Remember when every startup pitch deck featured “blockchain” on slide three? That was 2018. By 2020, most of those projects had quietly disappeared. The technology survived — it just settled into a much narrower set of useful applications than the evangelists predicted.

AI is following a similar arc, but with one important difference: the useful applications are genuinely broad. The question isn’t whether AI matters. It’s whether the current level of excitement matches the current level of capability.

The Peak Has Passed (Sort Of)

Gartner’s hype cycle is an imperfect model, but it’s useful shorthand. If generative AI hit its “peak of inflated expectations” in late 2023 and early 2024, we’re now somewhere on the downward slope toward the “trough of disillusionment.”

That’s not a bad thing. The trough is where reality sets in, where companies stop buying tools they don’t need, and where actual value starts to emerge.

You can see the signs everywhere. Venture capital for AI startups has cooled from its 2024 highs. Enterprise pilots that launched 18 months ago are being evaluated on actual ROI rather than potential. And the conversation has shifted from “AI will change everything” to “where specifically does AI help us?”

According to Deloitte’s State of Generative AI report, only about 25 percent of enterprise AI pilots move into production. That number hasn’t changed much despite all the excitement. The bottleneck isn’t the technology — it’s the organisational readiness.

What’s Actually Working

Let’s be specific about where AI is delivering real results right now.

Customer service automation. Chatbots and AI-assisted support tools have improved dramatically. They don’t replace human agents, but they handle the routine queries — password resets, order tracking, FAQ responses — that used to eat up 40 percent of support team capacity. Companies like Zendesk have baked AI directly into their platforms, and the results are measurable.

Code assistance. GitHub Copilot and similar tools are genuinely making developers faster. Not by writing entire applications, but by handling boilerplate, suggesting completions, and speeding up routine tasks. Studies suggest a 20 to 30 percent productivity gain for experienced developers.

Document processing. Extracting data from invoices, contracts, and forms. This used to require expensive OCR solutions with poor accuracy. Modern AI handles it far better, and the ROI calculation is straightforward.

Content drafting. First drafts of marketing copy, internal communications, and reports. Not finished work — but a starting point that saves hours of staring at a blank page.

What’s Not Working (Yet)

The gap between demo and production remains enormous for many use cases.

Autonomous AI agents that can handle complex workflows end-to-end? Still mostly vapourware in production environments. The demos look incredible. The real-world error rates are too high for anything mission-critical.

AI-driven decision-making in regulated industries? Moving slowly, and for good reason. When a model can’t explain why it made a recommendation, it’s hard to satisfy compliance requirements.

Full replacement of knowledge workers? Nowhere close. The people predicting that entire departments would be automated by 2025 were wrong. What’s happening instead is augmentation — existing workers getting more done with AI assistance.

The Smart Play for 2026

If you’re running a business in Australia or anywhere else, the smart approach right now is measured optimism. Don’t ignore AI. But don’t bet the company on it either.

Start with the boring use cases. The ones with clear ROI and low risk. Document processing. Customer service triage. Internal search. These aren’t exciting, but they pay for themselves quickly.

If you need guidance on where to start, firms like Team400 work with businesses to identify the AI applications that actually make sense — rather than chasing whatever’s trending on LinkedIn this week.

Build internal capability. Train your existing team. Create a small AI competency group that can evaluate tools, run pilots, and share learnings across the organisation.

And above all, be honest about what you’re seeing. If a pilot isn’t delivering results, kill it. If a vendor’s promises don’t match reality, walk away. The companies that’ll win with AI in the long run are the ones making clear-eyed decisions today, not the ones chasing hype.

The technology is real. The potential is enormous. But the path from here to there is longer and messier than most people want to admit. That’s fine. The best investments usually are.