Investing in the face of AI: beauties or the beasts? [Members]
For MOGULS by The Investor
on February 20, 2026
I don’t know about you, but when I’m confronted with a technology poised to vapourise vast tracts of the economy, to put hundreds of millions out of work, and ultimately to preserve the dregs of human society in a genius robot’s version of an ant farm, well… I look to profit.
What did you expect? This is Moguls. We’ll leave the penning of laments to the poets. Or at least to those people pretending to be poets by using ChatGPT.
This article can be read by selected Monevator members. Please see our membership plans and consider joining! Already a member? Sign in here.
Comments on this entry are closed.
Interesting article indeed. I’ve been in two minds about how the AI “bubble” might or might not pop, or rather if it is a bubble or not. The recent surrender of coders to AI supremacy highlighted in your article suggests to me AI isn’t as poppy as a sheet of bubble wrap many seem to think.
What swung me behind a belief there will be some kind of long future for AI was listening to a podcast by the astronomer David Kipping:
https://open.spotify.com/episode/5YKFoKysfHIskh7t67rn6v?si=EaZRogUYTuCFvDy5x9djVQ
He’d attended a meeting at the institute of advanced studies and came away astonished by the extent AI had penetrated into the work out the top scientists on the planet, despite him having used and even developed AI himself for over a decade.
Interesting listen (and so are Kipping’s other astronomy pods – try the one with Will Kinney on what preceded the big bang)
That’s a great article. It’s a struggle trying to keep up with all the news about AI, never mind trying to pick winners. I am reminded of the 1980’s advert for teletext, “if you want to know, page the ORACLE” and the even earlier evil queen in Snow White, ” mirror, mirror etc”
The productivity gains will only be unlocked if the right prompts are used in fields where gains are possible. So, my thoughts on winners would be those companies that sell the most licences, Microsoft, Apple, Amazon, Netflix. And then add into the mix a bit of uranium, silver and power generation components.
@Ian — Cheers for your sharing your thoughts. Yes I know all sorts of people using AI daily now, for mundane tasks and for more complicated data crunching. At the least we are at the useful power tools stage with them. As for your current view of the ultimate future of the technology (and the implications for the market and appropriate stocks) I think if one is uncertain and in two minds then that’s the exactly correct perspective on the existing evidence! 😉
@Bassavoce — Yes, how the technology is used in day to day practice will determine its ultimate utility, and whether it’s a distraction or a boost. I suspect that for some applications ever more of the prompting will be mediated by ‘thin wrappers’ where you’re essentially using the sort of interfaces we’re already used to in order to interact with the AI. Though that said it’s also true that the natural language element is a big ‘unlock’ in a lot of scenarios.
@all — As soon as I pressed publish I read the following article (also in Weekend Reading links today) that makes the good point that crashing share prices for software stocks could have a reflexive element if it means those companies are less able to hire the talent they need to compete:
https://finance.yahoo.com/news/fund-beating-99-peers-sees-090000453.html
Of course this won’t matter if the downturn is short-lived. But in a prolonged derating it’s a fair additional point.
“he’s resigned himself to never writing code again, except for fun.” – I think your friend needs to have a strong think about what constitutes fun.
It does look like coding is going to go the way of the dinosaurs now. It’s probably no bad thing (I’ve often though it an inhumane past time), and the end product can still be tested to whatever degree you require if you are worried about quality. Possibly this is in line with coding moving away from 1s and 0s over time through ever increasing layers of abstraction. This latest vibe coding is just a really dramatic advance where it’s gone suddenly to just looking like normal (human) language.
So all the software houses will, as you rightly point out, just have to scramble to learn how to orchestrate this new approach to best effect. Winners and losers defined by how well they navigate this step change. Software will get built a lot quicker. But as with everything, it won’t mean more leisure time for (desperately reskilling) engineers for all those hours freed up. It will just mean a ton more software will get produced.
