AI, tools and transformation

Big companies are stuffed with software: SAP, Workday, hundreds of vertical apps, plus a sprawl of spreadsheets and scripts nobody fully tracks. Yet the work is still boring and repetitive. So some people get excited: AI can sweep it all away.

Not so fast, says the author. That idea misunderstands where software comes from and how companies change.

Lawyers don't build their own tools

Silicon Valley is full of tool-builders, but most people aren't. A top matrimonial lawyer thinks about cases, not legal software. Excel tempts you with templates and wizards—and each of those templates eventually became a company—but that's a nudge, not the answer.

Hence the 'forward-deployed engineer': a builder who walks into a law firm and sees opportunities lying on the table that the lawyer can't. Which proves the hard part isn't writing code—it's knowing what to build.

Most tools emerged by trial and error

The truly useful software of the past few decades rarely came from a flash of insight. Many things we use daily first prompted 'Why would I want that?' The problem hides inside something else, and the fix often means redefining or unbundling it—something AI can't help with.

And even when you have a great idea, you need the whole company to adopt it. Changing accounts payable touches five departments, three systems, and four regulatory regimes. That requires a purchase, an 18-month sales cycle.

Software is a spectrum, AI just moves the threshold

At one end sit institutionalized behemoths like SAP; at the other, improvisational spaces like Excel, email, and shared folders. When a makeshift solution becomes routine and mission-critical, the company paves the desire path and turns it into formal software. Hundreds of apps grow this way.

AI rolls across all this: chatbots become a new freeform space next to Excel, taking over some tasks and losing others. A small firm might stick with Google Sheets longer because AI scales it up, or ask whether Claude should build something—then discover a startup selling just that.

So don't just hand everyone ChatGPT. The data shows it: after giving everyone Copilot, most people use it occasionally or not at all. It's like giving everyone a PC and Lotus 123 in 1983—the tool shipped, but transforming a company doesn't come from everyone having a tool.

Back to the old CIO conversation: pilots, lighthouses, heroes, quick wins. The CEO scratches their head: 'A hundred workflows and we've done five pilots? That doesn't scale.'

Every new technology forces three questions: how do we buy, build, deploy it? What does it mean for our operations? Does it threaten our business model? None are answered by giving everyone Claude for X. You'll still call Accenture, the Big Four, or McKinsey—ironic, given AI threatens their own business.

In short: AI didn't solve the problem of 'what needs doing and who'll use it.' It just swapped the pen for a smoother one. The hard question remains exactly where it was.