Do you think they'll continue to win in enterprise? As a casual office user, who's had to do some PowerPoint and word docs recently, I found the experience of using office 365 truly miserable. All of them are laggy and horrible to use.
I think by moving onto the cloud they've left themselves open to being disrupted, and when it comes it'll be like Lotus Notes, an extremely quick downfall.
They have enterprise users locked in mainly due to Active Directory, for which there is no good replacement, and to a some extent SharePoint. There's also Office, of course, and you are right that the migration to web tech isn't well taken. I'm thinking of "New Outlook" in particular. They probably plan to EOL classic Outlook when Office 2024 EOLs in 2029. The last stronghold will be Excel. If native Excel ever gets discontinued, then everything Microsoft will have been webshittified™.
Trust me, I really want that to happen, but who has the billions to burn (and the will to use them at that) to build a solid alternative? Most probably, the EU will have a misguided shot at it, out of desperation from the USA, and will subsidise some inadequate local actors. I'm not sure whether it will be good, timely nor sufficient.
Depends on your age. I remember being warned in my 20s that older people couldn't read 10pt font, 12pt was a stretch, I didn't really believe them.
Now I'm in my 40s, oh wow. Small, illegible, font is everywhere. Instructions on food is especially bad for this. At least on the computer you can usually force 125% font rendering.
Point being, the site is probably quite legible to people in their 20s.
It's so easy to measure too. 15k is 750 lines per working day.
There's no way any developer can meaningfully test, read and review 750 lines worth of changes per day.
And with AI code you've got to really go through it with a fine tooth comb, you can't scan the code. There's loads of subtle, but bad, bugs and assumptions it makes without telling you. I really want to emphasise the assumptions bit. A good example I saw recently was it assuming the sort order of a DB call should be something totally wrong, but easy to miss unless you read every line meticulously.
So either the OP is lying about how many lines their AI is pumping out, or they're checking complete AI slop untested and unreviewed into the codebase.
That's why you dont use LLMs as a knowledge source without giving them tools.
"Agents use tools in a loop to achieve a goal."
If you don't give any tools, you get hallucinations and half-truths.
But you give one a tool to do, say, web searches and it's going to be a lot smarter. That's where 90% of the innovation with "AI" today is coming from. The raw models aren't gettin that much smarter anymore, but the scaffolding and frameworks around them are.
Tools are the main reason Claude Code is as good as it is compared to the competition.
> The raw models aren't gettin that much smarter anymore, but the scaffolding and frameworks around them are.
yes, that is my understanding as well, though it gets me thinking if that is true, then what real value is the llm on the server compared to doing that locally + tools?
You still can't beat an acre of specialized compute with any kind of home hardware. That's pretty much the power of cloud LLMs.
For a tool use loop local models are getting to "OK" levels, when they get to "pretty good", most of my own stuff can run locally, basically just coordinating tool calls.
Of course, step one is always critically think and evaluate for bad information. I think for research, I mainly use it for things that are testable/verifiable, for example I used it for a tricky proxy chain set up. I did try to use it to learn a language a few months ago which I think was counter productive for the reasons you mentioned.
How can you critically assess something in a field you're not already an expert on?
That Python you just got might look good, but could be rewritten from 50 lines to 5, it's written in 2010-style, it's not using modern libraries, it's not using modern syntax.
And it is 50 to 5. That is the scale we're talking about in a good 75% of AI produced code unless you challenge it constantly. Not using modern syntax to reduce boilerplate, over-guarding against impossible state, ridiculous amounts of error handling. It is basically a junior dev on steriods.
Most of the time you have no idea that most of that code is totally unnecessary unless you're already an expert in that language AND libraries it's using. And you're rarely an expert in both or you wouldn't even be asking as it would have been quicker to write the code than even write the prompt for the AI.
I use web search (DDG) and I don’t think I have ever try more than one queries in the vast majority of cases. Why because I know where the answer is, I’m using the search engine as an index to where I can find it. Like “csv python” to find that page in the doc.
