First, a bit from Azeem Azhar , who is a font of useful, even-handed smarts about tech and all its tentacles.

My AI tools, the invaluable GAI Insights (see their three recent foundational important GenAI HBR articles via John Sviokla ), and Azeem help me stay ahead of the fast-moving AI curve...and my LinkedIn network pals who often connect me to interesting stuff.

"By 2027, a single artificial intelligence model could cost $100 billion to train – a figure bigger than the GDP of two-thirds of the world’s countries. This staggering projection illustrates the breakneck speed at which AI development is scaling up, both in size and cost."

(from Azeem's newsletter)

The scale does matter. Huge knowledge bases have obvious marvelous outcomes.

But the next five to ten years will be about Intelligence married to Agility and stealing a core concept from comedy, Timing.

The Achilles Heal will be tied to a step function improvement in hacking/cyber threats for two reasons:

• The Dark Web is adopting modern tech, including GenAI, smart architecture, and leading-edge service models, as fast as or faster than anyone else. It is starting to look like a criminal B2B mall with great products, services, construction concierge services, and attractive pricing.

• As the leader of the Newton gang (which robbed 80 banks and 6 trains from 1919 to 1924) said to the judge who asked him why he robbed banks, "That's where the money is." Their last train robbery netted $3M, which in today's dollars is about $57M!

Value and, therefore, dollar impact will come from applied knowledge (intelligence) done inside your competitor's operational cycle or just inside your current performance cycles (continuous improvement) or current consumer expectations at the exact time when that matters.

At FedEx Supply Chain back in the 90s, we figured out delivery wasn't really about being the fastest. It was about getting there at the exact time it was needed/expected. Of course, doing that with a 99.999% success rate helped. That, along with being the fastest/most responsive, was a killer combo.

Also, no one will care if your LLM has 50 billion more tokens than someone else (for several reasons).

The there reasons LLMs will hit a ceiling are:

- The astronomical cost...not that these are expensive, they are bonkers expensive ...for $100B, I can build the next-gen big LLM or acquire the Gap plus Kroger plus HP plus Southwest Airlines. I guess I could sell low-cost travel to a place where you can get decent groceries, very American clothes and a printer designed to addict me to their ink.

- Diminishing marginal returns (thanks Leon Walras 'Elements of Pure Economics' 1874)

- Historically Wave One paradigm shifting technologies are marvelous. But technology, materials, and science move forward, and some smart folk look back at Wave One and ask what we would do differently with today's gear and know-how. They generally obliterate Wave One with this new Wave Two. Sometimes it's even the exact same people...cheeky genius Aneel Bhusri and co literally did that from PeopleSoft to Workday. Workday = $60B market cap or 2/3's of a big next-gen LLM! Who remembers Wave One Internet search firms? This is starting to happen in the world of EVs, well underway for satellites/space stuff, and pretty soon, entertainment and, I hope, healthcare.

The entire history of complexity is characterized by four things:

• Humans have an innate capability in part due to the 85 billion neurons and trillions of network connection permutations we have in our heads to generate complexity.

• It is one of our best 'skills'. About 99% of the time, new things arise and gradually become terribly complex. Urbanization, the law, the winter NJ transit schedule, Phenomenology, Taylor Swift concerts, French politics, commerce, Charlie Parker's Anthropology, Kubrik's approach to movies, multi-layer supply networks, computational neurobiology (Salk Institute), the later versions of Catan, quantum photonics, and, of course, taxes are examples.

• Toby's First Law of Complexity: Complexity Breeds Curation - going all the way back to markets at the crossroads growing into towns or then Belgium. Also, think of all forms of brokers, layered supply chains, Amazon, entertainment, healthcare, the Dutch in the 16th/17th century, that internet thing, and GenAI...yes human generated complexity. Travel Agents are a great example of curating complexity. In 2000, in polite London, I gave a speech to the 3000 folks at the IH&RA and asked all the travel agents to raise their hands. I told them 90% were going out of biz that decade. I got booed loudly. So, travel complexity was curated by agents who were then curated by that internet thing. New tech often affords new plateaus of innovation. Travel agents are making a comeback in niches in curation/concierge ways.