The other thing that looks inevitable is we will all have to start buying AI tokens/credits both for personal use and through work. It’s going to be a bit like paying for gas, electricity and broadband. Finally the internet will get monetised. Effectively, the era of free services will end as the AI charging model is so simple to implement. Once the world has got to grips with AI services, they won’t be able to live (or compete) without them.
An exceptionally fine Friday Moguls’ piece @TI. Whilst James Emanuel’s Rock & Turner missive (as you linked to) partly reflects my concerns*, this one from 19:22 yesterday (22/2/26) does the heavy lifting for me:
https://open.substack.com/pub/citrini/p/2028gic
It’s not quite ‘AI 2027’ (it’s 2028!), nor ‘Situational Awareness’; but, differences of timescales aside, which informed observer isn’t worried where this all leads?
FWIW (i.e. almost nothing here, if not literally nada); if a gun were at my head (rather than a ML agent pointed at my trad ‘knowledge economy’/PMC, previously safe as houses, career), and if I had no choice but to go ‘all in’ (i.e. TINA time, which is never a good idea IMHO, as Ed Thorpe proved) on the atoms dominating parts of the ‘tech stack’; then it’d be:
15% Alphabet (TPUs, Google Cloud, search/ads, Chrome, YT, Weymo/LiDAR/FSD)
15% Microsoft (Azure, Copilot etc)
10% Nvidia (Rubin GPUs & CUDA)
10% Amazon (AWS)
50% SPX/Nasdaq and/or ASML, TSMC, AMD, ARM, Cadence, Teledyne, Vertiv, Applied Materials, Arista Networks, Broadcom, Qualcomm, CEG etc
Which looks nothing (whatsoever) like my current US large cap lite (& TF, gen ex US, Euro, Jap overweight; & SCV, commodo, infra, miners/PM miners, EM/Frontier Mkt heavy, plus some bonds) current portfolio.
[*For @Ian #1: as a fellow appreciator of David Kipping’s Cool Worlds’ podcast, I’ve included an AI/Fermi crossover themed link for you at the (current) end of the (now v long) May 2024 “First they came for the Call Centres” comment thread.]
VG, thanks. What a confusing time! I understand fully the feeling to be more diversified than usual, but am curious as to how many individual stocks you hold currently and e.g. what % is in top 10, if you’re happy to say. Thanks.
I have been trying to expand the old ‘circle of competence’ into Commodity Royalty Cos, and Consumer stuff (e.g Hermes) and even a few Industrials – feels like the days of the Fundsmith only style are gone for the moment.
Separately, I keep thinking things can’t get any worse for LTI, and then… what timing to switch more from Consumer names to Software stuff. Not saying he’s wrong, but hasn’t helped performance. I am perhaps stupidly holding as a) feels like can’t get much worse, esp. now that Mgmt Co is less % than it was, b) stock holdings at max pessimism c) discount d) divi (whatever it is this year) and e) liquidity (the share split hasn’t done much, has it?). Sorry to go longer off-topic than I meant to!
@Tom Grlla — Thanks for your thoughts. For number of stocks, I’ll just say “more than 20” and note this doesn’t count a very chunky clutch of investment trusts and a few REITs etc, which obviously increases the diversification meaningfully. (I tend to treat some of those ITs almost like operating companies that happen to invest though, especially LTI).
As ever I am very reluctant to go into granular detail about my portfolio, not so much for reasons of privacy or even regulatory concerns (I don’t want anyone explicitly following what I *do*, I want people reflecting and hopefully learning from what I *say*) but more for reasons of it not being helpful to members in the long run.
For example I trade far more than I should or that most active investors of my ilk would. This significantly alters the outcomes — for good or ill — of my choices, and is not reflected in point-in-time snapshots of my portfolio. For instance I too have a holding of LTI, which as of this moment is about 4% of my very active portfolio (I have a big more boring bucket walled off for mortgage repayment purposes) which is sporting a 12% loss.
