Weekly Report: March 1, 2026

#148 AI SaaSpocalypse, Sydney, private credit bubble

Stuart Loh

Mar 02, 2026

Observations

Sydney

We visited Sydney last week during Chinese New Year. One thing that struck me was the amount of infrastructure development occurring. Sydney now has a metro system that everyone raves about (in addition to light rail, and CityRail).

Sydney also has a sprawling network of car tunnels under the city, with the longest one being a whopping 22km long (WestConnex)! We drove through a large portion of it and it’s a bit unnerving traveling underground for that long. On our last trip, we missed our exit and ended up halfway across the city in the wrong direction because Google Maps stopped working in the tunnels (no GPS). This year, they apparently had fixed that problem by installing thousands of Bluetooth beacons in the tunnel that signal positioning to phones.
I contrast that to the Bay Area, where by comparison there is almost zero infrastructure development. It’s sad.
Robotaxis don’t appear to have entered the public’s consciousness there yet.
We had to drive a lot in Sydney. I packed a maths book in my daughter’s backpack and on a one hour car ride she, entirely of her own volition, pulled it out and spent the whole ride answering maths questions. And then when we arrived at our destination, she voluntarily sat down at a table for another hour and finished the rest of the book off. 🤯

AI SaaSpocalypse

The extreme outcomes of AI, and projecting AI trends out to infinity, are not scenarios that require a great deal of imagination to imagine. The idea of the technological singularity has been around since the 1950s and is the domain of apocalyptic sci-fi. But there is an intermediate extreme where, in this decade, AI does not exterminate humanity, but nonetheless causes massive economic and societal upheaval.
Many articles have written about the dire impact AI will have in coming years, such as Andrew Yang’s The End of the Office. But this week’s Substack post from Citrini, The 2028 Global Intelligence Crisis, has literally moved markets, accelerating a decline in enterprise SaaS company valuations.
Citrini’s post — “a scenario, not a prediction” — is an eloquent piece of “what if” science fiction that has sparked vigorous pushback. This is not surprising because extremes are easy to pick apart as they are to imagine.
AI will not be the death of software companies or software engineers any time soon. Maybe tech companies will shed staff by the thousands, as Block did this week, because AI means it takes less engineers to do the work. Maybe Block simply overhired (apparently only a quarter of the people let go were from the engineering team, even though AI uptake is currently the strongest in that function). Remember that Twitter slashed far more than half of its workforce without the service falling over when Musk acquired it, and that had nothing to do with AI. The truth is probably a mix of these reasons. This trend may continue in tech companies, but I doubt it will lead to massive structural problems, for Silicon Valley is not main street.
Block is also a public company, and public companies are motivated by their stock price. Whatever valuation metric is in vogue with the market is what they optimize for. Sometimes it’s revenue growth. Sometimes it’s profitability and cash flow. Now the pendulum seems to be swinging towards efficiency, with shareholders cheering RIFs as they go straight to the bottom line. Block’s stock price rose over 20% on the news.
(Interestingly, for most of the almost-two decades that I have worked in tech startups, headcount has been used as one indicator of success for private companies, whereas P&L numbers are closely guarded. What CEO doesn’t want to appear like they’re managing a larger and growing team than a smaller and static one? That longstanding tendency now appears to be reversing, which I think is a good sign. I have always been wary about conflating certain metrics like headcount with success.)
I predict the pendulum will swing the other way again. After all, if one engineer with AI can now do the work of five engineers, then as soon as you figure things out, why wouldn’t you hire back four engineers so they can now do the work of 25 pre-AI engineers? Visions of technology leading to a 10 hour work week have never come to fruition because productivity normally expands to fill the void. The feature request pipeline is never ending.
Second, companies moved from on-premises software to SaaS at the turn of the millennium because they no longer had to deal with all the hassle of managing that software — maintaining the infra it ran on, keeping the software updated, and so on. What company then, especially a non-tech one, really wants to rip out their SaaS vendors and write their own software? It’s not just that the switching costs are prohibitive. It’s that running a SaaS business is much more than simply writing code: those businesses take care of everything else — infrastructure, DevOps, technical support, security, reliability, software maintenance, feature development — and are experts at doing that because that’s all they do.
For a company to roll its own software, they still have to hire technical staff to do that. Do you really want non-engineers to try vibe coding their own production software to do anything really serious? It’s one thing for a finance person to use AI to help them analyze financial statements or build spreadsheets — it’s another thing altogether to build and maintain something to replace NetSuite, or even QuickBooks.
I do foresee a cottage industry developing for independent software engineers that will use AI to build and support small, tightly scoped internal “homegrown apps” for business customers.
To be sure, it is true that if you project AI out to infinity, you could have AI agents taking care of all of those ancillary tasks. It’s not far-fetched to believe that we will eventually have agents that are able to buy and spin up cloud infra autonomously, monitor server status, address downtime issues, and manage all those costs against a budget. If you project AI out to infinity, AI can basically run most functions in a software business. However, that more likely means that SaaS companies will get more efficient — not that they will be supplanted by their customers rolling their own software.
What is a worry is how this will exacerbate inequality. As assets become more productive, asset owners will reap an increasing share of that productivity, to the expense of non-owner wage-earners. The Gini coeffieicnt for wealth in the U.S. is purportedly around 0.83 — similar levels to France just before their revolution. Widening, unchecked inequality eventually leads to instability, which history has shown will either lead to revolution or metastasized oppression.
Ultimately no one knows how any of this is going to play out, and the narrative whipsaws between dreams of avarice and nightmares of catastrophe. The lack of certainty means that investing has become increasingly driven by emotion — amplified by the casino economy, retail investors, momentum trading, fear and FOMO — and that causes volatility that we shouldn’t read into as being indicative of how things will play out in the months and years to come. In the word of Benjamin Graham, in the short term the market is a voting machine.

Further Observations

Hegseth has had quite a week hasn’t he? He’s gone after Anthropic, the Ivy League, Boy Scouts, and Iran.
After Paramount sweetened its offer, Netflix has bowed out of the fight to acquire Warner Bros. Discovery. Incidentally, unless Paramount later decides to spin out WBD’s cable assets, CNN as we know it is dead.
In other big news, in this big news week: SCOTUS invalidated POTUS’ tariffs (which were then immediately reinstated under a different power). POTUS gave the longest ever State of the Union. Northeastern parts of the country were buried in snow.

Articles

Statement from Dario Amodei on our discussions with the Department of War and Statement on the comments from Secretary of War Pete Hegseth (Anthropic)
The Secretary’s extreme actions in response to its dealings with Anthropic is simply fury at being defied. They don’t make sense because angry people act irrationally. The banning of Anthropic is nothing more than “if you don’t do exactly what I tell you to, I will fuck you up, you whore” energy. Then this happened.
Marked to Fantasy and its companion site (Mispriced Assets)
Private credit is a bubble.
People Loved the Dot-Com Boom. The A.I. Boom, Not So Much (New York Times)
The 10am Drop: How Jane Street Broke Bitcoin Price (CryptoZeno on Binance Square)

Diversions

Reviews

🎬 Marty Supreme (2025)
If you like movies where things keep going wrong for the protagonist, this is a good one. I find them a little stressful, but I found it entertaining. ★★★
🎬 Don’t Look Up (2021)
I liked this movie but ultimately depressing. In the vein of shows like Idiocracy and Silicon Valley, Don’t Look Up is much closer to reality than is comfortable. ★★★★
📺 A Knight of the Seven Kingdoms (Season 1)
Thoroughly entertaining. TV shows in Westeros don’t need to be big budget to be great. ★★★★★
📖 The Real North Korea: Life and Politics in the Failed Stalinist Utopia (Andrei Lankov, 2014)
Some interesting insights, but a little dryly written, especially given that the author spent a good deal of time living in North Korea. ★★★