AI in 1, 5 and 10 Year - What Happens Next?

Episode transcript

Transcript provided for accessibility and search. Minor transcription errors may remain.

Introduction

SPEAKER_00 0:04

This is AI Security Multipliers, AI, Cybersecurity, Governance with Steve. This week, where is AI going? We look at its possible future. Artificial intelligence is moving extraordinarily quickly. But what happens when we look beyond the next model release, the next agent, and the next AI announcement? What will AI look like one year from now? Five years from now? Ten years from now? And what might happen beyond that? More importantly, what does it mean for the way we work, the way we secure our organizations, and perhaps even the way we think about intelligence itself? To look beyond today's AI hype, here's Steve.

SPEAKER_01 0:52

I should probably begin with a disclaimer. I don't know what artificial intelligence will look like in 10 years, neither do you. Neither does Microsoft, and neither does OpenAI, neither does Google. And despite what somebody on LinkedIn confidently posted this morning, they don't either. Predicting technology 10 years into the future is an excellent way of creating

Predicting the Future Is a Terrible Idea

SPEAKER_01 1:11

something people can quote back to you 10 years later. In 2007, the iPhone had just appeared. In 2012, most people hadn't heard of deep learning. In 2016, very few people outside the technology world knew what a transformer was. And in 2021, if you'd told most employees that within a couple of years they'd routinely have conversations with an AI capable of writing software, analysing documents, and creating images, they might have been slightly skeptical. So this episode isn't really about prediction, it's about direction. Because while we can't know exactly which models, products, or companies will dominate, we can see some fairly significant changes beginning to happen. AI is becoming cheaper, more capable, more autonomous, more integrated, and increasingly less visible. So rather than trying to predict the next 20 model releases, I want to look at three points on the horizon. One year, five years, ten years, and then a little bit beyond. I'm Steve, and this is AI Security Multipliers. Let's see how badly this episode ages. Let's start with the easiest one. One year from now, 2027. And my prediction is AI gets boring. Not because development stops. Quite the opposite. The technology will continue moving incredibly quickly, but using AI will become considerably less remarkable. Today we still talk about AI as something separate. We're implementing AI. We're developing our AI strategy, we're deploying Copilot. We're testing agents. Eventually that starts sounding strange. Nobody announces, but we're implementing databases. Or we've developed our internet strategy. Hopefully, AI increasingly becomes a feature of the technology we already use. Your email has AI, your browser has AI, your security tools have AI, your ERP has AI, your CRM has AI, your development environment has AI, your operating system has AI, and increasingly you don't necessarily even know which model is underneath it.

2027: AI Becomes Normal

SPEAKER_01 3:09

Nor do you particularly care. You ask for something, it happens. And I think that transition is important, because technologies often become most transformative when we stop noticing them. Electricity changed the world. We don't walk into a meeting room and say, excellent, I see you've implemented electricity. We just expect the lights to work. AI starts moving in that direction. But something else happens over the next year. Agents become much more normal. We've already started seeing this. AI isn't simply answering questions anymore, it's beginning to perform tasks. Research something. Prepare something, monitor something, schedule something, investigate something, change something. Increasingly you give AI an objective rather than a prompt. And that creates a significant shift. Because AI starts moving from being a tool you use to something that works on your behalf. At first, most organizations will probably keep the autonomy fairly constrained, and they should. Human approval. Limited permissions. Specific tools, defined tasks, good logging, but gradually confidence grows. Organizations identify tasks where agents are reliable enough, and people stop approving every individual action. That's when things get interesting. Because the productivity benefit doesn't really come from generating an email three seconds faster. It comes when AI can complete an entire process that previously required someone to move information between five different systems. And by 2027, I think we're going to see much more of that. Not artificial general intelligence, not robot overlords. Just lots of incredibly useful automation that happens to be able to reason. Which may actually be more disruptive. And here's where I think many organizations are going to get surprised. The biggest problem won't necessarily be the AI. It'll be everything around it. Identity, permissions, data, process ownership, integration. Governance, accountability, operating models. All the boring enterprise things that suddenly become incredibly important when something can act autonomously. An organization might have an extraordinary AI agent capable of analyzing customer information and recommending the perfect action. Wonderful. Except the customer data is spread across 11 systems. Half of it is duplicated, three systems disagree. Nobody knows which one is authoritative. And