Okay, that sounds bad (and it is!) but as of this instant it’s both worse than it looks and probably not so bad as all that. Why? Because at some point in the past few months I had well over 10% in my portfolio! However I traded around this position, selling some near £7 (I suspect to Michael Lindsell) and buying *at best* nearer to £6 (most much higher obviously). I’ve lost more in total than the paper loss the position is currently showing, but my total % gain/loss on LTI shenanigans vs money deployed is also probably nothing like so bad as 12% (though I’ve not worked it out to be honest).
What’s more, the money I realised reasonably recently went into LSEG and RELX around their lows, from which they’ve bounced.
Take all this churn into account and I don’t think telling you I have X% in AI-disrupted stocks but Y% in investment trusts informs members, but rather obfuscates.
Hence with Moguls I am more trying to honestly set out how I see the investing landscape right now, occasional opportunities (e.g. UK stocks over the past couple of years) and methods for running an active portfolio (and why you might) in a landscape with the ruthlessly efficient and effective global tracker as an easy alternative. The idea is to be a fellow practitioner not a role model as such, hence my reluctance to say much about the granularity of the portfolio. I don’t want to mislead.
(I’m already regretting the above comments! LTI could be doubled down if it stays below £6 or gone by next week if something else comes up…)
Re: LTI, yes the switch looks almost cruel. Nick Train to me is starting to seem almost like a Greek mythical figure cursed by the gods. To go from a near-100% premium to languish down here around a 20% discount, while your portfolio melts in front of you and the market turns on a dime.
LTL still has that big gob of cash, but I have almost given up on the board doing anything with it. As I told the chair, they should have acted when the premium was crazy high by issuing stock, and now they’d have more freedom and flexibility to buy back shares on a discount.
As you know the real problem with LTL is the evaporating FUM. You almost have to treat LTL as having zero value but the cash, discount it, and then apply a standard UK equity income or global income discount to the rest of the portfolio. Although even that is a bit generous given his stock picking performance of late.
Still, at least Diageo has bounced…
Is the antidote to the New New (agentic disruption of enterprise SaaS) the New economy (i.e. pre the (supposed) emergent Intelligence As A Service era, but post software) or the Old one (i.e. outright pre digital enterprises, or ones involving atoms over bits)?
Constellation Software is a halver now, but that only works out if there’s still a business model and a TAM. IJDK if that’s still the case. Contrastingly, Old seems to have fared better than New in dealing with New New in the payments space, to judge by PayPal versus Western Union in the face of Apple and Google Pay and whatnot. Perhaps if you’ve survived disruption once then you’ll probably survive it again..?
That suggests (again, perhaps) that the likes of LTI (although it’s a special case because of the ‘hidden value’ in the investment management business) might be a better punt than, say, Adobe or Salesforce.
The footnotes to that though are a). that, judging by investing Substack, hardcore deep value investors who would have run a mile away from PayPal at $300 were seriously looking at it as a run off proposition at the dip below $40 (every investor has his or her price), and b). you are getting 7% divis with LTI (and 10% with WU, and it’s an arguably value accretive cannibal buying back it’s shares).
Obviously, in aggregate, there is no reward for the idiosyncratic risk of single name stock selection (Bessembinder et al, lest we need reminding again). But there is the fun of chasing narratives, and if you keep it to <5% then what harm can it do? 😉
Given the mention of Duolingo above, and the discussion here on both the merits of active stock selection and individual shares ownership:
https://monevator.com/active-investors-are-engaged-investors/
Duolingo’s performance today is a reminder why, in aggregate, for the median investor with no discernable edge, stock picking probably will underperform broad index tracking approaches.
Related to the above mentions of LTI and LSE, IC today covering both in some detail. I think their views on LTI are too pessimistic though and may mark a sentiment bottom, with the IT trading below £6 now, compared with a £16ish ATH. Your getting a 7% yield on a 20% discount to NAV to wait for a turnaround. Of course, maybe AI has changed the whole game forever (very plausible IMHO); but, if you don’t buy that (and who’s to know, one way or the other), the Nick Train approach is one possible alternative.
