I see it differently. The companies that will get real value from AI in the next six to eighteen months aren’t necessarily the ones moving fastest today. They’re the ones doing the unglamorous work first. Sorting their data. Tidying their SharePoint. Deciding who owns what, and putting proper governance and security around it.
On the face of it, that can feel backwards. Why spend time on a platform that’s been around for years, when the board wants to talk about AI, agents and automation? But that’s exactly why SharePoint should be on the C-suite agenda. Microsoft is positioning it as the foundation for what comes next: the trusted knowledge layer that Copilot, agents, low-code tools and future AI experiences rely on. If that foundation is messy, everything built on top of it inherits the mess. That work might be seen as boring, but it’s the single biggest predictor of who gets a return and who ends up firefighting.
Why old foundations decide new returns
AI is only ever as good as the information it reads. Give it a clean, well-governed view of your business and it becomes genuinely useful. Point it at two decades of half-finished documents, duplicated folders and content nobody has owned since 2019, and it answers with total confidence using all the wrong material.
We’ve already seen versions of this happening – emails sent to clients referencing a CEO who left the business two years earlier, for example. They hadn’t got it wrong. The AI had. Now picture that inside an organisation of a hundred thousand people, with two decades of legacy content and no one quite sure which version is the truth. The opportunity for error is huge. And, unless you have a clear and tidy view of your data estate, much of it is probably in your blind spots.
Agents turn bad data into business risk
Agents raise the stakes again, because an agent not only reads your data – it acts on it.
The most useful way I’ve found to think about an agent is to treat it as an employee. You would never give a new starter a master key on day one, access to every system in the building and no one checking what they did with it. You give them the access their role needs, you keep an audit trail, and you can see what they’ve been doing. An agent deserves exactly the same discipline. It has a job, it should only reach the data that job requires, and you should be able to see where it’s been.
This is where we draw the line between AI experimentation and AI adoption. Experiments can run on enthusiasm. Adoption needs ownership, visibility and control.
And in Microsoft 365, that brings the conversation straight back to SharePoint. If SharePoint is the knowledge layer AI depends on, then permissions, lifecycle management and clear ownership are what make that layer safe to use. And that’s where governance comes in.
Governance is what makes speed safe
Governance has an image problem. It gets talked about as the brake, the thing that slows everything down and wraps good ideas in red tape. I think that badly undersells it. Good governance works like good brakes on a fast car. The reason you can take the corner at speed is that you trust them completely. Strip the brakes out and you don’t go faster, you just crash sooner.
In SharePoint terms, it’s slightly less glamorous: permissions, ownership, naming conventions, retention, review cycles and clear rules for what belongs where. None of that sounds transformational on its own. But together, it decides whether AI can find the right knowledge, respect the right boundaries and act with enough context.
The organisations Silicon Reef are working with that invested in their foundations aren’t moving more slowly than their rivals. They’re moving faster, because their people can build without anyone having to ask whether the whole thing is about to fall over.
We saw this play out at Warner Bros. Discovery. When they put Microsoft Power Platform into the hands of tens of thousands of employees, the governance went in before the building started. The environment strategy, the data loss prevention, the security roles, the central monitoring. Because that groundwork was done, every new app across the estate landed compliant, and the business could let hundreds of people build at real speed without losing control of any of it. The foundation was the thing that made the speed safe.
The cost of skipping the boring work
A strong foundation is the part the ‘rush-to-build’ crowd misses. Skipping governance feels faster for about a year. Then the bill arrives. Data ends up somewhere it shouldn’t. An agent or app does something nobody authorised. Trust evaporates, and the team that was supposed to be shipping value spends the next two quarters cleaning up instead. The companies that did the ground work are compounding a lead while everyone else is mopping the floor.
There’s also a more direct cost point starting to emerge. The introduction of more advanced AI tools like Microsoft’s Copilot Cowork is a useful signal of where this is going: agentic AI that can complete longer, multi-step tasks, with usage-based billing through Copilot Credits. That makes information quality a commercial issue, not just a governance one. If AI has to trawl through redundant, outdated and trivial (ROT) content to find something useful, you’re spending credits on noise. Cleaning up ROT means AI reads less, costs less and has a better chance of returning something valuable.
None of this is an argument against moving quickly. We are firmly in favour of building, and at Silicon Reef we help clients do it every day. It’s an argument about the order you do things in. Foundation first, then speed. Centre of enablement first, then a thousand apps and agents. The investment you make in governance now isn’t a tax on innovation. It’s the thing that lets innovation pay off later instead of blowing up.
If you’re a leader deciding where to put your AI budget this year, the unglamorous line item is the one to protect. It won’t feel like the exciting choice. In eighteen months, when you’re looking at returns rather than incident reports, it’ll turn out to have been the one that mattered.
Find the original article posted here.
