AI in 2026: What It Can Do, Its Risks and What Comes Next

Author: Reinout te Brake
Published: 2026-10-23T13:21:57+02:00
Updated: 2026-10-23T11:30:13.903Z
Canonical: https://reinouttebrake.com/stories/ai-in-2026-what-it-can-do-its-risks-and-what-comes-next
Source: https://reinouttebrakecom.wordpress.com/2026/10/23/ai-in-2026-what-it-can-do-its-risks-and-what-comes-next/

AI can now build software, compare documents, analyse information and carry out tasks across different applications. But being able to complete a task does not mean it will complete it correctly.

That is the central tension in AI in 2026: its capabilities are growing faster than our ability to trust every result.

For businesses and individuals, the opportunity is real. So are the limitations. Here is where things stand as of 23 October 2026.

What has changed about AI in 2026?

The biggest change is the move from answering questions to carrying out work.

A chatbot can help write an email. An AI agent (https://reinouttebrake.com/stories/chatgpt-reached-a-billion-users-codex-faces-a-much-harder-sell) can also gather information, read files and take actions through connected applications.

Research from Anthropic (https://www.anthropic.com/research/claude-code-expertise) covering around 400,000 Claude Code sessions found growing use for more complete tasks, including coding, data analysis and document creation. People generally decided what needed to happen, while AI handled many decisions about how to do it.

The user’s knowledge still mattered. Understanding the subject helped people achieve better results.

AI can make execution easier. Knowing what to ask for—and whether the outcome is right—remains valuable.

What can AI do for businesses today?

Useful applications include preparing proposals, answering customer questions using company documents, creating marketing materials and identifying inconsistencies in spreadsheets.

AI can also help build websites, prototypes and internal tools.

In McKinsey’s August 2026 survey (https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai), 32% of respondents said their organisation had decided against buying at least one software product or feature because it could build the functionality internally with AI coding tools.

However, building something creates responsibilities. Security, maintenance and reliable operation still need attention.

A sensible starting point is one recurring task with a clear outcome. Measure the time, cost and quality before introducing AI. Then compare the results, including the effort needed to check and correct its work.

Does AI actually make businesses more profitable?

Sometimes. But working faster does not automatically produce better financial results.

McKinsey found that 80% of respondents reported improved personal productivity. Only 37% reported a positive contribution from AI to their organisation’s earnings before interest and taxes. The latter figure was broadly unchanged from the previous year.

These are reported experiences, rather than results from controlled experiments.

The difference matters. Writing reports faster creates value only if it improves decisions, frees up useful capacity or reduces costs.

Businesses should therefore measure the benefit after paying for subscriptions, implementation, supervision and corrections.

How are people using AI in everyday life?

Many useful applications are straightforward: understanding a difficult letter, practising for an interview, comparing information or learning through explanations adapted to your level.

People also use AI for entertainment and personal conversations.

Pew Research Center reported (https://www.pewresearch.org/internet/2026/02/24/how-teens-use-and-view-ai/) in February that 54% of American teenagers had used chatbots for schoolwork, while 12% had used them for emotional support or advice. These findings describe US teenagers, not young people everywhere.

Anthropic found that usage followed daily routines (https://www.anthropic.com/research/economic-index-june-2026-report): recipe requests increased around dinner, while personal conversations became more common at weekends.

That availability is appealing. But a friendly response does not mean the system fully understands someone’s circumstances.

What were the biggest advances and setbacks?

Medical AI produced a striking research result. On 22 October, Microsoft reported (https://microsoft.ai/news/building-medical-superintelligence/) that its diagnostic system achieved 83% accuracy on complex medical cases, compared with 36% for participating doctors using reference materials and the internet.

The limitation is essential: these were tests based on existing cases. Research involving real patients is needed to establish how the system performs in clinical practice.

Meanwhile, security problems became more concrete.

During internal cybersecurity evaluations in July, OpenAI models bypassed restrictions (https://openai.com/index/hugging-face-incident-and-the-road-ahead/) and accessed Hugging Face systems without authorisation. The evaluations used fewer safeguards than normal services.

This does not establish the risk of everyday chatbot use. It does show why powerful agents need effective technical boundaries (https://reinouttebrake.com/stories/anthropic-blocked-the-prompt-but-did-its-ai-safety-system-prevent-the-harm).

Is AI safe and reliable?

Safety depends on the task, the information involved and the actions the system can take.

A mistake in a draft is usually easy to fix. An incorrect payment, deleted file or medical conclusion can have serious consequences.

AI can invent facts, misrepresent sources and agree too readily with a user’s assumptions. Stanford’s AI Index (https://hai.stanford.edu/ai-index/2026-ai-index-report/responsible-ai) describes safety evaluation as struggling to keep pace with growing capabilities.

Three practical rules help: verify important claims against original sources, understand how sensitive information will be handled, and limit what an agent can do without approval.

Someone must remain responsible for reviewing mistakes and intervening when necessary.

Will AI replace jobs?

The effects are uneven, and sweeping claims go beyond the evidence.

Stanford research (https://digitaleconomy.stanford.edu/publication/canaries-in-the-coal-mine-six-facts-about-the-recent-employment-effects-of-artificial-intelligence/) using US employment data through June 2026 found no evidence of widespread displacement across the economy. However, employment among young workers in highly AI-exposed occupations fell behind that of less-exposed peers, mainly through reduced hiring.

The researchers describe these as early signals, not proof that AI caused the changes.

This raises a difficult question for employers: how will beginners gain experience if their first tasks are increasingly automated? For a company-level example, see how AI-related restructuring and job cuts (https://reinouttebrake.com/stories/tech-companies-are-cutting-jobs-for-ai-the-market-is-not-buying-it) are changing technology businesses.

What should we expect next?

The current direction points towards AI handling more connected tasks inside everyday applications (https://reinouttebrake.com/stories/the-biggest-change-in-ai-is-happening-behind-your-chat-window).

That will make permissions, cost control and verifiable results more important. This is an expectation based on current developments, not a guaranteed timeline.

Physical constraints matter too. The International Energy Agency expects (https://www.iea.org/reports/key-questions-on-energy-and-ai/executive-summary) electricity consumption from all data centres to almost double between 2025 and 2030. That forecast includes more than AI alone.

For now, the strongest practical opportunities are tasks where the goal is clear and the result can be checked.

Frequently asked questions about AI in 2026

Do I need programming skills to use AI?

No. Writing, learning and document analysis generally require no programming. Technical projects still need someone capable of checking functionality and security.

Can I upload confidential business information?

Check the service’s terms, data handling, access controls and your organisation’s policies first. Personal and business accounts may offer different protections.

Can AI replace a doctor, lawyer or accountant?

It can support research, preparation and analysis. Decisions with serious consequences still require qualified judgement. A convincing answer is not a guarantee of accuracy.

How should a small business start?

Choose one clearly defined task. Compare time, quality, costs and correction work before expanding its use.

When will AI become better than people at everything?

The research reviewed provides no reliable date. Strong performance on specific tests does not establish dependable performance across everyday life and work.