2025 AI Year in Review: The Year AI Became Infrastructure

2025 will be remembered as the year generative AI stopped being a novelty and started being infrastructure. Not everywhere—and not without friction—but the trajectory is clear. Let me walk through the year that was.

Q1 2025: The Hangover Settles

Firms that had over-invested in AI in 2024 were dealing with the reality: the tool works, but people don't automatically use it. Professional services firms that had gone all-in on AI assistants and knowledge bases found themselves paying monthly subscription fees for software nobody used.

The conversation shifted from "Should we do AI?" to "Why isn't this AI thing being used?" That's progress. It's a management problem, not a technology problem.

The firms that thrived in Q1 were the ones that had treated AI as a change management project, not a tech project. They had trained people, incentivized adoption, and integrated AI into actual workflows.

Q2 2025: Agents Started Looking Real

For the first time, agentic AI—systems that could make multiple steps and handle complexity—started working reliably in production. Claude, GPT-4, and other leading models started shipping improved reasoning capabilities. This was the quarter where "AI can handle complex multi-step workflows" stopped being theoretical and started being real.

Professional services firms started experimenting with agentic workflows: research, analysis, draft creation, review, refinement—all orchestrated by an AI system with human checkpoints. The results were mixed, but the potential was apparent.

Q3 2025: The Consolidation

The wild west of AI tools consolidated. Hundreds of "AI for X" startups realized nobody was actually paying to use them. Some merged. Some pivoted. Some shut down. The survivor were the ones integrated with systems firms already used.

Firms also started moving from "What can we do with AI?" to "How do we govern what we're doing with AI?" Governance, compliance, and audit tooling became table stakes.

This was also the quarter where regulatory guidance started arriving. Not comprehensive, but enough to give firms some direction on how to think about AI use, AI liability, and AI disclosure.

Q4 2025: The Accountability Phase

By Q4, firms were being asked by clients, investors, and regulators: "What's your AI strategy?" Vague answers weren't good enough anymore. "We're exploring AI" became "We're using AI to automate X, improving Y by Z%."

The firms with clear stories won. The firms that had measured impact, documented process, and could articulate their strategy looked credible. The firms that had spent money on AI without clear outcomes looked irresponsible.

The Big Themes

1. Productivity Gains Happened, But Slowly and Selectively

By end of year, firms using AI seriously saw productivity improvements: 10–30% depending on the specific workflow and how well they integrated AI into process. Not the 40–50% I predicted. Not the 2–3% that skeptics said. Somewhere in the middle. Real, but not revolutionary.

2. Governance Is Now Part of the Conversation

Every board meeting about AI now includes a governance conversation. Who approves AI use? How do we audit it? What's the liability? These aren't technical questions, they're business and legal questions. And they're shaping how AI gets deployed.

3. Hiring Shifted

Firms stopped hiring "AI specialists" and started hiring "people who can integrate AI into our business." The AI expert is useful, but the person who can solve for the organizational, process, and data challenges is more valuable.

4. Data Quality Became Critical

Firms with poor data infrastructure couldn't deploy AI effectively no matter how much they invested. Firms with organized, consistent data could deploy AI and see results in weeks. This finally made the CTO's old rants about data quality governance feel vindicated.

5. Compliance Pressure Increased

Regulators started asking questions. Not in a heavy-handed way yet, but the questions are coming. What are you using AI for? How do you ensure it's accurate? How do you disclose it to clients? How do you audit it? Firms that didn't have answers are now scrambling.

What Didn't Happen

The jobs apocalypse didn't happen. Professional services firms weren't downsizing because of AI. If anything, the competitive pressure of AI adoption meant hiring for different skills, but not mass unemployment.

The regulatory crackdown didn't happen. We got guidance, but we didn't get the "AI is banned in professional services" kind of regulation that some people feared.

The moonshot AI breakthroughs didn't happen. We got improvements, not revolutions. Model improvements were evolutionary, not transformational.

What's True for 2026

Going into 2026, here's what I'm confident about:

• AI is now standard infrastructure, not a differentiator.
• Competitive advantage goes to firms that integrated it well, not firms that adopted it early.
• Governance and compliance will be more important than model capability.
• Smaller, specialized models will gain share against mega-models as costs matter more.
• The second-order effects (how AI changes client relationships, pricing, competitive dynamics) matter more now than the first-order effects (how much time AI saves).

The Real Takeaway

2025 was the year we stopped asking "Is AI going to matter?" and started dealing with "AI matters, now how do we manage it?" That's not as exciting as the hype cycle narratives. But it's more important. It means the technology matured enough to become boring infrastructure instead of shiny novelty.

That's the sign of genuine maturity.