Something remarkable is happening in plain sight, and most people—including the incumbents—aren’t paying attention. While everyone debates whether AI will replace jobs or transform industries, no one is talking about how AI is already systematically displacing the productivity tools we’ve used for decades.

This blog is an experiment in using AI tools. And while normally I write everything you read here manually, today’s entry is 95% fabricated by machines. The following is based on an actual conversation between myself and Bob Matsuoka about the stealth transformation reshaping how we work—and why the tech giants are missing this opportunity. Bob publishes his HyperDev blog, a technical publication focused on practical applications of agentic coding technologies. Drawing on his experience as former Head of Product Engineering at TripAdvisor and CTO of Citymaps, he provides hands-on reviews and strategic insights for developers navigating AI-powered development tools. He currently serves as fractional CTO for multiple companies while actively building with the technologies he covers. We last saw Bob’s work two decades ago when he penned a piece of why the relatively new iPhone was a big deal — for its Clock app.

Bob and I spoke earlier in the week, he recorded our meeting with Granola, a note-taking app. He then copied the transcript into a Claude thread that he uses for this purpose. While that was being developed, he made another Claude project to extract my previous work from my blog posts. We both manually made edits to a GDoc to clean up our respective dialog. We spent about 30 minutes or so on our own with all this work, not counting the hour that we spoke to each other.

Bob: You know, that VisiCalc comparison keeps rattling around in my head. But here’s the thing—what I’m seeing isn’t just a better spreadsheet. Six months ago, if I needed something done, I’d think “Gmail” or “Excel.” Now? Claude is my first stop for everything.

David: So when you say “go first”—you’re talking about bypassing the usual suspects entirely? Gmail, Calendar, the whole productivity stack that we’ve all been living in for the past decade-plus? That’s a pretty fundamental shift in user behavior.

Bob: This local MCP bridge I’ve got running—it connects to 55 different tools. Apple contacts, Google Calendar, Gmail. I can read emails, write responses, schedule meetings, all from what amounts to an AI dashboard. The thing just works.

David: Interesting. So we’re talking about the AI interface becoming your primary operating environment. I remember when browsers started feeling like the “real” desktop for most people—that was maybe 20 years ago? This sounds like the next evolution of that shift, but faster and more comprehensive.

Bob: Pretty much. And here’s what caught me off guard—Anthropic and OpenAI didn’t make any big announcements about this. They just shipped it. Last week, Anthropic rolled out unlimited Retrieval Augmented Generation training. You can dump a thousand documents into a project folder now and it trains on all of them. No press release. No marketing campaign. They’re moving so fast they don’t bother promoting major features.

David: The velocity here is what really gets my attention. I covered Google’s office suite launch back in the day—that was a multi-year enterprise sales campaign with migration consultants, pilot programs, the whole nine yards. These AI companies? They’re just shipping features like its continuous deployment. No marketing blitz, no sales engineering team, no six-month enterprise evaluation cycles. It’s almost like they don’t even realize they’re disrupting a multi-billion dollar market. Every day their AI tools are gaining functions.

Bob: Exactly that. But the underlying architecture is different this time. Let me explain—last month I had this travel client who needed a pricing model. Old me would have built some monster Excel sheet with formulas and maybe Visual Basic scripts. Instead, I uploaded the raw CSV data to a custom GPT, had it write Python code using NumPy, and boom. I had an interactive model where the client could ask “What’s my margin on a five-day trip to Miami during peak season?” No spreadsheet. Just answers.

David: Rather than handing clients a spreadsheet with 47 tabs and saying “good luck figuring this out,” you gave them something that actually responds to questions. That’s a fundamentally different interaction model from what we’ve been calling productivity software for the past couple decades.

Bob: And they never saw a spreadsheet. They just got answers. That’s the shift—from tools that require expertise to interfaces that provide insight.

The Pattern David Has Seen Before

David: I’ve covered enough of these platform transitions to recognize the pattern by now. IBM dominated when mainframes were everything—I remember those room-sized monsters that cost more than most houses. Microsoft crushed them with personal computers (which seemed crazy at the time, if you think about it). Google ate Microsoft’s lunch with web-based productivity tools. Each time I’m watching it happen, I think “this is unprecedented”—but the script keeps repeating itself with different players.

Bob: What’s the pattern you see repeating?

David: It’s the same story every cycle—the incumbent gets comfortable with their revenue streams and loses their edge. IBM couldn’t imagine computing without those room-sized mainframes (which, let’s be honest, still generate fantastic margins). Microsoft initially dismissed web-based productivity as “toys” when Google Docs launched. Now Google’s painted themselves into a corner with the advertising model—everything has to generate data for ad targeting, which fundamentally conflicts with what users actually want from productivity tools. They can’t see past their golden goose to imagine what work looks like when it’s not subsidized by surveillance capitalism.

Bob: That’s exactly what’s happening. But look, there’s another layer here. Google Office replaced desktop software with web apps—same basic concept, different delivery. But AI tools? They’re reasoning engines. They can process context, maintain state across tasks, generate stuff that didn’t exist before. That’s not an upgrade.That’s a different category.

David: Can you walk me through a specific example? I’m trying to understand how this “collaborative intelligence” differs from, say, really sophisticated autocomplete or code suggestion tools. Where’s the line between automation and genuine collaboration?

Bob:  I built this MCP Gateway project with pretty open-ended instructions—basically told it “solve problems however you think makes sense.” The thing that gets me is how it discovers its own solutions. Like, I asked it to set up a reminder for me, expecting it would use my Google Tasks tool. But because I said “reminder,” it went off and found the Mac OS Reminders app on its own, wrote AppleScript code to access it, and created the reminder there. I never told it about the Reminders app. It just discovered that path. That’s what I mean by collaborative intelligence—not that it’s actually intelligent, but it’s genuinely good at discovering new ways to solve problems. It doesn’t just follow patterns; it finds solutions I wouldn’t have thought of.

David: That’s helpful context, but help me connect this back to the productivity software question. How does understanding—excuse me, analyzing—code architecture translate to replacing the basic office tools that most knowledge workers live in every day? Excel, Google Docs, PowerPoint presentations?

Bob: The AI doesn’t just format code or fix syntax errors. I’ve configured it to analyze what I’m trying to build. It refactors entire architectures, suggests performance optimizations, debugs problems I didn’t know existed. That’s not a better text editor—that’s collaborative intelligence.

Why the Incumbents Are Missing It

David: Here’s what puzzles me about this whole situation—Google’s got more money than they know what to do with, they’ve practically got a monopoly on search, they hired half the world’s AI researchers, they invented the transformer architecture that makes all this possible. They certainly had plenty of AI-themed announcements at their IO conference recently, so they are building stuff. But they should own this space, not just be another player.

Bob: Two things. First, they’re trapped by their own success. Their business model depends on keeping people in their ecosystem, showing ads, and collecting data. But AI-first productivity doesn’t fit that model. When I ask Claude to analyze my calendar and suggest optimizations, the last thing I want is ads mixed into that analysis. Second, they’re thinking about AI as a feature to bolt onto existing tools. Google’s putting AI into Docs and Sheets. Microsoft’s adding Copilot to Office. That’s like adding internet features to desktop software in 1995. They’re missing that the entire interface paradigm is changing.

David: Yes, there’s also the organizational antibody problem that I’ve documented at pretty much every large tech company I’ve covered over the decades. You’ve got thousands of engineers working on incremental improvements to Gmail, Docs, Sheets—products that generate billions in revenue and support entire divisions. Walk into a meeting and suggest “let’s cannibalize all this for something completely different” and watch how fast you get politely but firmly shown the door. (Even if you’re absolutely right about the technology direction.) The incentive structures just don’t support that kind of self-disruption, especially when the current products are still growing and profitable.

And that brings us back to the deployment velocity problem. These AI companies are shipping major new capabilities weekly, sometimes multiple times per week. Google’s enterprise software division? They’re operating on quarterly release cycles if you’re lucky. That’s not just a technology gap—that’s a fundamental cultural and organizational chasm that’s very difficult to bridge in large organizations.

Bob: Plus, they’re not constrained by existing user expectations. When Anthropic ships a new capability, users adapt their workflows to take advantage of it. When Google changes something in Gmail, users complain that their familiar interface looks different.