The vast majority of artists in all fields don't really have their own style and are just copying other people's. Doesn't matter whether we're talking about art, literature, music, film, whatever.
It takes a rare genius to make a new style, and they come along a few times a generation. And even they will often admit they built on top of existing styles and other artists.
I'm not a fan of AI work or anything, but we need to be honest about what human 'creativity' usually is, which for most artists is basically copying the trends of the time with at most a minor twist.
OTOH, I think when you start entering the fringes of AI work you really start seeing how much it's just stealing other people's work though. With more niche subjects, it will often produce copies of the few artists in that field with a few minor, often bad, changes.
Sure, you can say that AI is just "stealing like an artist", but that makes the AI the artist in this scenario, not the prompter.
It bothers me that all of the AI "artists" insist that they are just the same as any other artist, even though it was the AI that did all of the work. Even when a human artist is just copying the styles they've seen from other artists, they still had to put in the effort to develop their craft to make the art in the first place.
While the multiplier is less for me ( perhaps 3x or 4x ) I think the assumption that productivity gain leads to directly to more money is bit optimistic. Unless you are self employed, or run your own company, and actually get paid by results, being more efficient is seldom paid much. ( with luck you get promotion every two years or so or pay raise every year )
I have worked for too long in the field, but this year and simply thanks to the LLMs I have actually managed to get 4 usable hobby projects done ( as far as I need them to go anyway - personal tools that I use and publish but do not actively maintain unless I need some new feature ), and I have been quite productive with stack I do not normally use at our startup. Most previous years I have finished 0-1 hobby projects.
The ramp up period for new stack was much less, and while I still write some code myself, most of it at least starts as LLM output which I review and adjust to what I really want. It is bit less intellectually satisfying but a lot more efficient way to work for me. And ultimately for work at least I care more about good enough results.
Completely unrelated, but what’s with the spaces inside the parenthesis here? It’s super weird (and leads to incorrect text layout with a standalone parenthesis at the end or beginning of a line…)
I love this! I mean, (a) I feel I am 20x faster / better. I have not yet reached the stage where I can scientifically put a "x" on how much better I have become.
Am I getting more money>> I think I am currently putting the correct amount of seeds which hopefully will flourish in 2026 with results. But now, it is easier and faster.
But yes, I am sorry for claiming a more "felt like" number.
When it comes to small focused OSS tools, you can basically code them at the speed of thought with modern LLMs. The slow part is figuring out what you want to do in the first place, the implementing is basically instantaneous. It's literally faster than googling a tool that already does the job.
And yes, that doesn't scale to all problem domains or problem sizes, but in some areas even a 20x speedup would be a huge understatement.
I met a seasoned game dev who complained to me he was only ever hired at the end of projects to speed up the code a bunch of mid/junior level game devs the company had used to actually make the game. Basically he said there was only so much time he'd get given, and he'd have to go for low hanging fruit and might miss stuff.
We've only got a couple of game dev shops in my city, so not sure how common that is.
Over 20 year I've had lots of clients on self-hosted, even self-hosting SQL on the same VM as the webserver as you used to in the long distant past for low-usage web apps.
I have never, ever, ever had a SQL box go down. I've had a web server go down once. I had someone who probably shouldn't have had access to a server accidentally turn one off once.
The only major outage I've had (2/3 hours) was when the box was also self-hosting an email server and I accidentally caused it to flood itself with failed delivery notices with a deploy.
I may have cried a little in frustration and panic but it got fixed in the end.
I actually find using cloud hosted SQL in some ways harder and more complicated because it's such a confusing mess of cost and what you're actually getting. The only big complication is setting up backups, and that's a one-off task.
As someone who has set this up while not being a DBA or sysadmin.
Replication and backups really aren’t that difficult to setup properly with something like Postgres. You can also expose metrics around this to setup alerting if replication lag goes beyond a threshold you set or a backup didn’t complete. You do need to periodically test your backups but that is also good practice.