It’s pretty sobering though that way back on the 13th October 2012 @TI/Monevator was reporting on how Nick Train actually wanted the premium to NAV to come down on LTI!
How times have changed. Or is it that Nick hasn’t changed with the times perhaps…..
@DH — DuoLingo’s performance today doesn’t tell us anything about whether the average investor will be able to beat the market (unlike the copious research that *does* tell and show us that).
Highly-rated stocks priced on cashflows discounted far into the future are extremely volatile. One might as well have posted about DuoLingo when it had advanced from $80 to over $500 in 2.5 years and said “this shows why picking stocks in the way to go” 😉
Moreover holdings don’t exist as points in time. You have infinite points of entry and exit, and many active investors, myself included, add and reduce positions very regularly. (I have done over 1,000 trades some years…)
This is a bugbear of mine with the often-quoted Bessembinder research, or at least the way it is widely-cited. Nobody is buying buggy whip IPOs then holding them for 100 years until they are superseded by EV manufacturers (to attenuate the point).
Volatile share prices and market over/under reactions is exactly why *some* people can make money from active investing. Most don’t, and even more shouldn’t try. 🙂
Re: Nick Train, yes he said the premium was way too high, but as I’ve noted to the LTI chair they did nothing about it. They should have issued tons of shares to bring that premium down, it’s not like most of their stocks aren’t super-liquid (the LTL holding aside). Now they are doing little while on a big discount, and they can’t really because the trust is so small and the LTL holding even after massive shrinkage is still so dominant.
Personally I think the way LTI has been managed from a corporate structure/IT perspective (i.e. portfolio picks etc aside) has been pretty woeful both when it was doing well and now it isn’t. Ho hum.
@TI #10: re: “copious research”: Going beyond Bessembinder, Sandvik’s and Hylin’s “Do All Stocks Fail to Outperform
Treasury Bills?” (Norwegian School of Economics Fall 2018): “we investigate the distribution of individual stock returns in United Kingdom, Japan, Germany, France, Italy and Sweden from 1986 to 2017. Specifically, our results highlight the strong presence of positive skewness in the return distributions. Consequently, the majority of stocks fail to generate buy-and-hold returns superior to the matching one-month Treasury bills over their lifetime (or sample period). The only exceptions are Japan and France, where slightly more than half of the stocks yield positive excess return. Measured in wealth creation, only a fraction of companies constitute the total net wealth created in the market. The numbers range from 0.5% in Italy to 10.9% in Sweden, whereas the remaining stocks in aggregate have produced returns equal to the Treasury bills. Thus, it is evident that stock markets are highly concentrated, where contributions from the minority of stocks more than make up for the poor performance by the majority. The results provide evidence to why most undiversified funds underperform against market-wide benchmark portfolios.”
On “infinite points of entry and exit”: do you use a system on your active holdings which involves avoiding both of the market timing and the doubling down into losers risks (‘losers average losers,’ as PTJ put it)?: e.g. using DCA on buys and calendar or % tolerance rebalancing for sells?
How do you know whether to maintain your conviction or to exit on a thesis invalidation? Do you have a checklist?
For every thesis there seems to be an anti thesis. For every pundit, an anti pundit. For every expert, an equal and opposite expert.
Although some stocks are especially polarising, e.g. Tesla and Google (a real world Tyrell Corporation and Weyland-Yutani Corporation?); and Palantir (a real world Cyberdyne Systems?), all stocks have their cheerleaders and naysayers.
While ‘The Dutch Investors’* (and others) favour today’s sharp faller Duolingo, ‘The Pragmatic Investor’ does not:
https://open.substack.com/pub/jamesfoord/p/win-some-crcl-lose-some-eose-avoid
Both present reasoned arguments and persuasive narratives.
How, therefore, can one have the necessary ex ante confidence to take meaningful position sizing?
