I also work as the head of a technology company, not the CTO, but high enough I could offline the company and bankrupt it tonight. I've worked here for 10+ years, and we've grown from three people to eighteen, and I know that's a hilariously small number still for the size of our company.

Here's my honest rules for when to hire:

1. Is anyone taxed continuously above 85%?
2. Do we have single points of critical failure?
3. Do we have critical infrastructure not covered with a backup person?
4. If someone leaves for vacation, is the company standing still?
5. If a client demands support on a weekend, can someone help?
6. Can someone reach an expert in their team at the company reliably?

If at any point I feel there is a gap that could introduce a lack of trust, support, or safety, we hire. If I feel that the infrastructure is at risk, we hire. Importantly, we never hire, just to hire, you need a role to go into, and you need responsibilities you'll manage. No one gets hired to sit on their ass, and no one gets hired just because head counts look good.

Just in case you're wondering if I could actually offline the company tonight no, I wrote all of our DevOps, and CI/CD tooling from scratch. If Azure went offline tonight, it would take ~1-hour to stand everyone back up on another cloud provider, with two commands from our tooling. It would take three people to offline the company, since I built in a forced approval mode, and everything is backed up properly in three locations, one of which I can't access, to prevent me from nuking all sources. I can request access, but it takes the owner, and a tech lead to both approve access.

I know companies our size who have hundreds of people. What I can't figure out is why, a person is expensive, let's assume 90k / head, on average, 11 people is $1-million / year. Hyperscale size is a different mindset, definitely, but 5000+ employees, again, just seems weird.

Regardless if we're talking about 30, or 300 people, I just can't figure out or wrap my head around 5260. When I worked at BlackBerry (Research in Motion, back then), some teams had four people on them, with four manages in the stack, reporting to their own VP, thats what this sounds like for that many employees.

I worked in QA, and above me, from memory, was Rob (Manager), Sue (Director), Mike (VP), then shift to the Project Management team, dear lord. In that team, I reported to a team of people, around 12, who managed communicating with some app developers. To report a bug / issue in an application, I had to file the bug report, include Rob, and wait. It took over a week before I would get a response, and it was always the same, "We need additional information.". Eventually, I snapped at the chain in an email, and started to email the app developers directly. I took a 5+ days process, down to 10-minutes, with better communication and better outcomes.

One day I removed 15 people from a process, and still CC'd my manager, and it was fine. The chain of communication had no value, and I'm not being rude, I'm being honest. We used to hold meetings, to plan the meeting, which planned the meeting, with 20 people on a call in multiple time-zone, all so I could get a response from a developer in a different company. The project management team would frequently modify my messages and requests, destroying them. At every meeting I was mangsplained what words meant, incorrectly.

The actual fallout from doing that, was pure rage from multiple teams, managers, directors and VPs. I had to sit in a meeting with a board of executives to explain myself. I still remember that meeting, and answering back to a furious VP (paraphrased): "Interjecting multiple teams and people, into a chain that I need to have control over is not only hurting communication, it's making it impossible. The developers are happier, they like taking to me directly, since we can get things done, which no one else can seem to do. Last week we had a meeting where a bunch of people argued me on an email I wanted to send, destroying the question so it could sound "professional", and then refusing to send it because it was a bad time. The developer and I resolved that issue within 20-minutes, over email, before lunch yesterday, and it's all recorded in email, with Rob CC'd, so stop this stupidity, this is not a functional environment.". That's very close to what I said, not a direct quote, but very close.

VP Grumpy demanded to see the email I wanted to send, read it, then yelled at the PM lady who refused to send it. I'm not joking that by this point we maybe had 40 people involved, 40 people for me to email an app developer. If they wanted insight, the email server had the logs, and emails, nothing was hidden, I CC'd my manager, so it was just literal waste. Many departments worked the same way, and I'm willing to bet Snap is the exact same way.

Interesting, I'll take a look at that later :) - Thanks!

Oh definitely, that's already in the works, if not present.

Does anyone believe this will not be used to profile and track users? If you have to use ID to verify / validate against an app, how is that processed? Unless it's done offline in a secure enclave, the government / body will know you've uploaded the ID, and have all device identification, resulting in a large fingerprint. Once they know that, any site you visit can likewise be linked, resulting in the government knowing what you visit and what.

I've not against age verification, I'm against bad age verification. I've explained the idea a few times, but the short version, an online enclave downloads databases full of ID hashes, then disables any network connectivity, a full blackout. The offline enclave starts with a hard kill switch if any network connectivity is detected. The DBs will be transferred into the offline enclaves and the ID will be privately verified, with an age range stored. Then the ID and all DBs for this process are wiped, the enclaves are destroyed, and securely wiped, and network connectivity is restored.

Once that's done, you've verified your age, without handing over your paperwork, it's private, and accomplishes the same goal.

I wasn't talking about the network scope, or the user reach. The application is simple enough, so accounting for a few dozen IT / Network / Computer specialists (IT), which I put in my estimate, I'm still baffled by how they could have more than a few hundred employees. Running at scale is hard, that's fair, but, you don't need thousands of employees to do that. My guess is: They have a core IT team that does 99.5% of all the work that's a couple of dozen people, and a few hundred of extra "find the any key" level of IT JR employees.

How do they have that many? I think they just over hired and now need to rapidly cut head counts because what could that many people do? Assuming you have a dozen developers, maybe a dozen IT specialists, a dozen computer specialists outside IT, some management, and accessory staff, how do more than a few hundred people work at Snap?

Ignore AI, I want to know what those people do, 250 employees would sound like a lot, but 5250, would be 21x what sounds as a reasonable number. I'm always amazed by how many people can work at a company whose product is reasonably simple. My general guideline is that when a person hits 85% utilization, steady, you need to hire, and every person should have a backup if they're critical, but how are 5250 people working at 85%?

