The Future of AI in 2025 and Beyond – Trends I'm Paying Attention To
By Don Hoang
Artificial intelligence is moving fast-but not always in the ways people expect. The hype cycles are loud, but beneath them, some very real, very practical shifts are reshaping how businesses operate, how products are built, and how work gets done.
I've spent the past decade working with high-growth companies at Uber, Revolut, Atomico and more recently as an advisor and investor across the AI, fintech, and infrastructure space. These days, I spend most of my time working with founders and early-stage teams navigating how to use AI responsibly and effectively.
If you’re here because you Googled my name-hi, I’m Don Hoang. This site is where I write about what I’m thinking, learning, and seeing. Below are 10 trends in AI that I’m paying close attention to in 2025-not just because they’re interesting, but because they’re shaping what gets built, what gets funded, and what actually works in the real world.
1. AI Agents Are Starting to Work for You
We’re moving past the stage where AI just answers questions. Now it’s taking actions.
Instead of asking a chatbot to “write an email,” imagine AI that:
Drafts the message
Schedules the meeting
Updates your CRM
Follows up if there’s no reply
This is the shift from static chat to agentic AI-systems that can plan, decide, and execute across multiple tools. The early results are promising. In some cases, they’re replacing the need for entire ops teams or entry-level roles. In others, they’re quietly boosting the productivity of small teams without any new hires.
Of course, we’re still early. Most agents today are brittle and prone to fail at the edge cases. But the direction of travel is clear: over the next few years, expect AI to become an active participant in your workflow-not just a smart autocomplete.
2. Your Data Is the Real Differentiator
Models are getting better, faster, and cheaper. But they’re also becoming easier to access. GPT-4o, Claude, Mistral, LLaMA-all powerful, all available. So how do startups stand out?
The answer is proprietary data.
The best AI companies I’m seeing today are winning because they:
Own hard-to-get datasets
Understand how to label and structure that data
Build feedback loops so their model keeps improving
This matters whether you’re building in fintech, health, legal, or logistics. The model doesn’t matter nearly as much as the data it's trained on-and how often that data updates and sharpens the product.
I always ask: What do you know that no one else knows? That’s where the moat is.
3. AI That Solves Real Problems in Specific Industries
The most exciting AI products right now aren’t trying to do everything. They’re built for one job, in one industry, and they do it better than anything else.
Examples I’ve seen recently:
Legal platforms that draft and review contracts
Construction tools that read technical drawings and flag risks
Healthcare products that help with triage or billing compliance
Fintech tools that underwrite loans faster and more fairly
These are the kinds of companies that rarely go viral on Twitter but they’re growing fast, delivering real value, and becoming embedded in workflows.
They also tend to build trust more easily. Instead of asking users to learn a new behavior, they slot into how the work already happens-just faster, safer, and smarter.
4. Humans Still Matter, In Fact They Matter More
There’s been a lot of hand-wringing about AI replacing jobs. Some of it is justified. But what I’m seeing more often is something different: AI as a tool for amplification.
Good AI products don’t take over-they partner. They help a junior team member get up to speed faster. They help an analyst find insights more quickly. They give a lawyer or a marketer or a designer a head start.
In most cases, AI still needs human judgment. It’s just giving us more leverage. For anyone building a team or a product in 2025, that’s worth remembering: the human layer is where trust is built. And trust is still what drives adoption.
5. The Infrastructure Layer Is Where the Real Race Is Happening
Underneath all the shiny apps is a fierce infrastructure race. The biggest AI companies are competing on compute, latency, and control.
Model providers like OpenAI, Anthropic, Mistral, and Meta are in a sprint to release the fastest and most powerful models. Chipmakers like NVIDIA, AMD, and Groq are pushing the limits of performance. And cloud providers-AWS, GCP, Azure, and newer players like CoreWeave-are figuring out how to scale it all.
Why does this matter? Because latency, reliability, and cost will define who wins in applied AI. Teams that get their infrastructure stack right will move faster, iterate better, and deliver a stronger experience for users.
If you’re building something AI-native, infrastructure is not a footnote-it’s a strategic choice.
6. Regulation Is Becoming Product Strategy
AI is moving into sensitive territory: health, finance, education, law. Which means the old move-fast-and-break-things mindset won’t cut it.
New rules are already rolling out. Europe’s AI Act is here. The U.S. is debating regulation. And global standards around bias, safety, and transparency are evolving quickly.
The smartest teams I know are baking trust and compliance into their product from day one. They’re documenting how their models work. They’re reviewing data sources. They’re thinking about audit trails and red-teaming and how to communicate risks clearly.
That’s not just about avoiding trouble. It’s becoming a growth advantage. If your customers trust your AI, they’ll adopt it faster. If regulators understand how it works, you’ll ship sooner.
Good governance isn’t a blocker. It’s a multiplier.
7. Emerging Markets Might Leapfrog the Rest
One of the things I’m most excited about: how AI might land differently in emerging markets.
In places where there’s less legacy software and more mobile-first behavior, AI could be a leapfrog moment. Imagine AI tutors in education systems that don’t have enough teachers. Or diagnostic tools where healthcare infrastructure is limited. Or AI-powered credit underwriting where traditional banking hasn’t reached.
These markets have real constraints-but they also have fewer assumptions. And that means they can often move faster when the tech is ready.
8. AI Is Becoming Invisible (And That’s a Good Thing)
The best AI products I’ve used lately don’t scream “AI.” They just feel smart. Gmail that finishes your sentences. Notion that summarizes your notes. Meeting tools that write your recaps.
AI is shifting from being a feature to being part of the fabric. That’s a good thing. It means we’re past the novelty phase. Users aren’t impressed that something uses AI-they care whether it works.
For builders, the lesson is clear: lead with value, not with tech. The AI should disappear into the experience.
9. Open Source Is Gaining Real Ground
Closed models like GPT-4 are impressive. But open-source models are catching up quickly and in many cases, winning on flexibility.
Companies that care about cost, control, or custom data often prefer to fine-tune their own stack. With the right team, an open model like Mistral or Phi-3 can outperform a closed one on the task that matters most to your users.
This isn’t just a technical decision-it’s a strategic one. Choosing open means owning more of your roadmap. It means trading convenience for control. For many teams, that’s the right trade.
10. This Is Just the Beginning
We’re still early. For all the progress, AI is still clunky in places. The tech breaks. The interfaces lag. The trust isn’t always there.
But the curve is steep. Every month, the tools get sharper. The primitives get better. The patterns get clearer.
If you’re building, now’s the time. If you’re investing, now’s the time. If you’re just trying to understand what this all means, now’s a good time to lean in.
Thanks for reading. I’m Don Hoang, and I write here at www.don-hoang.com about what I’m learning, building, and seeing at the edge of AI, tech, and startups.