I am not saying something like RDS doesn’t have value but you are paying a huge premium for it. Once you get to more steady state owning your database totally makes sense. A cluster of $10-20 VPSes with NVMe drives can get really good performance and will take you a lot farther than you might expect.
I think the pricing of the big three is absurd, so I'm on your side in principle. However, it's the steady state that worries me. When the box has been running for 4 years and nobody who works there has any (recent) experience operating postgres anymore. That shit makes me nervous.
More than that, it's easier than it ever was to setup but we live in the post-truth world where nobody wants to own their shit (both figuratively and concretely) ...
datasette and datasette-lite (WASM w/pyodide) are web UIs for SQLite with sqlite-utils.
For read only applications, it's possible to host datasette-lite and the SQLite database as static files on a redundant CDN. Datasette-lite + URL redirect API + litestream would probably work well, maybe with read-write; though also electric-sql has a sync engine (with optional partial replication) too, and there's PGlite (Postgres in WebAssembly)
So can the cloud, and cloud has had more major outages in the last 3 months than I've seen on self-hosted in 20 years.
Deploys these days take minutes so what's the problem if a disk does go bad? You lose at most a day of data if you go with the 'standard' overnight backups, and if it's mission critical, you will have already set up replicas, which again is pretty trivial and only slightly more complicated than doing it on cloud hosts.
> ...you will have already set up replicas, which again is pretty trivial and only slightly more complicated than doing it on cloud hosts.
Even on PostgreSQL 18 I wouldn't describe self hosted replication as "pretty trivial". On RDS you can get an HA replica (or cluster) by clicking a radio box.
For this kind of small scale setup, a reasonable backup strategy is all you need for that. The one critical part is that you actually verify your backups are done and work.
Hardware doesn't fail that often. A single server will easily run many years without any issues, if you are not unlucky. And many smaller setups can tolerate the downtime to rent a new server or VM and restore from backup.
Not as often as you might think. Hardware doesn’t fail like it used to.
Hardware also monitors itself reasonably well because the hosting providers use it.
It’s trivial to run a mirrored containers on two separate proxmox nodes because hosting providers use the same kind of stuff.
Offsite backups and replication? Also point and click and trivial with tools like Proxmox.
RAID is actually trivial to setup.l if you don’t compare it to doing it manually yourself from the command line. Again, tools like Proxmox make it point and click and 5 minutes of watching from YouTube.
If you want to find a solution our brain will find it. If we don’t we can find reasons not to.
One thing that will always stick in my mind is one time I worked at a national Internet service provider.
The log disk was full or something. That's not the shameful part though. What followed is a mass email saying everyone needs to update their connection string from bla bla bla 1 dot foo dot bar to bla bla bla 2 dot foo dot bar
This was inexcusable to me. I mean this is an Internet service provider. If we can't even figure out DNS, we should shut down the whole business and go home.
To summarize, other people are having to meticulously check the AI slop you're slinging into the system that looks good, but doesn't even do what its supposed to do. And you didn't even check it before submitting the PR?
Yes, I do have a map of the code in my head of any code base I work on. I know where most of the files are of the main code paths and if you describe the symptoms of a bug I can often tell you the method or even the line that's probably causing it if it's a 'hot' path.
Isn't that what we mean by 'learning' a codebase? I know my ability is supercharged compared to most devs, but most colleagues have it to some extent and I've met some devs with an even more impressive ability for it than me so it's not like I'm a magic unicorn. Ironically, I have a terrible memory for a lot of other things, especially 'facts'.
You can sorta make a crappy version of that for AI agents with agent files and skills.
There’s a company called driver.ai whose idea is to parse your codebase and provide the “map” (navigation of code structure and connectivity) to LLMs. (I haven’t tried it.)
I think by moving onto the cloud they've left themselves open to being disrupted, and when it comes it'll be like Lotus Notes, an extremely quick downfall.
reply