AI does not comment the generated code to a level I would consider proper. It can generate JR developer level comments, just like the classic meme, "This is a stop sign", which isn't useful.

The code might be the same as a SR developer, it might not, I've seen it generate brilliant code, and truly terrible code, it's a spectrum. If you're careful, and you review the code, and really understand it, there's no problem. The big issue is when people accept the generated code and move on without review.

Exactly, you need to read and analyze the code, and if you need to make sure it's accurate. There is no problem if you're careful, my issue is many people aren't careful, and just accept it.

Perfectly said, AI is really a massive negative for programming. AI is great is you need boilerplate, or if you're stuck and want a suggestion, but outside of that, you should avoid it. If you don't know what the code generated means, or how it works, you can't debug, you can't support it, and you can't claim it's safe. That's the danger of AI, it can generate a lot of code, but in my experience, that code needs to be carefully checked and commented.

I've said, in one form or another, that code should be 50% comments, and I say that because you should have logic, steps, explanations in the code. Someone sitting down should know what I wanted to do, how I wanted to do it, and why. If my idea was stupid, fine, but you know what I was thinking, and why. If my idea is wrong, at least you can follow along.

AI can't do this, it can't explain why, and it can't break down the steps, which means you're programming with blinders on.

No, you don't want me to do my job, you want me to do some other person's job, even when I try to sort the mess out.

There was another case with a different plant, where the IT team for that plant changed a firewall rule to block access to SQL. The issue, that test fixture would call out to our SQL Server to get a serial number, it was two UUIDv4's merged. After 6+ months of running, they called me that the boards started failing with an error regarding serial numbers. Weird, since I had nearly a trillion available, and it took 5-minutes to find the issue. What should I have done then? Contact the plant IT team, which I did, and explained the problem, which I did. They gave me some story about SQL being too dangerous to allow.

The boards actually worked fine because if the test fixture couldn't reach the DB it would generate its own UUID sequence, register it on the fixture, in some memory, and wait until it could access the DB again, then do the updates, and registers the serials. The error was generated because the memory filled up. In the documentation, that IT team signed off on, it clearly said we needed SQL access through the firewall. I had the signatures on the paperwork.

In that case, the solution was to drive out, and manually sync their fixture to my notebook which had a local copy of the DB running. We had to do that every 6 months, and I never charged them, I never yelled at them, and I supported that idiotic workaround. After a while, we gave them a local copy of the DB, and I changed the firmware, for free. The problem that caused, if replacements were needed in warranty, I had no way to validate the serial numbers, which we bit the cost of.

If you think that's reasonable, and part of my job(s), then you have grossly misunderstood responsibility.

Exactly, I love my job, it's enjoyable, but when you present a problem that I can't help with, not because I'm lazy and don't want to because the computer is a plant I can't access, and I'm not authorized to work at, what do you expect?

I even tried in that one example, I really tried. I changed the messaging, I took their calls and emails, and I found the IT person, Geoff, and provided him the information. He talked to the production manager, and told the production manager to call him, and they still screwed it up.

There was another batch of issue once where a large apartment / condo complex install our devices incorrectly. They installed it in every unit, under the water pipes, which started leaking, they leaked onto our electronics and destroyed the devices. The management office blamed us, for covering the water pipes when they installed the devices. In that case, I order new boards, had them conformal coded, and bit the cost to replacement them. Every single day I showed up with a batch of boards to install, they gave me a speech about how unacceptable it was that I had to do this. The installation manual, which every device came with, had a warning not to let the boards get wet, It was an entire page warning.

I can keep going, I have tons of these stories, where people don't listen, shove their head into their ass, and blame me for the air stinking. In every case, I try to help, even if it cost money, time, resources, until it goes so far into the ridiculous, that it's clear it's intentional incompetence.

I understand about the old lady, that's an outlier, but most people, unless they're being willingly ignorant and incompetent, can get up to speed pretty quick.

Take modern KDE and Gnome, if you honestly can come up with a list of ways people would have trouble adapting to it, please do! My mother came over recently, needed to check her email, and figured out how to do it almost instantly in KDE. Her computing competence is somewhere around the "What is the "any" key". It's no longer an impasse to use modern Linux with a modern GUI.

15-years ago, sure, Linux wasn't an option on the desktop for the average person, but now? Windows is now the problem operating system, I think most people would find it refreshing to use a modern, performant, light, fast, operating system that looks good, which Windows is not.

No, I don't care if you don't care. If you call over and point to the computer and say "box thing, nothing", then what do you want me to do? If you sport no competence, intelligence, or ability to communicate, what do you want me to do?

The screen tells you to call someone else, and even with that message, you call me? What do you expect me to do? I was nice at first, and changed the message, and you still screwed it up, and was still nice, until everyone did it.

That's when I don't care, and won't care. If you operate with such grotesque incompetence, that you're clearly making a mockery of being alive, and you ignore any form of common sense or competence, then ya, screw off.

On the other hand, I'll go to bat for you if you need help, and the help request is reasonable. I'll move mountains to help, if I can help, and you've clearly tried excluding pure incompetence as the root cause. I'm not a "jerk IT" guy, I only become a jerk when the people who are making demands clearly can't process their own reality. You can't call me about a computer in a plant, that I have nothing to do with, with a message on the screen telling you to call the right person.

I have to be honest, I haven't run into a serious printer issue, that was any worse than Windows, on Linux or Unix in 15 years. I really haven't, and in many cases the Linux / Unix experience is just easy, compared to Windows. For instance, on Windows 11 you can no longer scan, it just tells you that you need a new driver and to pay for a service. On Fedora, it doesn't care, it just scans and scans, and it's great.