Five Years From Now: Predictions Worth Making

Making predictions about technology is a fool's errand. Making predictions about AI specifically is even worse. The field moves fast, the hype distorts everything, and the distance between what's possible in a lab and what's deployed in the real world is measured in years, not months.

But predictions are also how we plan. How we invest. How we prepare. And there's a difference between reckless prediction (AGI by 2027! Robots replace all workers by 2030!) and grounded prediction, the kind that extrapolates from observable trends, acknowledges uncertainty, and is honest about what we don't know.

So here's my attempt at the latter. Five years from now, March 2031: what does the AI landscape actually look like? Not the fantasy version. Not the apocalyptic version. The most likely version, based on what we can see today.

Prediction 1: AI Is Everywhere and Unremarkable

By 2031, AI will be embedded in virtually every software product you use, and you'll barely notice it. Not because it's not doing anything, but because it's become infrastructure. Background technology. Like how autocorrect, spell-check, and recommendation algorithms are technically AI but nobody thinks of them that way.

Your email client will draft replies. Your spreadsheet will suggest analyses. Your IDE will write boilerplate. Your calendar will negotiate meeting times. Your search engine will synthesize answers. None of this will feel futuristic. It will feel like how software works.

This is the most boring prediction and the most confident one. Technology disappears into the background when it matures. The dramatic "look what AI can do!" demos of 2024-2025 will feel quaint by 2031, not because the technology is less impressive, but because everyone will have been using it daily for years.

The implication: if you're building a product, AI features won't be a differentiator by 2031. They'll be table stakes. The competitive advantage will shift entirely to execution, user experience, data quality, and domain expertise. "We use AI" will be as meaningful as "we use the internet" is today.

Prediction 2: The Model Layer Has Commoditized

In 2026, foundation models are a hot battleground. OpenAI, Anthropic, Google, Meta, and others compete intensely on model capability. By 2031, this competition will have settled into something resembling the cloud market: a few large providers, largely interchangeable for most use cases, competing primarily on price, reliability, and ecosystem.

The capability gap between frontier models will narrow to the point of irrelevance for most applications. The difference between the best model and the fifth-best model will matter to researchers and to a narrow set of cutting-edge use cases. For the vast majority of applications, any top-tier model will be sufficient.

This commoditization is already underway. Open-weight models have closed much of the gap with closed models. The rate of improvement in smaller, efficient models has outpaced the improvement in frontier models for practical tasks. By 2031, choosing a model provider will feel more like choosing a cloud provider: important, but not existential.

The winners in this scenario are the companies that build valuable applications and workflows on top of the model layer, not the model providers themselves (who will compete on thin margins). This is analogous to how AWS prints money but the most valuable tech companies are the ones that build on AWS, not AWS itself. Actually, that's not quite right. AWS is enormously profitable. But the point stands that value accrues to the application layer, and the model layer becomes infrastructure.

Prediction 3: The Hype Hangover Arrives, Then Passes

Somewhere between now and 2031, there will be a significant correction in AI expectations and valuations. It may have already started by the time you read this. The pattern is predictable: inflated expectations collide with messy reality, investment contracts, weaker players die, and the surviving companies emerge stronger.

By 2031, we'll be on the other side of this correction. The discourse will have shifted from "AI will change everything overnight" to "AI has changed a lot of things gradually." The breathless predictions will be replaced by pragmatic discussions about specific applications, ROI, and implementation challenges. It'll be boring. That's a good sign.

The hangover will be painful for people and companies caught on the wrong side of it. AI startups that raised at inflated valuations will face down rounds or shut down. Workers who over-indexed on AI hype without building real skills will face a tougher market. Investors who bet on "AI" as a category rather than on specific businesses will lose money.

But the technology itself will be fine. Better than fine. The post-hype period is when technology actually gets useful because the focus shifts from spectacle to substance.

Prediction 4: AI Regulation Is Real, Uneven, and Messy

By 2031, every major economy will have AI-specific regulation in place. The EU AI Act will be fully enforced. The US will have some form of federal AI legislation, probably less comprehensive than the EU's. China will continue its approach of tight control over AI capabilities and deployment. Other countries will fall somewhere on the spectrum.

The regulation will be imperfect. It will lag behind the technology. It will create compliance burdens that disadvantage smaller players. It will have unintended consequences. In other words, it will look like every other technology regulation in history.

What it won't be is absent. The era of "move fast and deploy AI with no guardrails" is ending. By 2031, any company deploying AI in consequential domains, healthcare, finance, hiring, criminal justice, education, will face meaningful regulatory requirements around transparency, testing, bias mitigation, and human oversight.

This regulation will create a new professional category: AI compliance. Similar to how GDPR created a demand for privacy officers and compliance specialists, AI regulation will create demand for people who understand both the technology and the regulatory landscape. If you're looking for a growth career path, this is one worth considering.

The regulatory landscape will also create geographic fragmentation. AI products that work in one jurisdiction may not be compliant in another. Companies will need jurisdiction-specific deployments, testing, and documentation. This is annoying but manageable, and it's exactly what happened with data privacy regulation.

Prediction 5: The Agent Hype Will Have Partially Delivered

In 2025-2026, "AI agents" are the hot concept: autonomous AI systems that can take multi-step actions, use tools, and complete complex tasks with minimal human oversight. By 2031, agents will have partially delivered on this promise.

Where agents will work well: narrow, well-defined domains with clear success criteria and limited blast radius. Automated customer support for routine issues. Code review and testing pipelines. Data processing and analysis workflows. Research synthesis. Scheduling and coordination.

