I worked at a major cloud provider and saw this first hand.
When I started, I was really bothered by the fact that everything took so long. I had joined from a startup where there was time/financial pressure. Everything needed to be done now, and there was incentive to get it done.
The cloud provider was the complete opposite. Everything was slow. Things that should take a day took a week, things that took a week took a month. I rationalized it that there were more checks in place, more i's to dot, more t's to cross.
But in hindsight, it was the domino effect of Parkinson's law in play. The cloud provider had enough money to keep paying everyone in perpetuity; that removes financial constraints. Time constraints could be explained away. So things just got done whenever they got done. The knock on effect of this is that when you depended on other teams, they got around to their external responsibilities whenever, that then slows you if you depended on them.
The larger organization accounts for this by expanding timelines (because what else are directors/vp's/etc to do... they're powerless to actually implement), which are then just filed with more waste.
Relatedly, I think many people overestimate how hard it is to compete with the big players. People think that because they earn billions, they are unbeatable. But this is untrue! Scrappy, creative people who just go for it can often beat the big players. As a recent example, look at Cursor AI. Microsoft was in the PERFECT place to come up with Cursor. Microsoft had the most AI knowledge via their own teams and via OpenAI. They have enormous data centers for AI compute. They had the most used code editor (Visual Studio Code). And still Cursor came in and made a better product!
Jeff Bezos said "your margin is my opportunity". I think we can say the same for bureaucracies. Whenever there is a bureaucracy, think "your bureaucracy is my opportunity."
> And still Cursor came in and made a better product!
On a long enough timeline Microsoft wins this battle. I don't think you appreciate the inertia of market dominance and a massive existing user base. They can simply copy features and push them to millions of developers overnight. More likely, Cursor either gets acquired or becomes a niche alternative while VSCode maintains its dominance.
Competing with established players when you have zero moat is always tough and I don't think Cursor are even close to winning their category yet. Bureaucracies exist in large companies precisely because they've grown successful enough to need them. While they do create inefficiencies, they also enable consistent execution at massive scale - something startups often underestimate until they try to grow beyond their initial niche. It may be slow, but they win by outlasting, outspending or buying you.
However if you simply mean that Cursor can carve a niche quick enough to become an attractive acquisition target then I might concede that fact.
It seems like it ought to be reasonably restrictive to install some new AI tool on dev machines.
* they don’t know who’s behind it, what’ll it output?
* they don’t know the business model. Is will it be used to exfiltrate code from the company, aka train on the company’s codebase? Other text files you open?
I’m not at all saying I think Cursor is doing that—training on customer data would be a completely unethical business practice, bordering on malware, usually companies whose names get bounced around here are not so bad. But, the hypothetical host company doesn’t know anything about them, so it is prudent to require some checking.
I don't even use copilot (yet) because the process of reviewing it through all the different obstacles (security, legal, budget) is on it's 24th month or something. I think many people in traditional industry can relate.
This basically describes Slack vs Teams as a case study. Teams launches, Slack says "good luck with that," MS basically gives Teams away for free and gains market share, then SAP acquires Slack.
> On a long enough timeline Microsoft wins this battle. I don't think you appreciate the inertia of market dominance and a massive existing user base.
So Cursor shouldn't even have tried? It is pointless what they have achieved? That's what you're saying?
I disagree. Having people who know about your product and have a positive experience with your product can be very sticky. Coca Cola is a famous example. Also look at AWS. Microsoft has many of the benefits you said. Microsoft has scale, many existing relationships with customers, own much of the related software, and many more benefits. Still, AWS has a 31% market share while Azure has only a 20% market share. The gap becomes more narrow each year, but it's still not closed. There is definitely a huge benefit to grabbing market share by having a superior product. Even while you only temporarily have that.
You also talk about the economies of scale, but that was my point. Even while Microsoft had all the economies of scale, still Cursor came in and took a large part of the market!
Theory of the firm states that the corporation will get as large as coordination costs allow, but the other way to view that is that large corporations need to compensate for their increased coordination costs in some way.
Empirically, they can often do that successfully, as evidenced by the fact that large cloud providers are, indeed, large.
But you correctly point out that over time this almost guarantees opportunities for firms with lower coordination costs.
Is Cursor really better? I tried it briefly but bounced off. I’m convinced that AI-assisted search and replace would be quite nice sometimes, but not enough to switch.
Then again, I don’t use most features in Cody, either. Basic AI code completion seems good enough for me and the fancier shortcuts don’t stick.
I have mixed experiences with the "chat" portion of Cursor AI. The "composer" has been much, much better, albeit on greenfield projects. I haven't explored it too much with existing monoliths, but I suspect it might not fair as well.
I'm currently using it to build an iOS mobile game, using Swift. I've never coded in Swift, or used SpriteKit, but it's going surprising well. When it does produce invalid code, I just copy the error directly out of xCode into Cursor Composer and 9 times out of 10, it fixes the issue first try.
I've used it on different languages, but the rust it produces, even if it runs, is often full of bad decisions. Choosing improper data structures, cloning everything, etc. Curious what the quality is on the scale of "running swift" to "good swift".
I don't think so. The amount of features something like Google cloud has + experience is not cheap to just build.
Plenty of companies try in Germany the schwarz group aka Lidl and the German Telekom too.
Also normal companies are not fast. It takes time for them to learn about something new, understanding it and then implementing it before they finally can us it.
If Ms tells you they will do it in 1-2 years everyone is fine with that. And GitHub announced GitHub spark. Google has this already.
In a bigger organization you're further away from the actual mission of the organization, which eliminates the internal motivation to perform well, even in the absence of a measurable reward. For example, I'll always go above and beyond when doing something together with a close friend of mine, but I give zero fucks about my big corporation.
I believe that this is a problem with western societies at large.
