forrester rate us at the top right on the custom software grid, so pound for pound we're probably the best custom software developers you'll find."

That backgrounder begged what was for TechDay a rather pressing question. Who better to ask about custom software and the advent of AI development, particularly in the context of what Rimini Street's CFO Michael Perica recently described to TechDay as 'the SaaS-pocalypse'.

For those who missed the Perica interview, he elaborated that this 'pocalypse' is pressuring SaaS providers because AI offers the (purported?) ability for businesses to rapidly create custom software with a much-lowered barrier to entry, while achieving systems more closely aligned with how they actually do business.

Yurisich comes back with an answer that harkens to the choices developers face when creating software. Sure, there's Agile and it has distinguished itself as an effective methodology for some software development, with its iterative cycles.

Some, but not all, because waterfall and other methodologies still have their place.

"Yeah, look, it's a great question," he says, "And the answer is that it isn't black and white. We are hearing this from a number of global clients, that the cost of COTS [custom off the shelf] versus SaaS is starting to intersect and potentially cross over. So it is definitely happening, and we're definitely some of the more advanced or forward-thinking companies globally saying, we're going to invest more in custom software."

Clearly having been around for long enough to know better than declaring the death of anything in the tech space (as early as the late 1990s and early 2000s, technologies including printers, tape storage, and even 'spinning rust' mechanical hard drives had premature eulogies written for them), Yurisich also points to the inertia of systems of record.

"I'll take, say, SAP as an example. Look, at the end of the day, it's probably still going to be the system of record for ERP and supply chain and finance, for the foreseeable future. Most of the Fortune 500 or Forbes 2000 run on SAP, and they're not going to be ripping that out, because that's a massive business risk."

But at the same time, the rise of agentic AI won't be ignored by these businesses either.  "You need to look at it layering in the agentic architecture and other custom pieces around that; certainly we're doing that with a number of clients where we're building capability around those systems of record…you won't want to be interfering with finance, but there are a lot of other use cases in which it may make sense just to build your own capability."

Just how that capability is built is another topic on which we felt Yurisich was ideally placed to comment, given that Thoughtworks is by no means a stranger to innovation in software development. What, we wanted to know, are his thoughts on vibecoding ?(Our view is that it is both hilariously ad hoc, and fascinating at the same time)

Yurisich grins broadly. "Llook, I think it's actually really interesting. And in the right context we use it occasionally, to code up a prototype or a demo," he says.

The kicker and the big caveat is that 'fun' and 'enterprise' don't make bedfellows. "You're not going to want to build an enterprise platform on it, and no, you're not going to open your enterprise up to some of the newer tools, no matter how amazing they are. Instead, it's more about the art of the possible. For example, we've used it to get a product concept from three weeks to three days, as a Minimum Viable Product. So, very much for prototyping and upfront stages. And very much not for production."

Horses, as they say, for courses.

And, as it happens, Thoughtworks has been putting its innovation cred to work, helping developers and the companies they serve take advantage of the emerging AI technologies and techniques - which may indeed include vibe coding and of course automated agents. 

That's where its AI/works platform comes in, Yurisich confirms, explaining that the brand new agentic development platform is designed to accelerate software delivery and modernise legacy systems using AI agents. Launched in early 2026, it integrates AI with proven practices (unsurprisingly featuring Agile and continuous delivery), helping automate code generation, testing, and documentation with integrated governance, security, and cost control. 

These are tumultuous times and Yurisich has seen and been part of plenty of disruption, so we asked him for words of advice for those grappling with the inevitability of legacy software. "Understand your total architecture. That's the first and most crucial step in understanding where to modernise and build custom apps. Have strong, strategic business cases that drive competitive advantage. Apply the lens of financial or EBITDA growth. And always apply good software development practices, the software development lifecycle holds true as do Agile and other methodologies. With AI, those things don't go away and done well, this can be about much more than faster horses."

So, perhaps better courses, then.