Let's get back to basics...

I think it's appropriate to start my Substack with something so important to me and how I work...

AI is awesome. AI is crazy. Let’s all calm down.

AI will change the way humans work, starting with white collar workers (wow, that’s different than most of the other revolutions since Leonardo Da Vinci), and within white collar workers, with anybody who ideates, designs, builds, deploys, or maintains software. As Scott Wu from Cognition, who’s pretty much maniacally focused on software, recently said, it’s going to allow us to focus on what’s important, what’s inspiring, how to prioritize it, and how to do the prioritized work amazingly well and fast. Along with us white collar workers who are going to get pretty topsy-turvy for awhile, it’s going to revolutionize most other people’s lives - apparently starting with how my family figures out where we’re going on vacation to how they fact check me (sigh). Oh, and it will likely help with amazing things (hopefully) like companionship and therapy, and maybe healthspan. Oh, and Sid is using AI to go founder mode on cancer (if you could keep the pace up, Sid, even now that you’re better - thank goodness! - that would be great.) So, wow, lots of awesome out there. And the hype / hesitation is indeed justified - it’s going to come faster and harder than it’s ever come before. The first people who can benefit are the enterprises spending $2-3T/year on software engineering - and their customers - but everything else will follow, because frankly, most businesses are becoming more and more software businesses. It’s like Jevon’s paradox but on meth - yet still the same principle.

But wow has the crazy gone non-linear. No, AI isn’t going to take over the world. Yes, people will use it for bad things, and faster than before. They’ll also use it for good things, faster than before. And no, an agent isn’t anything more than a fancy bot and a harness isn’t anything more than software and an eval isn’t anything more than a test. No, Fable and Sol aren’t going to cause security armageddon when they get out - they are going to be used by both the attackers and the defenders, and so are all the other models you hear about - and the ones govts are developing that you don’t. So let’s stop fear mongering and just get to work. Context is a fancy name for database access and document search. Tokenmaxxing was incredibly stupid and anybody who subscribed to it should be banned from ever holding a leadership job again. And, likely most important, yes, all the big players are in the process of promising to pay each other money they don’t have - that was given to them by others who also don’t have it, in a very strangely shaped incestuous cyclic graph. Yes, it’s all going to come crashing down - I’ll stick out my neck and say it’s going to be in Oct-Dec of this year, latest Feb ‘27. Hopefully the crashing doesn’t take down great companies like Cursor, Cognition, and Anthropic - but instead just rationalizes them.

Nobody knows what’s going to happen - but let’s slow down and look at it from all angles. The movie “The AI Doc: Or How I Became an Apocaloptimist” is a great way to do that.

My personal opinion is that autonomy is the new secret sauce - and developing autonomous systems is hard - when you win you win big, but when you lose, you lose even bigger, and we all know that people will fight to avoid a loss of size X 10 times more than they will to achieve a gain of size X. So the companies building autonomy are where the innovation and moats are going to be.

The basics remain the basics

The most important thing I think about is that the basic remain the basics. Delighting customers (who still want the same things) remains the single most important thing. It’s the North Star. Making commitments to the other parts of the org at your business is important - and so is keeping them, something software orgs kind of suck at. Sorting out people in your org who will flourish in the new world (likely ~70%?) and those who need to move on is just as urgent when the percentage of non performers was one quarter the size. If we focus on getting the people right, the skills right, supporting them without scaring or thrashing them, along with getting our relationships with our customers right, most other things have a way of taking care of themselves. Don’t go batshit adopting AI as your prime directive - and most definitely don’t do it in a top-down way. Reinforce for your team that everything you and they have always wanted can be done faster and better now - and let them come up with how to do it. (Hint, if they don’t choose how to use AI, then you just found your 30%). If you do things in a top-down way, when you leave the room, it all vanishes in a puff like Uber’s token budget. Yes, this is hard, and yes most of us screw it up, including yours truly.

I believe that AI is going to help software orgs deliver software at the speed, predictability, transparency, and quality the rest of the business units have always wanted. Historically, software has been held to a strange bar of being wizards and along with that has come a tolerance of software teams being deeply unpredictable and very opaque to the rest of the business. Changing this is crucial for software to actually mature into an engineering discipline - something it’s very far away from at most companies. AI can help us have PRDs in sync with architecture documents which are in sync with the code (gasp). Let’s have AI do all the testing (better than before), have reliable CI, better uptime (anybody else hope GitHub actually figures out how to have uptime like a mission critical service?) , and - real features that iterate quickly on the order of double digit hours.

Obvious thing: Somewhere around half our AI budgets should be spent on doing software right and the other half should be spent on building the right software.

Oh wait, that’s how it’s always been.

Ok, I admit that what’s important has changed a bit - for the better

Back in the 1980s (and yes, I was there and loved it…) urgency was everywhere. We moved as fast as we could. Of course, shipping tapes and CDs imposed a certain cycle time on anything - and enterprise update timelines were in mid-single-digit years in most cases. But we were as urgent as we could be! Oracle raised $31.5 million on March 12, 1986 and Microsoft raised $61 million a day later - and both companies were on a tear for a decade, innovating and out-innovating each other regularly.

But then somehow in the late 90’s and through the 00’s, it just became a matter of cranking out your next quarter, your next SaaS product, your next region, your next data warehouse format, your next migration. Most of software was pretty stunningly boring for 10-20 years. (Ok, I have to call out Linux and PostgreSQL as wonderful exceptions).

The Humdrumian geologic period of software is just so over. We’re back to the way things should be: Urgency and cleverness are the currency of the realm again. But not fake urgency (thank you John Kotter for your wonderful Sense of Urgency book). AI gives us all opportunities to shine, to work on what’s most important, in short to be amazingly positively non-linear, like we always feel about 25% of way through a project, before the boring parts set in. We now have ample assistants to help push through that and keep the exciting parts alive. It’s not just us engineering and design and product people who feel it. The world, starting with C-levels, is more open to software innovation than it’s been at any other point in my lifetime. A large TAM with an eager buyet - hooray! So let’s remember that the main thing is keeping the main thing the main thing - and that’s delighting customers with great software while taking care of our amazing employees by growing and supporting them.

I’m excited by this new age - let’s call it the Cleveraceous period. The puzzles change how they look and their shape. But they are still puzzles - and behind every puzzle are another 3 puzzles. That’s the magic of software and it shows no signs of becoming less magic.

I hope you enjoyed this. You can find more of this on marklovestech.com, especially the “Feed” tab, where I support and snark on what’s going on, a couple times a week. Would love your comments.

Lets go!

Mark

P.S. I don’t know who often I’ll post. Too many places to post. Substack seems to be the current favorite sushi roll. Let’s see how long that thats. If you want to know when I post, just smash the obnoxious orange thing below: