Monday was the day. A civilisation was going to be wiped out, he exclaimed. Then Mad King Donald realised he had a golf date, so it moved to Tuesday, and then an 11th hour ceasefire was announced and both sides declared triumphant victories. While Trump claimed a “total and complete victory”, Iranian officials said their country had dealt a “crushing historic defeat” to the US & Isreal, while they can’t both be winners, they can both be losers.
Markets were surprisingly calm over Easter, with all major indices sitting within a range between -1% and +1% over the first week of April. This week however has been dictated by the ceasefire announcement in Iran, it is anyone’s guess as to how this plays out from here, but markets greeted the ceasefire news with open arms, surging higher as investors breathed a collective sigh of relief…for now.
Despite the positive sentiment in the stock markets, hopes for rate cuts weren’t immediately reinstated, markets are still (sensibly) concerned on the medium-term effects that this conflict will have on global price rises. Swap markets in the US are expecting inflation of more than 3% over the next year, it was below 2.25% at the start of 2026.
Before I delve into the meat of today’s note (it’s AI again, please feel free to come back next week if you’re AI’d out) we had four Dynamic Portfolios up for review in April, and although the performance varied from boring to Bonkers (below), the unfolding conflict, and the volatility that it has bought with it, is clear to see throughout all portfolios. Despite the shorter term numbers for a couple of the Dynamic options faltering, the long-term numbers remain remarkable.
The Bonkers 3-month portfolio is a favourite of ours, you would have ended up +0.50% over the period, and had you been on a desert island for the past 3 months and returned to see that performance upon your return, you’d have been pretty underwhelmed.
But dig a little deeper and you can see a far more exciting story. In just 4-weeks you would have made 24%, after 8-weeks you would have made 31%, but after 11-weeks you would have been -7% down, all to end (give or take) exactly as you started.
This is why adjusting your stop-losses is so crucial, while some investors are happy to ride it out (if it wasn’t for he who shall not be named then that 31% return could have very conceivably carried on rising). But life happens, war happens, volatility ensues, and when you see gains that big in such a small period of time, it’s important you protect against giving all of it back.
Adjusting your stop loss levels requires accurate and consistent management, but it really is a crucial tool to longer term success. And it is exactly what we have been doing in our Discretionary Fund Management (DFM) service over recent weeks, notably on some European and UK equity funds.
Our recent webinar generated lots of feedback and questions, so thank you, as always. However there were recurring questions on AI, so I thought it might be useful to cover how far it’s come, where it is (and isn’t), and where it might go – with one or two investment angles. Here goes…
For a long time, the benchmark for artificial intelligence was whether it could convincingly imitate a human in conversation. The Turing Test, suggested by Alan Turing in 1950, captured that idea neatly: if you couldn’t tell whether you were talking to a machine or a person, the machine was considered intelligent.
This is how it works. A human judge has a conversation by text with:
One real human
One machine (AI)
Importantly, the judge does not know which is which. After that conversation, if the judge can’t reliably tell the difference, the machine is said to have passed the Turing Test, and you have machine intelligence.
By that test, there is machine intelligence. But there is a problem. This test just means it’s very good at mimicking human responses.
Today’s AI is best understood as an extremely powerful pattern-recognition engine. It absorbs vast amounts of information, figures what a good answer should look like, and then presents it to you.
Imagine you tell your bot (e.g. Chat GPT or Claude) that you have some leftover ingredients – some turkey, a leek, and some puff pastry – and that you want a recipe to use them up. The bot has read millions of recipes, and will look at those ingredients and remember the patterns. It will think "Usually, when these three things are together, the next step is to make a pie."
The bot didn’t do this because it is a chef. It just recognised a pattern of ingredients, from millions of possible recipes, that told it “Turkey and Leek Pie" was the answer. It professionally sets out the recipe, even if your bot has never actually stepped into a kitchen or tasted the food itself.
It deals superbly with “stuff” that is structured – from recipes to coding to searching legal precedents and writing a brief. It gives the impression of understanding because the outputs are coherent and often correct. However, this capability is rooted in recognising patterns, nothing more – it isn’t intelligence nor experience.
This distinction becomes clear as soon as we move from answers to judgement. In real business settings, decisions are rarely clean. Information is incomplete, objectives conflict, and human behaviour introduces layers of complexity that are difficult to formalise. In these conditions, AI does not fail outright, but it reveals its limitations. It can generate plausible recommendations, which a manager will find helpful, as it is better than starting with a blank bit of paper. But the final answer requires judgement, typically grounded in personal experience, and this is what AI cannot do in a messy world.
Most of the AI “experience” comes from existing text, and not that from participating in the world and learning from feedback. As a result, AI tends to be more reliable in situations where the rules are clear and the problem is well defined, and less reliable when context and incentives and emotions are dominant.
Nonetheless, progress is real, even if it doesn’t always look dramatic to those not using it regularly. AI has got much better at logically thinking in steps rather than jumping straight to an answer. This is what regular AI users would have noticed has improved fastest. Not judgement, but its ability to lay out options clearly.
I find it particularly good at researching a topic, drafting a response, and then refining it after feedback. Pushing back against an answer hugely improves the output. This does mean it is important that you have some basic knowledge of the topic. If not, you will be seduced by the quality of the answer, even when wrong or incomplete, and this is where the danger lies with AI – seductive confidence.
The other big danger is where it gives confident answers in spheres where the world is messy and emotional. Just like investing. This is a problem for those who might throw themselves into the arms of AI for investment answers.
Generic bots such as Chat GPT and Claude are still inclined to “cut and paste” traditional investment industry platitudes. Answers are certainly well structured. But they are fundamentally inadequate and dangerous in prevailing conditions. In this context generic AI bots become sophisticated parrots. Applying judgement under uncertainty, and in human systems, will remain a hard problem for some time. At the moment, all too often AI merely generates an impersonation of someone that sounds confident.
Autonomy is a word you will hear used more regularly in the context of AI. At its simplest it means the ability to act independently.
Self-driving cars are often used as the example of autonomous AI in action, but they’re actually quite a limited version. A car can get you from A to B on its own, which is impressive, but the task is tightly defined. It doesn’t decide where you should go, why you’re going, or change plans along the way unless something very specific happens.
A more meaningful example is something like a digital assistant to which you give a loose instruction such as “organise my holiday in Italy.” Instead of asking you constant questions, it would choose destinations based on your past preferences, find and book flights and hotels, adjust plans if prices change, and build out an itinerary. You’re not managing it step by step—it’s handling the whole process.
Now imagine it is doing that in your business. The real shift is that it’s not just doing tasks when asked. With full autonomy AI is taking responsibility for the entire cycle: working out what to do, doing it, checking how it went, and adjusting. It is now less like a helpful assistant waiting for instructions, and more like someone to whom you’ve delegated a job and trust them to get on with it.
This is why unemployment will persistently rise for years to come. Not overnight, but inexorably. And herein lies a huge source of investment uncertainty for a decade and more ahead.
Less people employed means more government support. More government support means more taxation… but there are less people working so tax per worker rises even more disproportionately, and higher taxation hits economic growth, a very negative loop. How about the government borrows more money by issuing additional bonds (gilts in the UK)? With the economy more fragile, bond buyers will want much higher yields to compensate for this fragility… much higher bond yields will harm the stock market and limit the government’s ability to pay back its debts.
This scenario can be applied to all heavily indebted developed economies. This is the context for Dalio’s Stage 6.
Although that dark picture is encapsulated in one dire sounding paragraph, it is merely one possibility, the other is far more rosy…
AI hugely boosts productivity as it makes each remaining worker dramatically more productive, GDP and therefore the tax base could grow even with fewer workers (the question is whether that transition happens fast enough).
Of course, not all jobs will disappear, historically large developments in technology shifts employment rather than eliminating it completely (again, the scale and speed of this change is the known unknown).
Inflation wildcard? AI could also be massively deflationary, notably if the cost of production falls and business become far more efficient, which actually gives central banks more room to keep rates lower, partially countering gilt yield concern we laid out above.
There are plenty of opportunities, there always will be. But, and this is the key message from that analysis, you just need to stay flexible enough to enjoy those opportunities (your attack), avoid the obvious trouble spots, and have a defined stop loss strategy in place for when things go wrong (your defence).