Mid-2025 Check-In: The AI Space Three Years After ChatGPT

November 2022 feels like yesterday and a decade ago simultaneously. ChatGPT launched, and within weeks the AI conversation shifted from academic to operational. Three years later, here's what actually happened, and what it means for your firm.

The 2022-2023 Period: Shock and Hype

ChatGPT's release was the turning point. Suddenly AI wasn't abstract—it was in people's hands.

The reaction was predictable: hype, fear, and widespread experimentation. Every firm scrambled to "have an AI strategy." Consultants (myself included) got busy. Models improved monthly. Regulations scrambled to catch up.

Reality check: Most of it was vaporware. Lots of pilots, very few production implementations.

2024: The Year of Consolidation

By 2024, the market consolidated. The conversation shifted from "should we use AI?" to "how do we deploy it reliably?"

What happened:

• Enterprise LLMs became standard. Claude, GPT-4, and Gemini separated from the field. Smaller models and open-source alternatives filled specific niches.
• Tool integration exploded. AI started appearing in Salesforce, HubSpot, Slack, Excel. Not as plug-ins—as core features.
• Regulation began to clarify. The EU's AI Act passed. The US took a framework approach. GDPR applied definitively to AI.
• Cost fell and capability rose. By end of 2024, the ROI was undeniable for early adopters.

2025: Where We Are Now (June)

Three years in, here's the true state:

Capability is Solved

Claude Opus 4, GPT-4, and Gemini 2.5 Pro handle almost every professional services use case. The question isn't "can AI do this?" It's "what's the ROI?" The capability plateau has been reached for most practical work.

Adoption Is the Hard Part

Deployment is the bottleneck now. Most firms have access to great tools and understand the technology. Very few have actually integrated it into workflows at scale. The ones that have are winning.

Economics Are Clear

By mid-2025, we have real data:

• Research and analysis work: 2–3x faster with AI. Error rate comparable or better with good prompting.
• Document generation: 50–70% faster. Quality requires human review but saves substantial time.
• Knowledge work automation: 30–50% of routine tasks can be automated or AI-assisted.
• New service offerings: Firms using AI can offer custom analysis or advisory at margins previously impossible.

The ROI is real if you deploy correctly.

Regulation is Stabilizing

As of June 2025, the regulatory space is still evolving but not chaotic:

• GDPR applies to AI like any data processing. Standard data governance covers most cases.
• The EU AI Act is in effect for high-risk systems. Professional services mostly falls in medium risk.
• The US hasn't passed comprehensive AI regulation but does have sector-specific rules (financial services, healthcare).
• Insurance industry is adapting; AI is no longer an automatic exclusion from liability coverage.

You can build compliant AI deployment without legal expertise now. Three years ago, you couldn't.

What Changed Most Dramatically

Three years in, the things that shifted most:

1. Accessibility

ChatGPT required OpenAI. Now you have Claude (Anthropic), GPT-4 (OpenAI), Gemini (Google), and others. Competition dropped prices and distributed capability. Any firm can access world-class AI now.

2. Enterprise Readiness

Early AI was fragile, expensive, and risky. Enterprise offerings from Anthropic and OpenAI now include data guarantees, SLAs, and clear contractual terms. You can build business-critical systems on these models.

3. Integration

Three years ago, AI was a separate system. Now it's embedded in the tools you already use: Slack, Excel, Salesforce, HubSpot. The distinction between "AI systems" and "regular systems" is disappearing.

4. Talent

Three years ago, AI expertise was rare and expensive. Now most mid-career consultants can learn to use these tools effectively in weeks. The bottleneck isn't skill—it's organizational will and adoption.

What Hasn't Changed

Some things I thought would shift but didn't:

• Custom models. Most firms thought they'd need their own models. By mid-2025, nearly no one does. Foundation models are so good that custom fine-tuning rarely makes sense below $50M scale.
• Disruption of professional services. AI is making work better, not eliminating the work. Lawyers, consultants, and advisors are still needed—just more of them are using AI.
• Easy automation. Knowledge work is harder to automate than expected. You need careful workflow design and AI isn't a drop-in replacement for people.

The Three Types of Firms, As of June 2025

Type 1: Leaders (Top 15–20%)

Have deployed AI across multiple functions. Using Claude, GPT-4, and others in production. Clear ROI. Recruiting talent explicitly for AI capability. Will dominate their segment in next 3 years.

Type 2: Followers (Middle 50–60%)

Using AI tools personally and in pockets. Some pilots. No comprehensive strategy. Aware they're behind. Planning initiatives. Three years to catch up is possible if they move now.

Type 3: Laggards (Bottom 20–25%)

Minimal AI use. Often due to risk aversion, regulatory constraints, or simply not prioritizing it. Still have time to catch up, but window is closing.

What Comes Next (June 2025 to Mid-2026)

My prediction for the next 12 months:

• AI integration deepens. The line between "AI system" and "regular system" disappears. Slack, Excel, Salesforce—all have integrated AI by end of 2025.
• Regulation settles. By 2026, most major jurisdictions have clear AI policies. Less uncertainty, easier compliance.
• Consolidation accelerates. Smaller AI companies either get acquired or find niche markets. The main models come from OpenAI, Anthropic, Google, and maybe one or two others.
• Talent competition increases. As more firms adopt, the competition for people who can actually deploy AI effectively will intensify.
• ROI becomes non-negotiable. The hype is over. Boards are asking for evidence. The firms winning are the ones with clear business cases and measured results.

The Bottom Line

Three years after ChatGPT, we're no longer in the "what is AI?" phase. We're in the "how do we win with AI?" phase.

The capability is there. The tools are accessible. The economics make sense. The regulatory environment is manageable. The question now is execution and adoption.

If you're still evaluating whether to invest in AI, you're already behind. The firms that are going to lead their markets in 2026–2027 are the ones deploying and iterating now.