The case for doing nothing...
...for six months
Every new year I find myself pondering how best to spend the coming year. And this year I’ve been wondering whether the smart move right now is to take six months off.
A crazy idea? Let’s find out…
My year in AI
Over the Christmas break I found myself explaining to friends and family how I use AI. And it got me thinking. What are all the things I’ve made in the last year using AI? So I wrote a list!
Last year I:
Created ~10kloc of code. Or ~8 years worth of code at human rates.
Created 48 hours of music which I like. And 144 hours that I didn’t.
Made three ~5 minute animated movies.
Created two Mr Men style books (story & pictures) for birthday presents for two ‘lucky’ family members.
Planned a summer holiday.
Wrote 50% of a book on “How to be an awesome software engineer”.
And those are the big things. I’ve used AI to get a second opinion on career options, legal matters, Spanish verbs, medical advice, debugging electronics… …the list goes on and on. On some measures my productivity has increased by more than 10 times.
The current state
But as awesome as it is, I have two observations:
AI needs oversight. For now, it’s incapable of getting to done without help. Coding depends on good prompting plus iterating with the model to direct it where you want to go. AI will happily generate a 3 minute song on its own. But it’ll be, err, disappointing. Success requires a human hand on the tiller, gently making corrections.
You need a sense of smell. The outputs from a model are one of: spot-on, obviously wrong or subtly wrong. The latter is hard to catch. It’s where a sense of smell comes in - a gut feel for things that might be wrong. Knowing what question to ask to flush out the problem. What stones to turn over.
These two limitations - the need for oversight and a sense of smell - have led to an unexpected outcome. Twelve months ago I thought AI would be a leveler. It would give everyone a superpower. The gap between the best and worst human ability would shrink across many domains.
But these days it’s becoming increasingly clear it is the smartest who benefit most from AI. Those with domain knowledge. Those who have experience of managing people. Those who can communicate clearly and concisely.
Sam Altman talks about 2025 as being the year of the agent. Now the term agent is poorly defined and has multiple definitions. But all definitions need both these problems to be fixed. For an agent to be useful in the real world it needs to be trusted. And current AI just can’t. Currently it’s like an exo-skeleton. Amazing when combined with a human, but must less impressive on its own.
The case for taking 6 months off
I’m skeptical 2025 will be the year of the agent; models need to improve significantly to achieve the reliability agents need to be useful. Likewise I’m skeptical about OpenAIs claims of achieving AGI (another poorly defined term). But, despite that, o3 promises to significantly advance the productivity of models. So while we might not get agents or AGI, we will get progress. And it’s conceivable o3 leads to another 10x improvement in productivity.
And if it does, then what? A 10x improvement means what took you 6 months now takes 2.5 weeks. So rather than write code now, perhaps we should all go to the beach and come back when o3 is available. All being well we can get the code written in a couple of weeks :).
The case for not taking 6 months off
The beach is tempting. But it’s more complex than this. o3 may bring extraordinary capabilities, but getting the best will likely still need a human in the loop. Developing those skills takes time.
Perhaps the real question isn't whether to take six months off, but how to best use the time before o3 arrives. Maybe focus on developing the skills that will complement AI rather than compete with it? And plan on how to leverage o3's capabilities the moment they become available?
And it’s not just o3 - we can be confident Anthropic (Claude) and Google (Gemini) won’t be far behind either. We just don’t know (yet) what their timelines are.
The decision
For the right problems, in the right domain the productivity gains from current AI are impressive. But this year threatens to make them look modest. Six months on a beach might be tempting, but the real opportunity lies in preparation. When o3 & the other models arrive, success won't come from raw productivity – it will come from being prepared to effectively use the new technology. The world is changing, and the winners will be those who spend these next crucial months developing the skills that complement AI rather than compete with it.