This is one of many tools I’ve seen pitched as a Palantir rival, so I installed it. It is quick to start and suggests a sensible map of your data. What it lacks is what makes an ontology run a business. There are no permissions, no audit trail and no way to act on what you find. Add a second user and it breaks. The fairest way to read this tool is as a personal workspace. It gives one founder a single view of their own data on a laptop, so they stop jumping between apps. That’s useful, and it is a different job. An enterprise ontology sits across the whole organisation and carries rules on who can see and change what. It also lets people act on the data, with every action written back and recorded. I read the wave of these tools as good news, now that the idea of an ontology has gone mainstream. The risk is that a board sees a free download and assumes the whole job is now cheap. Drawing the map is the easy part. Agreeing who can see and change what, across thousands of people, is where the cost and the value sit.

After an AI agent misbehaves, the government wants answers to four questions. What was it told to do? What did it access? Why did it act as it did? Could the same failure happen elsewhere? The first two can be logged today. The third is only partly explainable, and the fourth needs an investigation after the event. I'd be asking your teams which of the four they could answer this afternoon. Every boardroom conversation about AI should now cover governance and security, alongside growth. The speech makes cyber security a board-level responsibility, and AI agents widen what that covers. A board should know which agents act on the company's behalf, what each is allowed to touch, and who answers when one goes wrong. If those answers sit only with IT, the board carries a risk it can't see. Most of the organisations we talk with trade in several countries, and each of those countries is writing its own AI rules. The UK has this bill, the EU has its AI Act and China has just set out a new AI safety framework. I'd keep one page that maps each market you operate in to the rules and regulators that apply. It turns a compliance scramble into a planning question.

As AI accelerates software delivery, implementation speed risks outpacing architectural understanding and governance. The result could be growing complexity, inconsistent design decisions and technical debt that becomes increasingly difficult to manage. Traditional architecture practices, built around design reviews, standards and human oversight, may struggle to keep pace with the volume and speed of AI-generated software. Simply applying existing governance processes more frequently is unlikely to be sufficient. Architectural principles, constraints and quality attributes need to become more explicit, measurable and embedded within development workflows. This means moving towards architecture that can be continuously validated through automated controls, testing and observability, while retaining human accountability for significant trade-offs. AI should accelerate delivery within an intentional architecture, rather than allowing architecture to emerge as a by-product of generated code. Achieving this requires a shift from architecture as periodic oversight to architecture as a set of continuously enforced design principles and constraints.