What's funny is when the book was published, I remember someone telling me that I should have included more examples of failed startups in the manuscript. I remember answering them, "Oh, I have. We just don't know which ones yet."
This is the difficulty of writing any kind of business book. There's just no way to use any company as an example to illustrate some principle without people misunderstanding that you're holding them up as the perfect or even great company. I use case studies to illustrate the concepts that I think are useful. I can't guarantee success any more than an athlete can tell you that if you study the way that Tony Gwynn hits the baseball, you too will be able to hit 300 in the big leagues.
That's fair but I felt you held up these companies and founders as these demigods running this well oiled operation and yet I remember looking up those companies and all quietly folded within a few years of your book coming out.
My point is you were not able to demonstrate any correlation with your suggested methods with startup success. A company could do the exact opposite of what you recommend and be successful. Or follow it to a tee and fail which is what all of them did (happy to hear about any counter examples here from the original book). So what exactly is the point if you can't even move the needle a little bit?
Academics have spent the last 15 years attempting to measure these effects, and it's just a very hard measurement problem. There are so many confounding variables. I would say the early evidence is supportive of the thesis, broadly speaking. In the long run, I expect we will eventually be able to speak about this in a much more nuanced way. For example, some of the elements of LS are much easier to test than others, and those have been found to be broadly effective.
This project may take decades, and yet leaders need guidance and useful mental models in the meantime. So, to me, the much more useful data is the literally thousands of founders (and other leaders) who have reported that they have found the framework useful in their own lives and companies. They may all be deluded, of course, but at some point I think we have to acknowledge that _something_ useful is going on here.
In any event, I don't give it that much thought. If a new theory emerges that proves more useful, or if new research gives entrepreneurs new and better tools, I'd be delighted. My only loyalty is to the truth.
You should, however, ( assuming your ideas are correct ) be able to look at a company and tell if it is following your ideas or not and will therefore ( assuming your ideas are correct ) be successful or not, and therefore make billions investing.
The bitter lesson is that there is no domain that will be left which AI won't get good at eventually. So really you have two options: if you actually believe the timeline is long you can keep retreating to the sectors that will be taken over last (emotional support nurse etc) or you can just say if you can't beat me join em and try to supercharge your career/project/life with AI now so it improving helps you rather than hurts you.
The supercharge bit missed an important fact: that strategy is very temporary. Getting expertise in software development takes a long time. Getting expertise in these LLM tools takes a lot less time— the combination of LLM expertise and dev expertise is the useful part. If LLMs make working developers, say, 35% more efficient, that’s going to be many thousands of people out of work, many of them being the most experienced and expensive we have. It’s not like those people are all going to give up immediately and become DoorDash drivers — they’re going to fight tooth and nail to get a job that uses their existing hard-won expertise. That means they’re going to level up their LLM knowledge, be willing to work for a LOT less money, and bring down everybody’s wages in the process. Companies don’t pay people based on the amount they bring to the company — they pay people based on the going market rate. That’s about to be a whole lot lower. So no matter how much you supercharge, you’re only buying yourself a little while until the labor market catches up. Nobody in development is safe. The entire field was so busy seeing how fast they could saw branches off of a tree that they didn’t realize they were standing on the wrong side of the cut, and the business side of the industry could not be happier about it. You’re basically working as a manufacturing engineer in the 90s US specializing in moving processes to offshore facilities. Probably felt pretty clever for a few years until they got the pink slip.
Honestly, the only hope that the dev field has is this all being so economically inefficient that the industry as we know it collapses after the VC subsidies run out, and we’re going to pivot towards much more reasonable interventions with local models and such.
you're being a doomer and only looking at how AI will take your job or suppress your wage. try to flip it around and see how you can do more. layoffs are coming and I have said as much in other posts but the top AI engineers will make a lot more money. ultimately they will get replaced to but in that scenario we should have reached AGI abundance.
I’m in a trade now. I’ll be fine. I’ve got a pension.
And those AI engineers won’t be making bank for very long.
Looking at basic economics— labor supply and demand within this field— is not being a ‘doomer’, it’s just not letting boundless optimism trump rational analysis. It’s not like this is the first profession to get sucked into the air intake of the machine that replaced them.
Switching from software dev to AI engineer is an achievable lateral move, and we will have a lot of surplus software devs. Few of them will have skills that will transfer cleanly enough to keep the mortgage paid. Guess what job they’ll try to move into? Unless this turns out fundamentally different from what the big companies are implying, we’re not going to see demand that isn’t wholly satisfied by the current workforce for decades. With that much expertise desperate for a job, there’s no reasonable path to keeping wages high.
Sure there might be a few tech PhDs + MBAs at the director level that are making bank, but we’re building a system that replaces us, and the industry is not going to magically stop at this arbitrary level of replacement.
Milton Friedman was once visiting China when he was shocked to see that, instead of modern tractors, thousands of workers were toiling away building a canal with shovels. He asked his host, a government bureaucrat, why more machines weren’t being used. The bureaucrat replied, “You don’t understand. This is a jobs program.” To which Milton responded, “Oh, I thought you were trying to build a canal. If it’s jobs you want, you should give these workers spoons, not shovels!”
Yup I was going to comment that's the closest analogy to tulips. You might hate AI but at least that's one thing you can't do (if you're even trying to be fair).
Author here. Seeing comments that this is a doomer article for some reason. I'm an AI optimist. It's humans that scare me. The article is about managing the transition to Valhalla. My toy projects and my job can disappear and I'm fine with it.
Create shows and movies. I, like you and everyone else, will figure it out. And it'll be better than what we have today just like its better to drink coffee and code vs toil at a farm 20 hours a day.
Yes. There will come a day where an AI can create you a movie you want to see, on the fly, and will be able to watch your body language for when you get bored to pivot to another topic/twist in the movie etc.
reply