Every December brings AI predictions, most of which are either vendor hype or attention-grabbing doom. Here are predictions for 2026 grounded in observable trends and realistic extrapolation.

High-Confidence Predictions (>75% likely)

AI Governance Becomes Mandatory

By end of 2026, AI governance frameworks will be expected, not optional. Boards will ask about AI risk management. Auditors will assess AI controls. Customers will require governance assurance.

Why this will happen: Regulatory pressure continues building globally. Incidents have demonstrated AI risks. The EU AI Act creates compliance requirements that affect global enterprises.

What this means: Invest in governance now. Organisations without frameworks will face increasing difficulty.

Productivity AI Rationalisation

Many organisations will reduce productivity AI licensing. The gap between “seats licensed” and “seats actively using” will drive consolidation.

Why this will happen: CFOs will scrutinise ROI. Usage data will reveal low adoption rates. Competitive pressure will reduce per-seat costs, making excess licenses harder to justify.

What this means: Measure actual usage and value. Prepare for vendor conversations about right-sizing.

Model Commoditisation Continues

The performance gap between major foundation models will continue narrowing. Model selection will matter less than application design.

Why this will happen: Multiple providers are investing heavily. Open-source models continue improving. The frontier keeps moving, but followers keep catching up.

What this means: Avoid deep lock-in to single providers. Focus on building good applications, not chasing model upgrades.

AI Spending Growth Slows

Enterprise AI spending will grow, but at a slower rate than 2024-2025. The “throw money at AI” phase is ending.

Why this will happen: Budget scrutiny intensifies. ROI questions demand answers. Economic conditions force prioritisation.

What this means: Compete for AI budget on business case quality. Generic “AI is the future” arguments won’t secure funding.

Medium-Confidence Predictions (50-75% likely)

Agent Technology Improves But Remains Limited

AI agents will handle more tasks reliably, but the “autonomous agent” vision will remain mostly unrealised. Agents will work in constrained domains, not open-ended business contexts.

Why this might happen: Agent reliability is improving incrementally. But open-ended autonomous operation requires capabilities not yet demonstrated at enterprise scale.

What this means: Explore agent capabilities for specific, contained applications. Don’t restructure organisations expecting agent capabilities that don’t yet exist.

Australian AI Regulation Emerges

Australia will introduce AI-specific regulatory requirements, though likely principles-based rather than prescriptive. Something between EU’s comprehensive approach and current minimal regulation.

Why this might happen: The Australian Government consultation is underway. Political pressure exists. But Australian regulatory culture favours lighter touch than EU.

What this means: Monitor regulatory developments. Build flexibility into AI systems. Prepare for transparency and accountability requirements.

Custom AI Development Declines Further

More organisations will conclude that custom AI development is not core competency. Platform and vendor AI will capture more share.

Why this might happen: Custom development is expensive and requires scarce skills. Platform AI keeps improving. The “build vs buy” calculation shifts toward buy.

What this means: Challenge custom development proposals rigorously. Build integration and optimisation capability rather than development capability.

Privacy Enforcement Intensifies

Regulators will act more aggressively on AI-related privacy violations. High-profile enforcement actions will change organisational behaviour.

Why this might happen: Privacy regulators are increasingly focused on AI. Test cases are working through legal systems. Political attention on AI amplifies regulatory response.

What this means: Audit AI systems for privacy compliance. Document lawful basis for data processing. Don’t assume AI use exempts you from existing privacy obligations.

Lower-Confidence Predictions (25-50% likely)

Major AI Incident Changes the Conversation

A significant AI-related incident – whether bias, privacy breach, or operational failure – will trigger regulatory and public response affecting enterprise AI.

Why this might happen: Increased deployment increases incident probability. Stakes are rising. Media attention is high.

Why it might not: Many near-misses haven’t triggered major response. Enterprise AI is generally lower-risk than consumer-facing applications.

What this means: Governance and risk management protect against being the organisation that triggers response.

Open-Source Models Challenge Closed-Source Dominance

Open-source models may become good enough that enterprises shift significant workloads from commercial APIs to self-hosted open-source.

Why this might happen: Open-source model quality is improving rapidly. Cost advantages are significant. Data sovereignty concerns favour self-hosting.

Why it might not: Operational burden of self-hosting is substantial. Most enterprises lack capability. Closed-source providers will compete on integration and ease of use.

What this means: Monitor open-source developments. Build capability to evaluate self-hosting options.

AI Talent Market Cools Significantly

Combination of increased supply (training, education) and reduced demand (AI normalisation, automation of some AI work) could moderate talent shortage.

Why this might happen: Universities have scaled AI programs. Boot camps have proliferated. AI tools increasingly assist AI development.

Why it might not: Demand continues growing. Quality talent remains scarce even if quantity increases. Experience requirements create persistent shortage.

What this means: Don’t assume the talent market will solve itself, but watch for market shifts.

What Won’t Happen in 2026

AGI. General artificial intelligence will not arrive in 2026. This prediction has a long history of being wrong; it will be wrong again.

Mass job displacement. AI won’t eliminate large categories of jobs in 2026. Employment effects will be gradual and distributed.

AI vendor monopoly. No single vendor will dominate enterprise AI. Competition will remain robust.

AI hype ending. Hype cycles will continue. New capabilities will be over-promised. Realistic assessment will remain essential.

How to Use These Predictions

These predictions are tools, not certainties. Use them for:

Scenario planning: Consider what happens if predictions come true or don’t.

Investment prioritisation: Focus on areas with high-confidence predictions; hedge in uncertain areas.

Risk management: Prepare for both likely and potential scenarios.

Strategic conversations: Frame discussions about where AI is going based on realistic assessment.

Don’t use them for:

• Precise budgeting
• Specific technology choices
• Detailed timing

The future remains uncertain. Predictions help navigate uncertainty; they don’t eliminate it.

Final Thought

2026 will likely be a year of maturation, not revolution. AI becomes more normal, governance becomes more expected, and practical value becomes more demanding.

The organisations that thrive will be those that have built foundations – data, governance, capability, realistic expectations – through 2024-2025. Those still chasing hype will continue to be disappointed.

Plan for progress, not transformation. That’s the realistic expectation for 2026.