Ask any CTO whether their team can build an AI tool in-house and the answer will almost always be yes. Open-source models and lower development costs mean most companies can get a working prototype running in weeks, and many have. The prototype works well enough to impress leadership in a demo and generate the momentum to become embedded in how the business operates. And that is precisely where the story gets interesting, because the question has always been what it costs to maintain and operate the system once it is running.
Sean Hopwood, Founder and CEO of Day Translations, has experience of both building and buying language technology. His company works with more than 10,000 linguists across offices on three continents and has developed its own large language model for translation. Clients had asked for a multilingual model that could work with or without internet access. In a separate example, when a client canceled an interpretation contract on a Friday to pursue an AI-powered alternative, Sean’s development team spent the weekend connecting and testing an AI service they were already working on. By Monday the service was ready, and the client returned after responding within three hours. “It’s the perfect example of adaptation,” Sean said on the In Other Words podcast. “Make an adjustment, find out what the client wants and give them what they want.”
Sean’s clarity comes from two decades of entrepreneurial decision-making. He founded Day Translations without a computer and without outside investment, and grew it into a company operating across three continents. Entrepreneurs learn early that every hour spent maintaining something generic is an hour unavailable for the capability that differentiates the business. That instinct for knowing where effort should go is the judgment organizations need when they evaluate their AI investments, and leaders who want better build-vs-buy decisions should be hiring and promoting people whose instinct is to ask “should we carry this” before “can we build this.”
For Sean, building created a way to meet specific client requirements. He is equally clear about where he buys, subscribing to platforms for the layers that are not his product. Companies investing in internal AI need to make that second decision with the same discipline, or risk taking on obligations they did not budget for. Elaine Barsoom, who built Nike’s first AI Center of Excellence, described what the right buying decision looks like on an earlier episode of In Other Words.
“Great ecosystems just disappear in use,” she said. The best platforms become invisible, and internal builds tend to do the opposite.
The obligation nobody budgeted for
Every internal AI build creates two things simultaneously, and most organizations are only counting one of them. Stephen Lumenta, Chief Product and Technology Officer at Phrase, described this clearly in MultiLingual Magazine earlier this year. “Every internal build creates an asset and a corresponding liability,” Stephen argued. The asset is capability, and the liability is the obligation to operate that capability safely and consistently as it becomes embedded across languages and regulatory jurisdictions. Stephen warned against treating these projects as one-time capital investments when they are perpetual obligations that accumulate technical debt. “AI lowers the barrier to building,” he observed, “and significantly raises the bar for operating at scale.”
The obligations are specific and they compound. The engineering team inherits accountability for uptime across time zones, auditability as AI enters regulated content, version control across models that change constantly, and incident response when something breaks in a language nobody on the team speaks. None of this appears in the prototype. It tends to arrive months after launch, when a regulator asks how a customer-facing communication was produced in a local language, or when the engineer who built the system moves on and takes all the undocumented architectural decisions with them.
Your best engineers are reinventing plumbing
Scott Kinka, Chief Strategy Officer of Bridgepointe Technologies and a 30-year technology veteran, identifies the core miscalculation on In Other Words.
“The other challenge that businesses have is believing that building is a one-time event.”
This is where the failure data starts to make sense. A January 2026 Gartner analysis found that at least half of all generative AI projects had been abandoned after proof of concept by the end of 2025. The reasons include escalating costs, poor data quality, inadequate risk controls and unclear business value
Laszlo K. Varga, Chief Insights Officer at Nimdzi, asked the harder follow-up in his research on the build-or-buy trap. “What are we NOT building because your best engineering talent is busy reinventing existing plumbing?” Drawing on research with more than 100 enterprise localization leaders, the article argues that getting to 80% readiness with AI coding tools is easy enough, but the remaining work involves the coordination and governance infrastructure that platform vendors have refined over decades.
“Engineering teams are eager and capable to build. Let them. But let them build novel solutions to unsolved problems.”
The technology company scaling on a platform
For technology companies, established platforms can support growth without adding the responsibility of developing and maintaining the underlying language infrastructure.
Pega, the enterprise software company with over 5,000 employees worldwide, is a clear example. The company needed to deliver multilingual content faster without adding headcount. By adopting the Phrase Platform alongside changes to its content processes, Pega reduced translation turnaround times by 75% and more than doubled translation volume year over year, processing more than 14 million words through Phrase Portal in a single year. Marketing content that once lagged English-language releases by weeks now launches in as little as one week.
Even the builder buys
Stephen reduced the framework to a single line. “Build only the differentiating layer. Buy or partner for the layers where liability weight grows faster than competitive advantage.”
Sean illustrates how the framework can work in both directions. He built where the advantage was his to own, and he is equally clear-eyed about where buying made him stronger. “I was able to use other services, like Phrase or all the language tech tools,” he said. “I was just able to subscribe to those services and grow the company.” Asked what he would do differently if starting over, Sean went further. “If we had adopted these tools earlier, it would have made a big difference.”
Most leadership teams have at least one AI project that started as a promising prototype and has since become embedded in how the business runs, without anyone revisiting whether it still belongs there. The question has always been whether carrying it is the best use of your company’s talent and capital. If the honest answer is no, the framework is clear. Buy the platform and build what only you can build.
Watch the full conversation
Sean Hopwood founded Day Translations in 2007 without owning a computer and grew the business without outside investment. Nearly two decades later, the company works with more than 10,000 linguists and has offices on three continents.
Sean shares his experience of staying independent and deciding which AI capabilities to build and which technology to buy.