Agents Start Doing the Work

SPEAKER_01 5:21

apparently the only person who understands the integration retired in 2019. AI doesn't magically remove organizational complexity. Sometimes it simply discovers it much faster. So I think 2027 becomes the year many organizations realize something important. Buying AI was the easy part. Changing the organization so AI can actually work safely and effectively, that's harder. Now let's jump forward. 2031. This is where prediction becomes considerably more dangerous. So please keep this recording somewhere safe and resist the temptation to send it back to me. But if the current direction continues, I think five years from now we'll stop thinking primarily about AI assistants. We'll increasingly think about AI workers or digital workers. Or agents, or whatever marketing department wins the naming competition. The terminology doesn't particularly matter. The operating model does. Imagine a finance department. There are people, there are applications, and there are hundreds of agents. One monitors invoices, one investigates exceptions, another prepares forecasts, another reconciles accounts, another monitors contracts, another produces management reports. Humans aren't necessarily performing every step anymore. They're managing outcomes, handling exceptions, making judgments, setting priorities, and supervising systems that perform much of the underlying work. That's a very different organization. And inevitably, we have to talk about jobs. Will AI replace jobs? Yes, some. I don't think we should pretend otherwise that technology has always eliminated certain types of work. AI will too. But that's not the whole story. Most jobs aren't one task. They're collections of tasks, some repetitive, some analytical, some creative, some interpersonal, some administrative, some requiring judgment. AI will absorb different portions of those jobs at different speeds. So for many people, the more immediate question

2031: AI Becomes a Workforce

SPEAKER_01 7:15

isn't will AI take my job? It's which parts of my job will AI take? And then what does my job become afterward? That's much more interesting. Maybe an analyst who spends 60% of their time gathering information spends 10% doing that. What happens to the other 50%? Hopefully they perform more valuable analysis, but there's another problem. What if gathering that information was how junior analysts learned? What happens when AI performs the entry-level work through which humans traditionally developed expertise? Law, finance, cybersecurity, software development, consulting, medicine, engineering. We've built professions around apprenticeship. You start with simpler work, you develop experience, you make mistakes. Someone more experienced corrects you. Eventually you become the experienced person. If AI removes large amounts of junior work, we need another way of creating senior people. That's a generational problem I don't think we're talking about nearly enough. And here's another prediction. I don't think AI necessarily makes managers less important. I think good managers become more important. Imagine managing five employees today. Now imagine managing five employees and fifty AI agents. Who's deciding what work gets delegated? Who's evaluating the output? Who's deciding when human judgment is required? Who's resolving conflicts between agents? Who's accountable for the outcome? Who's deciding which processes should be automated? Who's making sure people are actually developing? Management becomes less about supervising activity and more about designing work. That could be a significant change. The best managers may become something closer to orchestrators. People, AI, systems, vendors, data, all working together toward an outcome. And somewhere in the middle of all of that, there will still be a meeting that could have been an email. Some technologies are apparently beyond disruption. Cybersecurity

What Happens to Jobs and Management?

SPEAKER_01 9:05

changes too. Today, most security architecture ultimately assumes people initiate activity. Someone logs in, someone opens an application, someone accesses information, someone executes something. By 2031, that assumption may no longer hold. Machines initiate activity. Agents authenticate. Agents access information. Agents call APIs. Agents communicate with other agents. Agents execute processes. Agents make decisions. And some of those decisions happen at machine speed. So cybersecurity increasingly protects two workforces, the human workforce and the digital workforce. Identity becomes even more important. Authorization becomes dynamic. Monitoring becomes behavioral. Least privilege becomes essential. And security controls increasingly need to operate at the same speed as the systems they're protecting. You can't have an autonomous agent making 10,000 decisions in an hour. And a governance process that meets every second Thursday, the controls themselves have to become more automated. Which means, yes, we'll increasingly use AI to govern AI. That sentence should make everyone feel completely relaxed. Now let's go to 2036. Ten years? This is where I'm going to become considerably less specific. Because ten years in technology is enormous. But here's the direction I find interesting. I don't think we'll talk about AI nearly as much. Because AI will be everywhere. It becomes infrastructure. Not a product category, not a project, not necessarily even a strategy, just part of how technology works. Software understands intent, systems adapt, interfaces become conversational, agents coordinate work, machines increasingly understand context. And perhaps most importantly, we stop caring quite so much about the underlying application. Think about how enterprise software works today. You want customer information? Open CRM. You want financial information? Open ERP? You want a document? Open SharePoint? You want communication? Open Teams or email? You want analytics? Open another app. We've spent decades teaching humans

Cybersecurity in 2031

SPEAKER_01 11:03

how to navigate software. But if AI can navigate the software for us, the interface changes. Instead of asking, which application do I need? You say, show me why customer profitability declined last quarter and recommend what we should do about it. The AI figures out where the information lives, which systems it needs, which tools it should use, which analysis it should perform. The application starts disappearing behind the outcome. That's potentially enormous. I've sometimes wondered whether conversational AI could eventually become something close to the last major computer interface. Not literally the last interface we'll ever invent. That would be a particularly foolish prediction. But think about the progression. Command lines, graphical interfaces, web browsers, touch screens, apps, and now intent. Instead of learning how the computer wants us to work, the computer increasingly learns what we're trying to accomplish. Voice, text, images, video, context, perhaps eventually gestures and environments. The interface becomes less about navigating menus and more about expressing intent. And if that happens, there are enormous consequences. For software companies, for SaaS, for user experience, for enterprise architecture, and definitely for cybersecurity. Because if users no longer interact directly with applications, how do we authorize actions? How do we establish intent? How do we prove the

