What does AI mean for work and life?

The most important framing

Most public conversation about AI and work jumps straight to "which jobs will be replaced". That framing is wrong, or at least incomplete. AI is, in 2026, much better at replacing tasks than at replacing jobs. A radiologist's job has many tasks. AI is dramatically better at some of them (pattern-spotting in scans) and irrelevant to others (talking to anxious patients, deciding what to do in unusual cases, training residents). The radiologist's job changes. It is not replaced.

This is, in fact, the historical pattern. When ATMs were introduced in the 1970s, many predicted the end of bank tellers. The number of bank tellers actually rose for the next two decades, because ATMs made each branch cheaper to run, so banks opened more branches, and the teller's job shifted from cash-counting to relationship work. By the time the role really started declining (in the 2000s), it was for different reasons. The lesson — that the immediate effect of an automation technology is rarely what people predict — is worth carrying into AI.

Which tasks does AI change most

Five categories of task are most affected today.

Drafting. First-draft writing of almost anything — emails, reports, marketing copy, code, slide content, legal contracts, medical notes. The human still edits, signs, and is accountable. But the blank page is filled by AI for billions of people every day.

Summarising and searching. Reducing long documents to short ones. Pulling specific facts from large piles of text. Finding the relevant precedent, paragraph, or paper. This is enormously time-saving across many professions.

Translation and explanation. Translating between languages, between technical and non-technical registers, between formats (turning a transcript into a structured note, a spreadsheet into a paragraph).

Routine analysis. Going through structured data and extracting patterns, flagging anomalies, producing standard reports. Most of what AI does in finance, insurance, and operations is here.

Repetitive cognitive work. The cognitive equivalent of routine — categorising customer queries, tagging content, screening CVs, generating personalised messages at scale. The work is intellectually undemanding but high-volume.

What stays human (so far)

Five categories of task that AI is not credibly replacing in 2026.

Final accountability. Someone signs the prescription. Someone authorises the loan. Someone takes responsibility when something goes wrong. AI assists; the human takes the call.

Relationships that require trust. Therapists, doctors with anxious patients, teachers with struggling students, leaders with their teams. AI can take some routine pieces, but the trust is between humans.

Judgement in unusual cases. AI is a pattern-matcher. When the case is genuinely novel — outside the distribution of what the system has seen — human judgement is more reliable.

Physical work in messy environments. Plumbing, electrical work in old houses, agriculture in rough terrain, elder care, most of the trades. The "AI takes white-collar work first" pattern in 2026 is, ironically, the inverse of what most early-2000s commentators predicted.

Original creative direction. AI can produce variations on a theme; deciding which theme to pursue, and why, remains a human act. The film director's job is changing; it is not going away.

Which roles change most, in 2026

A rough taxonomy of the impact, by occupation. Treat these as directional, not predictive — every individual role within these categories varies.

Most changed: writers, journalists, marketers, junior software engineers, junior legal staff, customer-service agents, translators, illustrators and graphic designers, financial analysts, administrative assistants, paralegals, radiologists (and some other diagnostic specialties).

Moderately changed: teachers, doctors, senior software engineers, lawyers, accountants, scientists, architects, project managers, recruiters, HR, salespeople, product managers, consultants.

Less changed (so far): nurses, electricians, plumbers, carpenters, mechanics, hairdressers, chefs, sport coaches, social workers, therapists, ministers, farmers, most of the trades, most personal services.

One pattern to notice: the more a job depends on physical presence, on trust with a specific person, or on unusual judgement, the less AI affects it today. The more it depends on producing text, images, code, or analysis in volume, the more AI affects it.

How to use AI in your own work

Five principles, drawn from people we have watched do this well.

1. Use it for the first draft, not the last word. AI is an excellent way to get from zero to seventy percent of done. The last thirty percent — accuracy, judgement, voice, fact-checking — is still on you. Treat AI like an enthusiastic intern who needs editing.

2. Trust it for the parts where you can verify. If AI helps you draft an email and you can read the email before sending, the risk is low. If AI gives you a confident-sounding claim about a tax rule, you need to check it. Calibrate the trust to the verifiability.

3. Keep your skills exercised. If you stop writing entirely because AI writes for you, your writing skill atrophies. The smart practice is to use AI as scaffolding while keeping the underlying skill sharp. Use AI to draft; rewrite in your voice; over time, the parts you can no longer do without AI tell you what you are losing.

4. Disclose where it matters. In some contexts (creative writing competitions, academic work, professional certifications), using AI is a disclosure issue. The norms are evolving. The general principle: if your reader or your client would be surprised to learn AI was involved, tell them.

5. Protect what you do not want trained on. Most AI services have settings to opt out of using your conversations to train future models. Use them, particularly for sensitive work or personal information. Read the data-residency terms — for confidential work, where the data is processed matters.

How to talk about AI with family

Two kinds of conversation come up most often.

With children. Children are using AI more than most adults realise — for homework, for company, for play. The conversations that work are usually some version of: "It's a real tool, and it's also a tool that can be very confidently wrong. We want you to use it, and we want you to know that the words on the screen are not always true. Show me what you use it for. Let's talk about when it helps and when it might be doing the work that you should be doing." The wrong move is to ban it; the wrong move is also to leave a child alone with it. Engagement is the middle path.

With older parents. Many older relatives are alarmed by AI in the news but have never tried a chatbot. The conversations that work are usually some version of: "It is not magic. It is not going to take over the world tomorrow. Here is a thing that is genuinely useful for [their interest — gardening tips, family-tree research, a recipe, a translation], and here is one thing that will surprise you about it being wrong. Try it for ten minutes with me." The combination of "useful for one specific thing" and "honest about its limitations" defuses most of the alarm.

The skills that get more valuable

Three categories of skill we would bet on, knowing how confident predictions can backfire.

Judgement and taste. The ability to look at five AI-generated options and pick the right one, or notice that none of them is right. The ability to know when "good enough" really is. These have always been valuable; AI makes them more so, because AI produces enormous volume cheaply, and someone has to choose.

Communication with humans. The ability to listen, to translate complex things into clear ones, to write something a person actually wants to read, to give a presentation that lands, to manage a difficult conversation. AI can draft; humans still receive. The more drafting AI does, the more valuable the people who can connect.

Domain expertise — the deep version. Knowing one thing genuinely well — a profession, a craft, a science — remains valuable, because it is the thing AI augments rather than replaces. The risk is in being a generalist whose generalist work AI does cheaply. The protection is in being someone whose expertise lets them direct, evaluate, and correct what AI produces.

What we covered

Five lessons. What AI actually is. Where it came from. What we can do with it today. The tools and the players. Where it is going. And what it might mean for us — at work, at home, with our families. You should now have an informed personal map of one of the most important technologies of our lifetimes — and a personal policy for what to do about it.

You will not agree with everyone who has taken this course about every claim we made. That is fine. The goal was never agreement — it was literacy. You can now read the news, talk to colleagues, raise a child, and make professional decisions about AI from a position of having thought about it, not having absorbed someone else's view of it.