2025 Predictions: AI Agents, Regulation, and the End of 'AI Strategy' as a Standalone Thing

It's the end of 2024, and I'm looking ahead to 2025 with five predictions about where AI goes next. Some of these are technical. Some are organizational. All of them will shape how you should be thinking about AI strategy if you're running a professional services firm.

Prediction 1: AI Agents Start Doing Real Work

The hype around AI agents has been deafening. "Autonomous agents that do your work for you!" Reality so far: most AI agents are overblown. They struggle with complex, multi-step tasks. They hallucinate. They need constant human oversight.

In 2025, that changes. Not because agents become truly autonomous—they won't. But because firms figure out how to use agents for specific, well-defined workflows where human oversight is minimal. Agents that schedule follow-up meetings. Agents that draft initial contract summaries. Agents that route documents to the right person. These things are coming, and they'll be genuinely useful.

What matters: start experimenting with agent frameworks now. By the time they're ready for production deployment, you want to know how they work.

Prediction 2: Model Diversity Becomes The Norm

In 2024, most firms used one or two models. "We use ChatGPT" or "we use Claude." In 2025, sophisticated firms will use five or six different models for different tasks. Haiku for routine work. Sonnet for medium complexity. Larger models for hard problems. Open source for cost-sensitive tasks.

This requires more sophisticated infrastructure—model routing, prompt management, cost tracking—but the ROI is real. You're not paying for peak capability on routine tasks anymore.

What matters: start thinking about multi-model strategy. It's more complex but also more efficient.

Prediction 3: The Regulation Pendulum Swings Back

The new US administration will initially take a light-touch approach to AI regulation. But the lawsuits will pile up. Someone will win a case about AI-generated liability. An AI system will cause real harm to someone important. By Q3 2025, you'll see regulatory pressure building again—less from federal government, more from sector-specific regulators and state AGs.

The result: firms will be caught between "build fast" and "comply with unclear rules." This creates opportunity for firms that built governance early.

What matters: don't abandon compliance thinking just because the regulatory climate feels permissive. Build defensible practices now.

Prediction 4: The Training War Begins

Every firm wants AI to increase productivity. But nobody has trained their people to use AI effectively. In 2025, this gap becomes urgent. Firms that invested in training in 2024 will have competitive advantage. Firms that didn't will struggle.

This leads to a wave of "AI training for professionals" offerings. Some good, most mediocre. The firms that build their own training, tailored to their specific workflows, will win.

What matters: training isn't a nice-to-have. It's a strategic advantage. Invest in it.

Prediction 5: "AI Strategy" Merges With Business Strategy

Right now, many firms treat "AI strategy" as a separate thing from "business strategy." That distinction breaks down in 2025. The firms that win won't have a "head of AI strategy." They'll have business leaders who understand AI, practice leaders who deploy AI, and technology leaders who enable it.

This is harder to organize but also more sustainable. It means AI isn't something you bolt onto your business. It's integral to how you operate.

What matters: start integrating AI conversations into your regular business planning, not just into innovation committees.

The One Thing I'm Genuinely Uncertain About

Will AGI happen? Will the next frontier models have significant capability jumps, or is the curve flattening? I genuinely don't know. Most of my predictions assume continued, incremental improvement. If there's a major breakthrough (or if progress slows), everything changes.

So if these predictions feel off in 6 months, the first thing to check is: did the model space shift dramatically? If not, my track record usually holds.

Looking Ahead

2025 will be less about "is AI real?" and more about "how do we scale this thoughtfully?" That's a more mature question. It's also more boring. But it's also where real value gets created.

Happy New Year. Let's see how I did on this.