• Toby's Second Law of Complexity states that complexity and cost eventually get designed out with model, mechanism, and design-level innovation. That often goes hand in hand with Lego-ization. Does anyone remember tech before middleware or app servers or the auto industry before the production line? Or what coding and debugging used to be? AI used to have no Lego, and you would need PhD smarties to do a lot of things. 80% of the hard stuff turned into Lego. Look at Amazon SageMaker for one example where complex AI is now easily composable Lego. Unfortunately, the Dark Web has gone down this path. In October 2015, TalkTalk, a UK Mobile Telco, was hacked, severely damaged, and acquired for a song. A 15-year-old with a little computer experience assembled rented bits of Lego on the Dark Web. The Dark Web Lego came with simple online instructions.

By 2027, GenAI base platforms/back office 'stuff' will be 200x more efficient, and by 2030, it will be 1000x.

Wave One Wreckage

There will be a lot of Wave One wreckage. I have said variations of these two maxims a lot, but it is worth repeating:

• In a Schumpeterian Creative Destruction Cycle (like the one we are entering now), very few, especially top performers, think they will be on the destruction side

• The Board of Directors - if you aren't having a strategy refresh discussion, you are not paying attention. The tectonic shifts you feel now are just the tip of the iceberg. No amount of mixed metaphors will save you from inaction. The odds that your existing strategy and its cascade into mission, structure, talent, and culture (incentives/disincentives) are future-ready is near zero.

The innovation is often at the model/design level or software versus hardware. Ocient burned through 250 person-years of top engineering talent and created a massive exabyte-capable big data platform that's a 1000x leap forward (cost/performance). These kinds of leaps don't just improve unit costs. They open up completely new plateaus of opportunity.

They don't all have to be super sexy whiz-bang things. Tea bags were invented by an Isle of White company. By the end of the 1800s, they were the biggest tea company on earth. Corrugated metal bottle caps (1892) are another one. The same guy (William Painter) also invented an ejection safety seat for trains - that never caught on.

Look at VANTIQ for intelligence at the exact time you need it, which is both an output and a sensing/real-time awareness problem.

Arqit has a very different model for hyper-secure, quantum-proof, NSA-approved standard encryption. It would have saved TalkTalk or ATT last week.

The Evidence and the Worry

I read a little management book from Andy Grove when I was 22 that changed my views on leadership. Positive paranoic worrying had its useful place.

The S&P 500 was born in 1957. Around 80% of those listed are no longer there. My best estimate is that the churn rate in its first decade has grown by 600% since then, accelerating in the last 20 years. It was always easy to start a company. This year, over a million new companies will implode in the US alone.

It would be best if you didn't worry about that. Worry about the growth in new co-launches, which was five million in the US last year alone, and the odd one or two from there that scales and comes after you.

Also, worry about boards that effectively pursue a strategy refresh before you do.

What you should worry about is that it has become massively easier to launch a company AND to scale it.

It used to take much work to scale and was truly difficult. Talent, capital, manufacturing, marketing, sales and logistics were all difficult. They are 20x easier today. Look at Baidu and Prologis, who can solve a lot of that. Yesterday I spoke with a genius team covering product market fit, market research, tech including solutions architecture and AI ready to be variable resources. Can't tell you where, you must find your own, but it used to be the Dacian Kingdom.

There is always funding, and there has been more lately (except for diverse founders, which is statistically clearly a domain where we should be the opposite of proud). The statistics are almost unbelievable, except if you have been inside those sorts of raises, then it's pretty apparent.

These days, if I pitched a GenAI-anchored combined frozen yogurt, nail salon, petting zoo, and auto repair parlor and used my US or British citizenship and not my Mexican one, I could get funding as long as I had GenAI somewhere in each slide. Our AI would find the white spaces and cross-sector synergies and be the first to hedge revenue fluctuations/output capability against weather patterns. The wallet share synergy of what would be a four-sided AI-infused network model would be compelling. We would, of course, be super sustainably focused. If I could get a top engineer out of OpenAI or Google or Mistral, the market cap would double. I jest, but you get the point.

Thank you both for reading this far.

(edited slightly Q4 2024)