The frontier · Last reviewed 2026-09-17
What AI can and can’t do yet
Area by area: what current AI can do, its real limits today, and the human skill worth developing alongside it. Last reviewed 2026-09-17.
Free — no account needed to read
This page describes the frontier as of 2026-09-17. It moves, so nothing here claims a limit is permanent — only where things stand now, and what is worth developing either way.
Writing
- Where things are now
- Moving fast
- What AI can do
- Draft, edit, summarise and adapt tone for most everyday writing.
- Its limits today
- Having something worth saying, a genuine point of view, and knowing your specific reader.
- What to develop
- Clear thinking, voice, knowing your audience.
- How to practise it
- Write your first draft yourself, then use AI to edit.
Coding
- Where things are now
- Moving fast
- What AI can do
- Write, explain and debug a large share of routine code, and build prototypes quickly.
- Its limits today
- Deciding what to build, designing larger systems, and taking responsibility for reliability and security.
- What to develop
- Problem framing, system design, reviewing and testing code.
- How to practise it
- Review every AI-written change as if a new colleague wrote it.
Image generation
- Where things are now
- Moving fast
- What AI can do
- Produce high-quality images and variations from a description in seconds.
- Its limits today
- Taste, brand consistency, knowing which image serves the purpose, and rights and consent.
- What to develop
- Visual taste, art direction, ethics of images.
- How to practise it
- Write a design brief before generating, then choose against it.
Research
- Where things are now
- Improving
- What AI can do
- Search, gather and summarise sources quickly.
- Its limits today
- Reliability of sources, invented citations, and judging what is actually true or relevant.
- What to develop
- Source evaluation, asking the right question, synthesis.
- How to practise it
- Open and check every source an AI cites before using it.
Data analysis
- Where things are now
- Improving
- What AI can do
- Clean data, run analyses, and produce charts from plain-language requests.
- Its limits today
- Knowing whether the data measures the right thing, and interpreting what a result means in context.
- What to develop
- Statistical sense, asking what the numbers miss.
- How to practise it
- For any chart, write one reason the data might be misleading.
Planning
- Where things are now
- Improving
- What AI can do
- Generate detailed plans, schedules and checklists.
- Its limits today
- Real-world constraints, priorities between competing goals, and adapting when plans meet reality.
- What to develop
- Prioritisation, trade-offs, adapting.
- How to practise it
- Cut any AI plan to its three most important steps yourself.
Physical tasks
- Where things are now
- Slower progress
- What AI can do
- Robotics is advancing in controlled settings such as warehouses and factories.
- Its limits today
- Varied, messy, unpredictable physical environments remain difficult for AI-driven machines.
- What to develop
- Hands-on skills, craft, working safely in real environments.
- How to practise it
- Learn to fix, make or maintain one physical thing.
Social interaction
- Where things are now
- Mixed
- What AI can do
- Hold natural conversations, role-play scenarios and help draft messages.
- Its limits today
- Building real trust over time, accountability to a relationship, and being physically present.
- What to develop
- Listening, empathy, trust.
- How to practise it
- Have one important conversation in person this week.
Leadership
- Where things are now
- Mostly human-led
- What AI can do
- Help prepare, analyse team data and draft communication.
- Its limits today
- Earning trust, taking responsibility for people, and making values-based calls under pressure.
- What to develop
- Responsibility, courage, fairness, communication.
- How to practise it
- Lead something small: an event, a study group, a project.
Judgement
- Where things are now
- Mostly human-led
- What AI can do
- Lay out options, risks and arguments on several sides.
- Its limits today
- Weighing values, owning consequences, and deciding what matters for a specific person or community.
- What to develop
- Decision-making under uncertainty, responsibility.
- How to practise it
- Write down why you made your next decision, then review it in a month.
Creativity
- Where things are now
- Mixed
- What AI can do
- Generate large numbers of ideas, variations and combinations.
- Its limits today
- Original direction, lived perspective, and choosing what deserves to exist.
- What to develop
- Taste, point of view, experimentation.
- How to practise it
- Start one creative piece with your own sketch before any AI.
The areas
- Understand the Machine — You can only direct what you understand.
- Think — Develop your own mind.
- Judge — AI can generate possibilities. You still need judgement.
- Create — When creation becomes cheap, taste becomes more important.
- Understand People — Technology can simulate communication. Relationships still need people.
- Build — Ideas are cheap. Execution is not.
- Learn How to Learn — Don't memorise the world. Learn how to keep learning it.
- Experience Reality — Not everything important can be learned from a model.
- Purpose & Agency — If AI can help you do almost anything, you need to know what you want to do.