2036: AI Becomes Infrastructure

SPEAKER_01 12:25

user actually requested something? How do we distinguish the user from the agent acting for them? What does accession mean? What does least privilege mean? What does non-repudiation look like? The interface may become simpler. The architecture underneath it becomes considerably more complicated. Technology does enjoy doing that. But I think there's an even bigger possibility. For most of human history, intelligence has been scarce. Expertise has been scarce. If you wanted legal advice, you needed a lawyer. Engineering, an engineer, software, a developer, cybersecurity, someone who understood cybersecurity. Research, a researcher. Expertise takes years to develop, and because it's scarce, it's valuable. Now imagine a world where reasonably good cognitive capability becomes incredibly cheap. Not perfect intelligence, not omniscience, just competent assistance available to practically everyone. Twenty-four hours a day, in every language, across almost every subject. What becomes valuable then? Maybe it isn't simply knowing things. Maybe judgment becomes more valuable. Curiosity, creativity, trust, relationships, leadership, the ability to ask good questions, the ability to decide what actually matters, and perhaps most importantly, the ability to distinguish a technically excellent answer from a genuinely wise decision. AI may make intelligence cheaper. I'm not convinced it makes wisdom cheaper. And then we get beyond ten years. AGI, artificial general intelligence, superintelligence,

When Intelligence Becomes Cheap

SPEAKER_01 13:50

robotics, autonomous scientific discovery, AI designing AI, AI developing medicines, AI designing materials, AI solving problems. Humans currently can't. Maybe? Some of those things seem increasingly plausible. Some may take much longer than people expect. Some may happen considerably faster, and some predictions will look ridiculous in hindsight. Including almost certainly several I've just made, but I think there's a mistake we can avoid, trying to build a strategy around one precise prediction of the future. Because the future isn't going to cooperate. The model you standardize on will change. The vendor you choose will change, the architecture will change, the threats will change, the regulations will change, the capabilities will change. So perhaps the goal shouldn't be predicting exactly what AI looks like. In 2036, perhaps the goal should be building an organization capable of adapting when we find out. That means governance needs to change. Traditional governance often assumes change happens relatively

Beyond Ten Years

SPEAKER_01 14:48

slowly. We create a policy. We establish a committee. We perform an annual review. We update the standard. Everyone goes home feeling appropriately governed. AI doesn't move like that. Models change, capabilities appear. Employees discover new tools, agents gain new integrations. Vendors introduce new features. Risk changes continuously. So AI governance has to become more adaptive. Principles rather than endless rules. Risk-based decisions rather than treating every AI system identically. Technical controls rather than relying entirely on policy. Continuous monitoring rather than annual assessment. Clear accountability. And an operating model capable of changing. Because the best AI governance framework in the world is useless if it describes technology that stopped existing 18 months ago. Governance can't simply control change. Increasingly, it has to be designed for change. So

The Multiplier

SPEAKER_01 15:44

here's this episode's multiplier. Don't build a 10-year AI strategy. Build an organization capable of adapting for 10 years. Look at four things. Your people. Are you teaching them how to work with AI rather than simply how to use today's AI tools? Your data. Is it governed well enough to support systems you haven't invented yet? Your architecture. Can you change models, platforms, and providers without rebuilding everything? And your governance. Can it respond to new capabilities in weeks rather than years? Because I can't tell you which AI model will dominate in 2036. I can't tell you what the leading AI company will be. I can't tell you exactly which jobs will disappear. And I certainly can't tell you whether we'll have AGI. But I'm fairly confident about one thing. The technology will change probably much faster than we're comfortable with. So don't optimize your organization for today's AI, optimize it for change. That's the multiplier. One year from now, AI probably becomes considerably more normal. Five years from now, AI may increasingly become

The Next Ten Years

SPEAKER_01 16:50

part of the workforce ten years from now. We may barely think about AI as a separate technology at all. And beyond that, anyone confidently telling you exactly what happens next is either considerably smarter than I am, or trying to sell you something, possibly both. But I'm optimistic. I think AI can dramatically expand what people and organizations are capable of doing. It can democratize expertise, accelerate discovery, remove enormous amounts of repetitive work, help us solve problems we currently struggle to solve, and perhaps allow people to spend more time doing the things humans are actually good at. Judgment, creativity, leadership, empathy, relationships, making sense of ambiguity, and deciding what matters. There will absolutely be risks. Cybersecurity risks, economic risks, social risks, governance risks, and probably risks we haven't thought of yet. Our job isn't to stop the future arriving, and it isn't to blindly embrace everything that arrives either. It's to build organizations capable of using these technologies deliberately, securely, responsibly, and with enough flexibility to change when our predictions inevitably turn out to be wrong. Because the most important AI capability over the next 10 years may not belong to the machine. It may be our ability to adapt. I'm Steve. This is AI Security Multipliers. And remember, AI should multiply your capability. Not your mistakes, I'll see you next time.