Where agents will still struggle: open-ended tasks requiring judgment, tasks where errors have high consequences, and tasks that require understanding context that isn't in the data. Fully autonomous AI agents that can replace a human's judgment across diverse situations will still be more aspiration than reality.

The practical reality will be "human-in-the-loop" agent systems: AI that handles routine steps autonomously and escalates to humans for decisions, exceptions, and quality control. This is less sexy than the "fire your entire team and replace them with agents" narrative, but it's also far more useful and safer.

The companies that will have succeeded with agents by 2031 are the ones that were realistic about the technology's limitations and designed systems with appropriate guardrails, fallbacks, and human oversight. The ones that deployed fully autonomous agents in high-stakes domains without adequate safeguards will have generated cautionary tales and possibly regulatory backlash.

Prediction 6: Multimodal Is the Default

By 2031, the distinction between "text AI," "image AI," "video AI," and "audio AI" will feel as dated as the distinction between "text websites" and "multimedia websites" does today. AI systems will work natively across modalities: processing text, images, audio, video, and structured data as naturally as humans do.

This means: you'll have a conversation with an AI, drop in a photo, and get a response that references both the conversation and the image. You'll ask an AI to analyze a video and get a written summary with timestamps. You'll describe what you want and get a presentation with text, images, and layout generated together, not as separate steps.

Multimodal AI won't be a feature. It'll be the baseline expectation. Products that only work with text will feel limited, like websites that only display text feel limited today.

The implications for developers: building AI-powered products will require thinking across modalities from the start. The implication for users: interacting with AI will become more natural and less text-centric. The implication for content creators: every format, text, image, audio, video, will be both easier to create and harder to differentiate.

Prediction 7: Privacy and Data Sovereignty Are Competitive Advantages

By 2031, running AI locally, on your own hardware, with your own data, never touching a third-party server, will be a mainstream option. Not just for enterprises with compliance requirements. For individuals who value privacy.

The trajectory of model efficiency makes this inevitable. Models that required data center GPUs in 2024 run on laptops in 2026. By 2031, very capable models will run on phones, tablets, and standard consumer hardware. Apple, Google, and other device manufacturers are already investing heavily in on-device AI.

This creates a split in the AI market: cloud AI for maximum capability and convenience, local AI for privacy, speed, and independence. Many users and organizations will choose local-first approaches, especially as the capability gap narrows.

The companies and products that handle data respectfully, that give users real control over what gets shared and processed, will have a genuine competitive advantage. Privacy won't be a niche concern. It will be a mainstream buying criterion.

Prediction 8: The Skills Landscape Has Shifted, Not Collapsed

By 2031, the workforce impact of AI will be clear, and it will be messier than either the optimists or pessimists predicted.

Some job categories will have shrunk meaningfully. Not eliminated, but reduced. Basic content production, routine data analysis, standard code generation, and simple customer support will require fewer people than they do today. The people still doing those jobs will be more productive and using AI tools extensively.

Other job categories will have grown. AI operations, AI compliance, AI training and evaluation, human-AI interaction design, and various domain-specific AI application roles will have created genuine new employment. Not enough to fully offset the displacement, but enough to matter.

The net effect will be a shift in what's valued: critical thinking, judgment, creativity, domain expertise, and the ability to work effectively with AI tools will command premiums. Routine execution, regardless of domain, will command less. The people who invested in learning to work with AI will have benefited. The people who ignored it will have felt the pressure.

This won't be the apocalypse that doomsayers predicted or the utopia that optimists promised. It will be a gradual, messy, uneven transition that benefits some people and disadvantages others, much like every previous technology shift.

What I'm Not Predicting

Honesty requires acknowledging what I don't know:

I'm not predicting AGI. Whether artificial general intelligence arrives by 2031 is a question I can't answer, and neither can anyone else with confidence. It might. It might not. The timeline estimates from experts range from "never" to "already happened" depending on definitions. I'm making predictions that don't depend on AGI arriving.

I'm not predicting a catastrophic AI event. Could AI cause significant harm through misuse, accidents, or unintended consequences? Yes. Could that harm be catastrophic? Possibly. But predicting specific catastrophic events is not useful planning. Building robust safeguards is.

I'm not predicting specific companies. Which AI companies will exist in 2031? I have guesses, but they're just guesses. The specific winners and losers depend on execution, timing, luck, and factors that aren't predictable.

Why These Predictions Matter

If these predictions are roughly directional, correct even if the specifics are off, they imply some actionable conclusions:

Invest in adaptability over any specific tool or platform. The tools will change. The ability to learn and adapt won't become obsolete.

Build domain expertise. As the model layer commoditizes, domain knowledge becomes the differentiator. Know your industry deeply. AI amplifies expertise. It doesn't replace it.

Take privacy seriously. This is going to matter more, not less. Build systems that respect data boundaries. Choose tools that give you control.

Plan for regulation. If you're building AI products, invest in compliance and governance now. The regulatory wave is coming, and being ahead of it is cheaper than being caught behind it.

Stay grounded. The most useful response to AI isn't excitement or fear. It's pragmatic engagement. Learn the technology. Understand its capabilities and limitations. Apply it thoughtfully. That's always been the right approach to new technology, and it always will be.

Five years is a long time in technology and a short time in human adaptation. The AI of 2031 will be more capable, more accessible, and more integrated into daily life than what we have today. It won't be magic. It won't be terrifying. It will be a tool, a powerful one, doing what tools have always done: extending what humans can accomplish.

The question isn't whether AI will matter in 2031. It will. The question is whether you'll be ready for how it matters. The time to start preparing is now, not with panic, but with purpose.