AbbeFaria

I haven’t watched the lectures yet. