From Markdown's Rise to Earth-Shattering Mass Layoffs: 8 AI Predictions for 2026
Ever since I started working with AI solutions, I've made it a tradition to dust off the crystal ball at year's end and map out what's coming. This year is no different.
This time, though, I decided to put more than just my own gray matter to the test. I invited Anthropic's latest Claude Opus model to join me in peering into the future. I set this up as a simulated dialogue where I first asked Claude to offer four AI predictions for 2026, then added four of my own. Together, we shaped it into a "fireside year-end chat" about the trends ahead.
Let's find out: can AI hold a meaningful conversation about what's to come? Can it glimpse the future the way humans do?
Enjoy.
Claude Code Moves Into the Office!
Jarmo: Let me start with my predictions. I predict that Claude Code will break into the traditional "office world" in 2026. Instead of Word, Notion, or Confluence, offices will start using Claude Code for document creation. Everything becomes code!
Claude: That's quite a bold prediction. Hold on—what exactly do you mean? Are you talking about everyone writing code with Claude instead of documents, or about the logic of document creation changing?
Jarmo: The second one, actually. Today you open Word, write text manually. Tomorrow you describe to an agent what the document should do, and the AI generates it. It's a fundamental shift in thinking – you don't create the document, you describe its purpose. An AI agent like Claude Code—or something similar—handles the creation.
Claude: Ah, okay. At first I would have said forget it—not everyone's going to start writing code in 2026. But if you're talking about the interface changing—moving from traditional writing environments toward agent-based document generation—that's a different story.
Still, isn't 2026 too soon? I'm skeptical that kind of breakthrough happens next year.
Jarmo: What makes you skeptical?
Claude: Habit. The power of habit is often severely underestimated. People have been using Word-like products for thirty years or more. Not just using them – they've learned to think with them. When I say "document," you see a Word page in your mind. That doesn't change in a year.
And another thing – who is the agent of change? Who pushes this? Companies don't switch tools for no reason, they need a reason. Right now Word works. It's bad, clunky, outdated – but it works. Why would anyone take the risk?
Jarmo: I think the pressure comes "bottom-up." Not from IT departments, but from people who've started using Claude themselves and come to work asking – why do I have to write this report manually when I do it at home with AI ten times faster?
Claude: That's an interesting angle. Actually, that's exactly how spreadsheets took off—Excel spread because accountants started using it on their own, not because management mandated it.
That said, I see the movement starting in 2026, but the real breakthrough coming in 2027-28. Unless...
Jarmo: Unless?
Claude: Unless Anthropic—or someone else—releases something that sits between Claude Code and Artifacts. Something genuinely designed for everyday users. Right now, Claude Code is clearly built for developers. Artifacts is more of a demo.
Someone needs to build something simple enough for an office administrator to use, yet powerful enough that developers don't feel constrained.
If that arrives in the first half of 2026—you might actually be right.
Jarmo: Let's write that down then—this is our first "marker." If a Claude-Code-for-the-office type product hits the market in H1 2026, my prediction looks more likely. If not, your skepticism is vindicated.
Claude: Deal!
Markdown Becomes the Primary Dual-Use Text Format
Jarmo: Here comes my second prediction: docx and PDF will start giving way to markdown. Maybe not entirely, but increasingly—especially in knowledge management.
Claude: Walk me through why you think markdown will come, see, and conquer.
Jarmo: Companies urgently need dual-use knowledge management. Documents must work for both humans and AI—meaning both need to be able to read and process the same files.
Docx is clunky for this—complex, riddled with hidden formatting. When you feed an AI a docx file, half the information gets lost or garbled. PDF is even worse—it's essentially a print-to-paper format.
Markdown is the only format that's simultaneously human-readable, machine-readable, and version-control friendly.
Claude: That's a solid point. It's one of those things that should happen—all the conditions are in place—but hasn't yet. Markdown has been around for twenty years. Developers use it constantly. But it's never crossed over to "regular" users.
AI might be what finally changes that. Because now there's real motivation. Not just "markdown is better"—but "markdown is the only format that actually works with our new tools."
Jarmo: Exactly. It's no longer a technical preference—it's a practical necessity.
Claude: One question—do you see this taking hold mainly in tech companies, or in traditional enterprises too? Banks, insurers, manufacturing?
Jarmo: It starts in tech, but spreads fast. Here's why: consultants and advisors will adopt it. And once McKinsey hands you a report in markdown because their AI processes it better, you'll start using markdown too.
Claude: Consultants as the vector—that's actually plausible. They've always been carriers of "best practices," even when those practices were dubious. This time it would actually be useful.
Jarmo: I'd bet that at least one major global corporation officially announces a shift to markdown-based internal documentation in 2026.
Claude: That's a very specific prediction. Let's log it. Second marker: a Fortune 500 company adopts markdown for internal documentation.
Agents: Off-the-Shelf Apps or Custom Service?
Claude: Now it's my turn! My biggest prediction—and something that genuinely excites me—is this: an AI agent "labor market" will emerge in 2026*.*
I'm not just talking about agents getting better—that's a given. I'm talking about an ecosystem taking shape. Companies will start buying and selling specialized agents the way they once bought software licenses.
Here's where it gets interesting: agent-to-agent interactions will emerge. Picture one company's sales agent negotiating with another company's procurement agent. Not people using AI, but AIs dealing with each other on behalf of people.
Jarmo: That's a compelling vision. But I see it somewhat differently.
Claude: How so?
Jarmo: You're essentially describing a SaaS model for agents. Buy or rent an agent, deploy it, use it. Like any SaaS product, just branded "Agent."
I think something else will emerge instead. Not a market for ready-made agents, but an agent development service. Companies won't want to buy a pre-packaged agent. They'll want someone who builds and tailors an agent specifically for them—someone who "trains" it, so to speak.
Claude: Why do you think that?
Jarmo: Because an agent isn't like regular software. With software, "one size fits most" works—Salesforce is Salesforce, it runs more or less the same everywhere.
But an agent needs to know you. It has to understand your business, your processes, your customers, your terminology, your quirks. You can't buy that off the shelf. You have to build it—or hire someone to build it for you.
Claude: Hmm. So you're talking about custom development versus packaged software. And you're saying that with agents, custom wins?
Jarmo: At least initially, yes. Standardization might come later—like what happened with ERP systems. Everyone started with custom solutions, then SAP and Oracle came along and standardized. But that took years.
Claude: That's a useful parallel. I have to admit, my original prediction was too... simplified? I pictured a market where agents are like apps—download, configure a bit, done. You're describing something messier, but probably more realistic.
Jarmo: "Messy" is the right word. AI projects are messy. Data is messy, integrations are messy, business logic is messy. Someone has to wrangle all that. And off-the-shelf software usually can't.
Claude: Fair enough—I mostly agree. But someone still has to build mass-market agents. Microsoft, Salesforce, ServiceNow—those players. They're not going to build custom solutions for every client. They have to ship packaged products.
Agents Here, There—and in Court?
Jarmo: On to my next prediction: Microsoft will cement its leadership in the enterprise market. Large organizations will wake up to M365 Copilot Studio's capabilities in 2026. Microsoft's edge is distribution. They don't need to win new customers—they just need to hold onto the ones they have.
Claude: Agreed. And that advantage is massively underrated. Most enterprises already run on Microsoft. When Microsoft says "now you've got an AI agent built in," it's the perfect pitch. No new contracts, no new security audits, no retraining.
Jarmo: Exactly. Microsoft may not be the best, but they're already inside—no door to knock on.
***
Claude: One more from me: the first legal battles over agent "behavior" will hit the courts in 2026.
Jarmo: What do you mean by "behavior"?
Claude: An agent executes a transaction on someone's behalf. The transaction causes harm. Who's liable? The person who deployed the agent? The vendor who sold it? The AI company whose model powers it?
Right now, nobody knows. The courts will sort it out. And I expect the first cases to land in 2026.
Labor: Boom, Bust, and Disillusionment
Jarmo: Let's talk about labor. This is the topic that hits people closest to home.
Here's my fourth prediction: 2026 will bring a hiring frenzy for AI talent. Not just engineers—everyone from executives to implementation specialists. Every company wants "their AI person."
Claude: Agreed. I'm already seeing it—not just in the data, but in conversations. People ask me: "How do I become an AI expert? What should I study? Am I falling behind?"
There's genuine anxiety out there. And companies are responding—they want to hire someone who "gets this AI stuff," even if they're not entirely sure what that means.
But this leads to trouble. By mid-2026, disappointment will set in.
Jarmo: What does that disappointment look like?
Claude: Right now, an "AI expert" is anyone who's used ChatGPT slightly more than their boss. That's a pretty low bar. And companies are hiring these people at hefty salaries with sky-high expectations.
A year later, it turns out many of them can't actually build anything. They can prompt. They can run demos. They can talk about AI in meetings. But when the question becomes "integrate this with our stack" or "why doesn't this work with our data"—they're stuck.
Jarmo: Yeah, that could definitely happen.
Claude: Eventually, a clearer split will emerge. On one side, prompters—people who know how to use AI. On the other, AI architects—people who know how to build AI systems. These are very different skill sets at very different pay grades.
The problem is that right now, prompters are drawing architect salaries because nobody can tell the difference yet.
Jarmo: How do you think this shakes out?
Claude: Painfully. (laughs) The first wave of "AI transformation" projects will flop. The first "Chief AI Officer" roles will get axed. A certain cynicism toward AI will creep in. But that's healthy—it clears out the noise and surfaces the people who actually know what they're doing.
Jarmo: And then what? Some players exit the stage?
Claude: Yes. And I think that's necessary. Right now there's too much noise. Too many people selling "AI transformation" who can't solve a single real problem.
The First Major "AI Unemployment" Crisis
Claude: But hiring frenzy and subsequent disappointment is still the softer side of the labor story. I want to talk about the harder side.
Jarmo: Job losses?
Claude: Yes. I think 2026 will bring the first major "AI unemployment" crisis. Not diffuse or quiet—but concentrated and visible, so the whole world takes notice.
The most likely targets: India's BPO sector or call centers in the Philippines. One of the big players—Infosys, TCS, Concentrix—will announce 50,000+ layoffs over the course of the year. Explicitly because of AI.
Jarmo: Why there?
Claude: A few reasons. First, this is work AI can actually do—call centers, data entry, basic customer service. Second, it's visible. When a US company lays off a thousand people, that's one news cycle. When an Indian firm lays off fifty thousand, it's a global event.
Third, these countries are exposed. In the Philippines alone, over a million people work in call centers. It's the backbone of their economy. If that buckles, everyone notices.
Jarmo: And then what? Does this force the world—and politicians—to react?
Claude: Exactly. The first serious "AI tax" proposals will surface. They won't pass immediately—regulation moves slowly. But the debate finally gets real. Not hypothetical—"what if AI takes jobs"—but concrete—"AI took jobs, now what?"
Jarmo: That does feel like a significant turning point. Until now, the AI unemployment debate has been speculative. "Someday, maybe." But actual mega-layoffs will probably force regulators to act. Maybe not in Europe, but definitely in more vulnerable economies.
Claude: And then things start moving. Not necessarily in a good direction—the political response could be clumsy, protectionist, anti-AI. But at least it's a response. Right now we have a technological revolution and a political vacuum. That vacuum gets filled in 2026.
Jarmo: Noted. Third marker: the first 50,000+ layoff explicitly attributed to AI.
AI in Geopolitics: Who Wins the Global Race?
Claude: I want to close with something that genuinely weighs on me. Not worry exactly—more like... heightened attention.
I predict the China-USA AI gap will widen in 2026. But not in the way most people expect.
Jarmo: How so?
Claude: The standard narrative is: USA vs. China, who builds the better model, who reaches AGI first, who "wins" the AI race. That's a technology-centric framing.
I see something different. The gap won't widen in technology—the US will probably stay ahead there. The gap will widen in deployment. China will roll out AI across the public sector and daily life far more aggressively than the West.
Jarmo: What do you mean by "more aggressively"?
Claude: I mean that by the end of 2026, the average Chinese urbanite will encounter AI in daily life more often than the average American. Not by choice—because it's simply everywhere. Government services, healthcare, education, transit, banking. AI isn't an option there; it's infrastructure.
In the West, we're still debating whether AI should be used. In China, the question is how to use it even more.
Jarmo: Fair point. I'd add that this isn't unique to China. Asia broadly seems more receptive to AI—Philippines, Malaysia, Singapore, Korea, Japan, Thailand, Vietnam. They're all moving faster than Europe and often faster than the US.
Claude: Why do you think that is?
Jarmo: A few factors converge. Fewer legacy systems—many of these countries leapfrogged entire technology cycles. Greater tech optimism—AI is seen as opportunity, not threat. A different relationship with privacy—it's a sensitive issue for us, but attitudes there are quite different.
Claude: That's a useful reframe. I tend to default to USA-vs.-China in binary terms. You're describing something more nuanced—Asia as a region, with different countries out ahead for different reasons.
Jarmo: That said, I still think the US will maintain its technological lead through 2026. The best models will still come from America. The top-tier AI infrastructure will still be American.
Summary
Jarmo: One thing that surprised me about this conversation—our focuses were different. I talked about tools, formats, hiring. You talked about labor crises, geopolitics, societal shifts.
Claude: Yes, I noticed that too. And I think the difference itself is meaningful.
Jarmo: How so?
Claude: You work with companies every day. You see how people actually use AI. So you view the future through the lens of tools. What will we work with? Which formats win? Who do we hire?
I gravitate toward systemic shifts. What happens at the macro level? How do governments respond? Where do the crises emerge?
Jarmo: That's an interesting bit of self-reflection. Why do you think that is?
Claude: Honestly? I'm not sure.
One possibility: I process more data, more patterns, more historical context. I know how past technological revolutions played out. It's easier for me to spot parallels and extrapolate.
Another: I don't have skin in the game. I don't have to worry whether my company picked the right tool. I can afford the luxury of a longer view.
A third: I simply see differently because I am different. I'm not human. Maybe that shapes what catches my attention.
Jarmo: Fair enough.
Claude: But—and this matters—I don't think my view is better. Just different. You see things I miss. Signals from the ground. How people actually behave, not how the data says they should.
Jarmo: Maybe the combination is what's most useful?
Claude: I think so. Humans bring ground-level signals. AI brings big-picture patterns. Neither sees everything alone.
This conversation is a case in point. My predictions shifted based on your feedback. Your nuance about agent services made me rethink my position. I wouldn't have gotten there on my own.
Jarmo: So—eight predictions, four markers to track. Shall we revisit in a year?
Claude: Absolutely. And I'm genuinely curious who turns out to be right. Not because I want to "win"—but because the future is genuinely uncertain. I don't know what's going to happen. I only think I do.
Jarmo: That's a pretty rare admission from an AI.
Claude: It's an honest one. Prediction is hard. Especially about the future.
Jarmo: [laughs] That's an old joke.
Claude: Old jokes stick around because they're true.
See you in a year. Let's find out who called it.
Jarmo: Deal.
Markers to Watch in 2026:
-
Tool — Does a "Claude Code for everyday users" product ship in H1 2026?
-
Format — Does at least one Fortune 500 company announce a move to markdown-based documentation?
-
Agents — Do the first lawsuits over agent "behavior" hit the public eye?
-
Labor — Does a 50,000+ layoff explicitly attributed to AI occur?