Buy Microsoft stock as they add AI to every one of their products
I started consulting on AI innovation for an enterprise manufacturing company - a day later, I went out and bought a bunch of Microsoft stock.
Why? My client always looked for the Microsoft solution first. And I could see an endless march of Microsoft press releases, weaving GPT-4 into into their entire product portfolio. I'm betting they are representative of the way many large company IT departments think.
Secondly, Microsoft’s bet on OpenAI will continue to pay off - it’s easy to forget that OpenAI had an internal version of GPT-4 working 2 years ago. So even if the pack has caught up with GPT-4-based models, that’ll change when the 5th shoe drops.
Transformers will change the face of knowledge work, and Microsoft can see it. They can see that we’ll readily use Copilot for Excel to analyse our data, Copilot for Dynamics to observe and improve customer interactions, Copilot for Power Automate to do our repetitive pointing and clicking, Copilot for Sharepoint to open up our institutional memory… And each customer company, team, and employee will build endless custom Copilots with a drag-and-drop low-code flowchart interface, thankfully assisted by Copilot for Copilot. It’s a grand vision, and I’m bullish about the value this will create in the long-term.
But in the here-and-now, I sat down to try and build stuff with Microsoft’s AI products. I started simply with an Excel file of HR survey data, nestled in OneDrive or SharePoint or some other double-barrelled Microsoft storage hell, and we set out to find sentiments and trends and clusters from some scattered, hasty, wordy, multilingual employee free-text responses. You know the drill.
But soon, I could feel bile building in my throat, bile with a flavour … that I can only describe as tasting like ‘Microsoft Windows’. I struggled to even attach my Excel file to the Bing For Work chatbot, struggled to make it read the Excel file format, struggled to point Copilot for Excel to the data right in front of it, to deal with empty cells, to generate and run code. Given that Excel is a flagship product, Copilot for Excel is really flimsy (as of summer 2024). More often than not, it would simply shrug and say “I can't help you with that”.
I had been spoilt by ChatGPT’s relentless improvements, which would have swallowed this problem whole in 30 seconds. But our information security team were (rightly IMHO) wary of handing business-sensitive data to OxymoronAI.
We also tried building a custom search. After a couple of person-weeks, with help from a couple of Azure consultants, we almost had a prototype working. But it was bewildering, it didn’t work that well, and we knew we still had to face the remaining 80% of the effort to productionise it. Azure’s power and complicatedness makes the easy things just about possible - while the hard things are also just about possible.
So what did I conclude from this? Microsoft's fighting on too many fronts. They're trying to integrate AI into their entire product portfolio at the same time, while building a new data centre every three days, and who knows what else behind the scenes. They’ve never had great taste for interfaces. Microsoft products often become a churning quicksand, accreting new features and occluding key functionality with each new version. The frenetic pace of AI progress has only exacerbated that.
And yet I have kept my newly-acquired Microsoft stock, because they will get it right enough. Companies will prefer to painstakingly build a Microsoft-based solution that no one can get fired for. The alternative would be to deal with the integration and information security hassles of working with third-party products. Microsoft makes $100b/year from Office and Azure, and AI services will double that for them in the next few years.