Back in December I wrote here about my first weeks with ChatGPT. Four months later, the pace has changed completely. On 14 March, OpenAI released GPT-4. On 23 March, ChatGPT plugins were announced. On 22 March, an open letter called for a six-month pause. On 31 March, Italy blocked ChatGPT. And in April, half the internet is playing with AutoGPT.
Inevitably, the question is back. I hear it from friends, from the entrepreneurs I work with and from people who have nothing to do with tech: “Will our jobs still exist in five years?” My short answer: yes, most of them. The long answer is more interesting, and I will try to give it as calmly as I can.
GPT-4 is a real leap, not a marketing stunt
Let me be honest from the start: I am not in the camp that calls it “just a better autocomplete”. GPT-4 is noticeably better than what we had in December. It reasons better, makes mistakes less often (not never), keeps track of long conversations and, according to the report OpenAI published, passes professional exams, including the bar exam, at high percentiles. OpenAI also announced that the model can understand images, not just text, although that feature is not yet open to everyone.
Plugins may be the bigger news for business. A model that can search, book, calculate or pull data from an app is no longer just a conversation partner. It becomes an assistant that gets things done. The AutoGPT experiments, where the model breaks a task into steps and executes them one by one, are more spectacular than useful for now. But the direction is clear, and it deserves a clear-headed discussion. I recently wrote about how I use ChatGPT in business, in practice.
Tasks get automated, not whole jobs
Every job is a bundle of tasks. A lawyer researches case law, drafts, negotiates, calms down a frightened client and takes responsibility for a piece of advice. A marketer writes copy, analyses numbers, persuades a boss and understands a customer. GPT-4 can do a few pieces of the bundle very well. It can rarely do all of them and, above all, it cannot put its name on them.
When a technology takes over some tasks, the job does not disappear. It gets reorganised. People spend less time on the mechanical part and more on the part where judgment matters. A Goldman Sachs report estimated in late March that generative AI could expose the equivalent of around 300 million full-time jobs to automation. The key word there is “expose”: it does not mean those jobs vanish, but that part of what is done in them can be done differently.
We have seen this movie before
When the spreadsheet arrived on personal computers, many thought it was the end of accountants. The opposite happened: calculations became cheap, so companies started asking for more analyses, scenarios and reports. Accountants stopped adding up columns by hand and started doing more valuable work.
The same goes for the internet and retail. I lived through it myself: I launched the online store MarketOnline in 2005 and the online florist Floria in 2009. Back then, many said physical shops would die. They did not. They changed: some closed, others learned to sell online too, and new jobs appeared, from logistics specialists to people handling ads and customer experience. Retail did not end. It transformed, and those who refused to transform paid the price.
I am not saying history repeats itself exactly. AI is faster and more general than a spreadsheet. But the basic mechanism is the same: technology makes an activity cheap, and demand for everything around it grows.
Which tasks are most exposed, and what stays with people
From what I see in day-to-day work and in the startups I advise, the most exposed are:
- repetitive text: product descriptions, standard emails, summaries, routine reports;
- first drafts of almost anything: an article, a proposal, a presentation, a template contract;
- basic customer support, for the questions that get asked a thousand times;
- simple code: scripts, standard functions, tests, translating from one programming language to another.
Less exposed are the things that require judgment in new situations, relationships built over time, taking responsibility, physical work in the real world and, above all, trust. A client who hands you their business wants to know who answers for it if things go wrong. A language model answers to no one. A plumber, a nurse, a good negotiator or a manager who can hold a team together in a crisis has no reason to panic any time soon.
The real risk: people who refuse to learn
If I had to sum up this whole article in one sentence, it would be this: AI will not take your job, a colleague who knows how to work with AI will. A copywriter who uses GPT-4 for first drafts and puts their energy into the idea and the editing will deliver more, and better, than one who stubbornly ignores the tool. The same goes for programmers, lawyers and salespeople.
I have often told people starting out (I also wrote about what I tell young people who want to become entrepreneurs) that the most valuable skill is learning new things fast. Now it is becoming a matter of professional survival, not just an advantage.
Transitions hurt, and we should not pretend otherwise
Here I want to be as honest as I was at the start. The fact that, in the long run, technology creates more jobs than it destroys does not help much a 50-year-old whose position disappears tomorrow. Transitions have real costs, and they are usually paid by those with the fewest resources to adapt.
A fair society does not leave people alone in the face of change. Companies that gain productivity from AI should invest part of that gain in retraining their own people, not just cut headcount. Governments should fund real retraining programmes, not box-ticking courses, and modernise their own services. And the public debate should be about how we manage the transition, not a choice between “we are all doomed” and “nothing will change”.
The letter calling for a six-month pause, signed by Elon Musk, Steve Wozniak and many others, and the decision by Italy’s data protection authority show that it is not only people worried about unemployment who have questions. I am not convinced a pause is realistic. But questions about safety, personal data and accountability are legitimate and deserve serious answers.
Practical advice
For employees:
- use an AI tool every day, even for half an hour, on real tasks from your work;
- identify the repetitive part of your job and automate it yourself, before someone else does;
- invest in what is hard to copy: relationships, judgment, domain knowledge, communicating with people;
- always check what the model produces. It still makes things up with great confidence.
For entrepreneurs and managers:
- do not cut people out of fear or because it is fashionable. First learn what AI can actually do in your company;
- give your team access, time and permission to experiment, and ask them to share what works;
- set clear rules: which data never goes into a public chatbot, and who checks the output;
- use the time you save for customers and quality, not just for cutting costs.
Conclusion
GPT-4 is not the end of jobs. It is the end of some tasks and the beginning of a new way of working. Those who learn now will have a huge advantage in the coming years. Those who pretend nothing is happening, or who panic, are wasting precious time. And we, as entrepreneurs and as a society, have a duty not to leave anyone behind just because change arrived quickly.
How are you using GPT-4 or ChatGPT in your daily work? And which task would you be happy never to do again?