The Enterprise Implications

David: Let’s shift focus to enterprise implications, because this is where things get really messy. I’ve been getting calls from CIOs and IT directors who are completely caught off guard by this bottom-up adoption pattern. Their employees are using these AI tools for mission-critical work—and reporting dramatic productivity gains—but IT has zero visibility into data governance, security policies, or compliance implications. It’s shadow IT on steroids, and frankly, most organizations aren’t equipped to handle it. This sounds familiar to those of us that started using PCs back in the 1980s.

Bob: That’s the other thing that’s different about this transition. Previous platform shifts were top-down. IT departments evaluated productivity suites, negotiated enterprise contracts, managed rollouts. But AI tools are getting adopted bottom-up by individual workers who see immediate gains. I know developers using Claude or Cursor for all their coding because it makes them probably 3x more productive. They’re not waiting for company approval. They’re just using the tools and dealing with governance later.

David: Shadow IT on steroids—and I say this as someone who’s been tracking unauthorized technology adoption in enterprises since people were sneaking Dropbox accounts onto corporate networks back in 2008 and spreadsheets in the 1990s. The scale and speed of this AI tool adoption is unlike anything I’ve documented before.

This puts IT departments into an impossible situation: they can try to block these tools and essentially tell their organization “we’d rather be less productive, thank you very much.” Or they can try to manage technologies they don’t understand well enough to create appropriate governance policies. I’ve watched CIOs try to navigate this, and honestly, it’s not pretty. The usual enterprise software playbook—pilot programs, vendor evaluations, compliance frameworks—doesn’t work when your employees are already using these tools daily and seeing immediate benefits.

Bob: Here’s what the AI companies figured out. They’re offering enterprise versions with security features, compliance controls, audit trails. But the value proposition isn’t “replace your existing tools.” It’s “make your people probably more effective at their jobs.” Hard to argue with that.

David: What’s your timeline prediction here? Because in all my years covering enterprise software transitions—and I’ve tracked everything from mainframe migrations to cloud adoption—they usually take forever. Even cloud-based email adoption still took most enterprises three to five years to complete. But this feels fundamentally different from anything I’ve documented before.

Bob: Based on what I’m seeing with my clients? Eighteen months, maybe less. When I can build sophisticated data analysis in three hours that used to take weeks, I’m not going back. And once you experience that kind of leverage, everything else feels like working with oven mitts on.

What Comes Next

David: What should people in the industry be watching for? What are the early warning signs that this shift is accelerating?

Bob: Multi-agent systems. Right now, most people use AI tools one at a time—Claude for writing, Cursor for coding, maybe Perplexity for research. (Though in fact, both Anthropic and OpenAI are actually using multi-agent systems in their desktop tools now.) But I’m starting to see orchestration tools that coordinate multiple AI agents on complex tasks. That’s when this really explodes, because you’re not just replacing individual productivity tools. You’re replacing entire workflows.

David: And what about the incumbent giants? Google, Microsoft—what’s their play here, assuming they wake up before it’s too late?

Bob: They need to pick a lane. They can survive losing the innovation high ground—IBM did, (old) Microsoft did. Both companies are more profitable now than when they dominated. But if they want to remain relevant for the next generation of workers, they need to build AI-native productivity platforms, not bolt AI features onto legacy products. The talent migration tells the story—I’m seeing engineers leave Google and Microsoft for AI companies, not because of money but because that’s where the interesting problems are.

That’s the real signal right there. When the smartest people in your industry start working elsewhere, you’re no longer the future. You’re just the past that happens to be profitable.

The Bottom Line

As our conversation wound down, I kept thinking about David’s observation that this transition feels both familiar and unprecedented. The pattern of disruption—incumbent complacency, technological shift, talent migration—follows a predictable script. But the speed and scope of change feels fundamentally different.

“We’re not just watching productivity software get replaced,” I found myself reflecting as we wrapped up. “We’re watching the entire concept of ‘software’ evolve into something more like collaborative intelligence.”

The implications extend far beyond which productivity suite you use. We’re looking at a fundamental shift in how humans and machines work together, happening largely under the radar while everyone debates the bigger questions about AI’s future.

The companies that recognize this shift early—and adapt their workflows, their hiring, their entire approach to knowledge work—will have a massive advantage. Those that don’t may find themselves wondering how they went from industry leaders to legacy providers in the span of a few quarters.

The arms race is real. It’s happening now. And the winners aren’t necessarily the companies with the biggest marketing budgets or the most enterprise sales reps. They’re the ones building the tools that knowledge workers reach for first every morning.