Two recent articles on the Martin Fowler site examine, from the angles of data foundations and engineering capability respectively, the practical obstacles enterprises face in adopting agentic AI. In an Aug 27 article, Pramod Sadalage and Prem Chandrasekaran note that many organizations are excited about AI's promise to streamline processes, save money and juice margins, but AI's capabilities are founded on the data it accesses — and for many organizations that foundation is "little more than sand." An Aug 19 article focuses on a widening perception gap: between non-technical people building apps with AI over a weekend and enterprise-grade software delivery, there is a difference in understanding of what software engineering actually is.
Data foundations: the "sand" beneath AI's capabilities
According to the Martin Fowler site article, AI's capabilities rest on the data it accesses. The Sadalage and Chandrasekaran piece sets out to discuss how to build a reliable, accurate and trustworthy data foundation. The abstract states plainly that many organizations' data foundations are "little more than sand," meaning data trustworthiness must be addressed before agentic AI can be deployed.
The article does not lay out a detailed methodology in its abstract, but it clearly sets "accurate and trustworthy" as the goal for building a data foundation. For enterprises hoping to use AI for process automation, the judgment points to a basic fact: what AI can do depends on the quality of the data it can access.
Perception gap: a weekend app is not enterprise software
The author of the other article observed an interesting gap opening up over the last six months. It isn't a gap in technology; it's a gap in what different people think software engineering actually is. The article describes a common scene: a non-techie, maybe an executive, excitedly describes something they built over the weekend — sometimes a chatbot, sometimes an internal workflow, sometimes a surprisingly polished app that solves a real business problem.
The question that usually follows is: "If AI can do this now, why can't our engineering team deliver ten times faster?" The author considers this a completely reasonable question — everyone has seen the demos. The author's first reaction is "you don't know what it takes to build enterprise software," but then realizes that explanation can come across as condescending.
"We've spent years emphasizing how to write good software, to the point where everyone assumes writing software is the same as software engineering."
The article stresses that a weekend-built app is real software; it very likely solved a real problem or validated an idea, and sometimes it is genuinely impressive. The author does not want to belittle that, but seeks to clarify the difference between "writing software" and "software engineering." The article's title sums up the division of labor it advocates: "Citizens Build, Agents Execute, Experts Govern."
What to watch next
Together, the two articles point to two preconditions for enterprise AI adoption: a trustworthy data foundation, and a clear-eyed understanding of the engineering discipline and expert governance that enterprise-grade delivery requires. The clues suggest that as agentic AI moves from demos to production environments, data quality and engineering governance will become the key differentiators between "weekend apps" and "enterprise-grade systems." Full details of the relevant methodologies can be found in the two original articles on the Martin Fowler site.