The “Build vs. Buy” Trap Is Back, and This Time AI Is the Salesman

(SeaPRwire) –   By: Oliver Hawthorne

There is a particular kind of confidence spreading through enterprise IT right now, and it smells familiar. The pitch goes like this: why keep paying SAP, Workday, or HubSpot when Claude Code can generate a custom equivalent over a long weekend? CIOs are hearing that the subscription era is over, that bespoke software has been democratized by LLMs, and that millions in licensing fees are ripe for reclamation. I have watched this movie before. It was called “we’ll build it ourselves in Java” in 2004, and “our internal team can replicate that SaaS tool” in 2014. The ending rarely changed. What makes the current wave different is not the economics of building, which remain stubbornly brutal, but the seductive speed of the demo. A prototype that once took a quarter now takes an afternoon, and that afternoon is exactly where the danger lives. The person who wrote the source piece knows this from both sides. He spent years in the CIO chair, once burned four months at a beverage company building a custom analytics dashboard before killing the project and buying instead, and now sits as a leader at a global workflow software company. That dual vantage point matters, because the build-versus-buy debate has never been about engineering capability. It has always been about opportunity cost, maintenance gravity, and the quiet question of who answers the phone when the architect quits.

The evidence assembled in the original piece deserves to be taken seriously, because it avoids the usual vendor hysteria and instead piles up operational arithmetic. Start with cost. Companies building their own LLM-based software solutions are typically spending five to ten times more than they would on an established workflow automation platform, once development, running, and maintenance are honestly accounted for. The subscription line item is visible and annoying; the iceberg beneath it is not. At one large financial services company, a CISO proposed rewriting a core system with an LLM, a process that could easily consume eighteen to twenty-four months, all to save the equivalent of 0.5% of the company’s annual operating budget. Then there is the security ledger, which AI has made worse, not better. Building something that works on day one is the easy part. Defending it on day 1,000 is the actual job. Roughly half of organizations have already seen AI agents exceed their permissions, and the car rental management platform PocketOS had its production database wiped by a coding agent in nine seconds. Governance, auditability, regulatory compliance: these are not features you prompt into existence. Finally, the human fragility problem. Average CIO tenure runs about 4.5 years. A custom platform takes two or more years to build and longer to mature. One CIO inherited a custom system from a departed predecessor and found no documentation, no support team, no vendor to call. Just executives staring at infrastructure they could not explain and could not afford to shut down. That is not a technology failure. It is an organizational design failure dressed up as innovation.

Here is where the commercial logic resolves, and it is more nuanced than either camp admits. The piece is written by someone with a product to sell, and readers should hold that fact plainly in view. But the framework survives the conflict of interest because it draws the boundary in the right place. Build when the software expresses your secret sauce. The shipping and logistics CDIO who greenlights only projects that harness decades of proprietary routing data is making the correct bet, because no vendor can sell him that institutional knowledge back. Buy when the workflow is undifferentiated plumbing: onboarding, time-off requests, billing, HR. These processes are unsexy, heavily regulated, and identical across industries, which is precisely why a vendor amortizing security and compliance costs across thousands of customers will always beat your internal team on total cost. The endgame, then, is not the death of enterprise software but a sharper stratification of it. AI coding tools will compress the middle market, killing off mediocre internal projects that never should have existed, while simultaneously raising the bar for what justifies a custom build. The vendors most exposed are not the workflow giants but the thin-wrapper SaaS companies whose entire product is a CRUD app with a login screen. For CIOs, the gut-check questions in this piece are really one question wearing five costumes: are you building a moat, or are you building a liability with a nice demo? The companies that answer honestly will find AI makes buying smarter, not obsolete.

Author bio: Oliver Hawthorne is a Principal Correspondent permanently stationed at an international technology review, covering enterprise software economics, AI infrastructure adoption, and the shifting fault lines between vendors and in-house engineering teams.