The tempting efficiency of infant capitalism was its decentralized and small scale structure. You can’t govern a country top-down, and letting the people at the bottom do their own work is most efficient.
But now some companies are larger than most countries and we are back to where we started… but without transparency and democratic regulations.
I think we need better ways to deal with large systems and complexity.
There's the Chinese saying "Heaven is high, and the emperor is far away". The idea being that low level bureaucrats are actually the ones running the show. Effectively, that's where China and the CCP sit today.
However, that has similar issues to the ones I described above. Top down mandates, the low level feigns to appease them, but marginal progress is actually made. Instead of expanding the timelines as I spoke about, in the CCP(/Soviet) case, the actual accomplishments are made up to keep "going forward". Think of all the ineffective building done to meet bullshit quotas in the China currently. It harkens back to the Soviet era.
I completely agree. But the intend I wanted to capture was that if you have complete transparency over your own system, can act independently and are not bound by top-down mandates, this is where you have the highest efficiency.
This is identical to the idea of small scale teams working on systems they completely understand and own. This is where you have the most productive teams and it is the same in terms of bureaucracy. And the job of an architect is not to create "a beautiful system", but to lower the complexity of the overall system.
I think the topic of complexity and large systems is one of the defining ones of our time. Similar to entropy, you start with a low-state system where you can generate energy from putting entropy into the system. But with a higher state of entropy, your energy gains decrease.
We act like you can just pure more and more onto the same pile, but eventually, everything stagnates. And only collapse can create a blank slate.
It seems like the real problem is finding a way to address that.
We've seen this play out over and over. Take something like building and zoning codes. They start out doing something worthwhile, like requiring fire exits. But soon you have a bureaucracy micromanaging everything, imposing minimum parking requirements even on a building sitting on top of a subway stop, imposing minimum lot sizes or density restrictions, requiring unnecessarily costly or labor-intensive construction methods at the behest of device makers or trade unions, etc.
What you need is a mechanism, something like checks and balances or a challenge process that allows any rule to be repealed if it doesn't still have enough votes to pass in every branch, to inhibit that process of ossification and corruption and clear out the cruft accumulated by its past operation.
The problem isn't so much that you need to throw sand in the gears. We've eroded some of the checks there used to be (e.g. moving to US Senators having statewide elections instead of being elected by state legislatures) and maybe we should put them back, but in general the system of checks and balances works kind of okay.
The issue is that it's not perfect, and it's never going to be perfect, so you're occasionally going to have things get past the gauntlet and make it into law when they shouldn't. And over time those things accumulate.
So what you need is a functioning process for clearing out the cruft even once it's already there. Something like, make it much easier to repeal rules than enact them.
> What you need is a mechanism, something like checks and balances or a challenge process that allows any rule to be repealed if it doesn't still have enough votes to pass in every branch, to inhibit that process of ossification and corruption and clear out the cruft accumulated by its past operation.
I agree in the sense that some way of "caretaking" is absolutely necessary. Like in a badly maintained IT-system, many managers think you can just keep piling stuff ontop of the old system and you can just happily going forward forever. But of course that is a terrible misunderstanding.
In an IT-system, you either restart from scratch or start a refactor. I think these are the only options here either. This needs talent and money, something the public services have been massively drained of. You can't make a system more lean and efficient by cutting costs, you mot make investments. If you involve Civil Servants, they will tell you exactly what kind of issues they are facing each day processing official documents.
Ideally, you'd start a digital transformation of public services and also streamline the legal cruft at the same time. But the problem is that much of this "cruft" is there on purpose because somebody profits. Such as the impossible Zoning Rules in California to prohibit any additional housing space... because that would make housing more affordable and lower their prices.
That's not my point. My point is that overhead increases and efficiency decreases the larger your system is. And companies at the size of states are just as inefficient.
Ergo, the idea that the market handles their duties more efficiently is a lie. It has nothing to do with corporate vs state owned enterprise, but a matter of complexity.
True but also the sheer number of people now trying to use it. Other factors include the difficulty of getting an appointment with the first call GP (once known as the family doctor) and the utter madness of many A&E departments. They do a great job but can't cope. 12-hour waits are not unknown. I see no reason to expect a turnaround with the Labour Government.
It would seem building better systems is not enough. We must strive to maintain, continuously improve, and scale them to meet demand.
Sadly culture in the US is diluted into believing that profit motive is a silver bullet that fixes all problems. And many are raised in religions that teach one to deny critical thinking, embrace appeals to authority, and love confirmation bias.
I work in a large law firm and see similar outcomes, coming from a similar fast paced background.
I’ve drawn a different conclusion as to “why” though. Law firms are inherently risk averse. As such things do take time to dot the i’s, and that is more important than the time or financials.
I look at it through the project management lens of the three constraints: Quality, Time and Cost.
The law firm (and potentially your cloud provider) has prioritised Quality. Either Time or Cost take a hit.
High quality + quick = costly.
High quality + low cost = a long time.
When I started, I was really bothered by the fact that everything took so long. I had joined from a startup where there was time/financial pressure. Everything needed to be done now, and there was incentive to get it done.
The cloud provider was the complete opposite. Everything was slow. Things that should take a day took a week, things that took a week took a month. I rationalized it that there were more checks in place, more i's to dot, more t's to cross.
But in hindsight, it was the domino effect of Parkinson's law in play. The cloud provider had enough money to keep paying everyone in perpetuity; that removes financial constraints. Time constraints could be explained away. So things just got done whenever they got done. The knock on effect of this is that when you depended on other teams, they got around to their external responsibilities whenever, that then slows you if you depended on them.
The larger organization accounts for this by expanding timelines (because what else are directors/vp's/etc to do... they're powerless to actually implement), which are then just filed with more waste.