*TDI piece on DUOL from June 2025:
https://open.substack.com/pub/thedutchinvestors/p/duolingo-vs-spotify-whats-the-better
@Delta Hedge — I didn’t say there was copious research about how stock prices move, although I am sure there is. 🙂 I alluded to the copious research that most active investors fail to beat the market over the long-term.
If an investor has edge then it doesn’t matter two hoots what stocks do as a group, all that matters is that the investor has some ability to select some subset of them to long/short, and over particular time frames. The volatility of the stocks provides the opportunity, it’s a feature for active investors not a bug.
Again, most active investors have no edge so it’s moot. As we’ve been writing for nearly two decades they should be passive, at least until that too blows up… 😉 (Green et al)
p.s. Rereading this comment it sounds a bit more combative than I intended. Blame a long week, and a desire to finish my Weekend Reading links. 😉 But I stand by the points I’m making.
As for your wider questions, it’s completely unfeasible to even begin to reply to them in a blog comment. (In some ways all my articles on Monevator are my response 😉 )
Thank you for all the hard work on the blog, inc. the w/e links. I didn’t read it as combative 😉
I have to ask the questions though, even just to myself. “My basic approach is to figure out what’s going on by asking a lot of questions”, per David Einhorn.
One underrated, or at least under repeated, observation (which might be useful to bear in mind in this the fourth major revolution in the means of production in the last few hundred years (e.g. agricultural -> industrial automation/ energy -> information/ communication/ software -> artificial intelligence/ high performance compute)) is that asset classes can’t go bust. An enterprise can disappear. But they’ll always be an energy sector, a need for metals, or for a tech sector.
Re @TI #10: “I have done over 1,000 trades some years…”‘: sounds like the approach of Mark Ellis’ Nutshell Growth OEIC:
https://www.trustnet.com/News/13460844/the-top-performing-global-fund-with-an-eye-popping-700-turnover/
“How high is your turnover? Very high. In the past 12 months it’s north of 700%. When I tell that to a traditional fund buyer, their eyes pop out….[But] our commission rates are virtually zero. Over the past 12 months, our total trading costs were 20 basis points and we added 5.9% [CAGR] through our trades.”
@DH — Yes, I think a lot of the thinking/literature about trading is out of date, especially in the US where you don’t have stamp duty either. Not to say it isn’t still a big drag on potential returns at these levels but the behavioural/misjudgement risks loom much larger for an active investor.
On the one hand, without getting too existential here, there’s the aspect of wanting a simple system to deliver lower but acceptable returns with less risk (the AWP, the Permanent Portfolio, risk parity, the Golden Butterfly portfolio), or just the market return (cap weight global index tracking B&H) or potentially much more return for much more risk (the Triple Accelerator idea I DM’d you the other day).
On the other hand, the market is unknowably dynamical. Sorosian reflexivity rules. To quote a couple of ‘Herbertisms’:
“Deep in the human unconscious is a pervasive need for a logical universe that makes sense. But the real universe is always one step beyond logic.”
“A process cannot be understood by stopping it. Understanding must move with the flow of the process, must join it and flow with it.”
Trading a lot at least acknowledges that truly we don’t know anything about what comes next, that the past is not prologue, and all our investing ideas should probably be lightly held. Still, when markets drop expected returns rise.
As a p.s. to #10, 16 and 17 above, this piece on Friday usefully summarises some research around high turnover trading:
https://open.substack.com/pub/danielsdeepdive1/p/the-most-expensive-mistake-investors
I thought about posting this in the ‘Active Investors are engaged investors’ thread, but this is a slightly different meaning of active investor! 😉
@Delta Hedge — Well I’m a stuck record, but again if an investor doesn’t have edge then more trading will amplify the impact of their negative alpha, compounding it ever more quickly on top of the (albeit now smaller) negative impact of fees.
Worth also noting that US talk about taxes in these sorts of articles isn’t as relevant here. E.g. Investors in active fund managers there have to declare realised capital gains in un-tax-sheltered accounts, whereas ours don’t. (This has been a big driver of ETFs in the US, which are taxed differently.) On a practical level most UK stock picking investors will mostly be operating within ISAs and SIPPs, too.
Finally, with respect to my own trading, as I’ve said many times on Monevator it’s most often in the same relatively small universe of names. I am not buying a Norwegian fishing company one day then selling it two days later to buy an AIM copper miner. Hence while ‘1000 trades a year’ might sound like a lot, but if half of those are incrementally adding and reducing the same set of positions perhaps a bit less so. 😉 Divide by 250 for a daily tally and it’s even less dramatic than it sounds.
But again, I’m not trying to convince anyone to go this way. It’s stressful and I probably trade too much for my own mental health to be honest, hence it’s been steadily coming down. (More background here: https://monevator.com/days-of-being-wild-part-one/)
Just to add a final point from me on the turnover question, I was listening to this interview with Sean Peche of Ranmore where he pretty much echoes my thinking:
https://www.investorschronicle.co.uk/content/44b58e4e-7c55-44e9-9e91-eeb2a4b552df
@TI #21: IIRC in a previous thread you answered that you were constitutionally disinclined to systemic/quantitative strategies.
But would you ever reconsider?
I’ve read more dynamical asset allocations (and, indeed, fixed % optimisations variants of the AWP and the PP) than you could shake a stick at 😉
For me the problem with active single name stock selection and trading (even in and out of the same names) is the base odds are stacked against you and (given the inherent asymmetry of needing 100% gain to recover a 50% loss, and a 400% one to get back to even on an 80% one) mistakes compound badly.
At least with B&H it’s closer to being the House than playing in the Casino, given that, in the S&P 500, 54% of days are green, 75% of years, 95% of decades and 100% of 20 year or longer periods.
Unlike a casino or trading with no edge, an edgeless investor has more chance of being ahead with B&H the longer that they persist.
And, back test overfitting / ‘p hacking’ concerns acknowledged here, B&H returns pale against the historical returns of many dynamical asset allocations.
@Delta Hedge — I just wrote you a huge reply, but pressed ‘Cancel’ instead of ‘Reply’ FFS. 🙁 I can’t face typing it again! Maybe another time or in a post.
For now I’ll just summarise as “It’s not cope (as the kids put it) to say I love the challenge and the game and the interaction with the real world of stock picking and that’s a large part of why I do what I do, whereas those other strategies wouldn’t deliver that aspect for me, plus I see no reason why I’d have edge versus PhDs and unlimited compute”. Cheers!
That’s horribly frustrating when that particular FUBAR happens. I’ve done similar with work emails a few times, several being drafted and open at once, then accidentally closed one or more without saving. Maybe OpenClaw (or its successor agent) will save the day in the near future 😉
When you’d first mentioned doing up to 1,000 trades a year (so about two roundtrips per trading day) I kinda had in mind that maybe you were looking to emulate this trading legend.
https://open.substack.com/pub/paretoinvestor/p/bnf-trading-system-ai-prompts
But it seems a little less ambitious than that – i.e. more Nutshell fund than BNF (or those Turtle traders that Denis and Eckhardt funded in the 1980s)
Tbh I find the idea of stock picking too exhausting for more than token single name exposure using bottom up fundamentals. (Following Metcalf’s Law) just 100 available stocks (say the Footsie, or the NASDAQ 100) is already a network of 4,950 unique cross relationships, with a network size of 10,000, with the factorial of 100 (i.e. 100!, or 9 followed by 157 digits) of unique stock combinations, and (following Reed’s law) about 2 raised to the power of 100 different sub groups of possible stocks (itself a number with 30 digits). It’s incomputable, never mind being utterly incomprehensible in it’s combinatorial complexity! Consequently, I now spend more time on the sizing (the K factor*) of the asset class (choice of instrument) than on the security selection process per se.
* i.e. Kelly Criterion all the way down (per this one from today):
https://open.substack.com/pub/jimmysjournal/p/how-to-size-positions-correctly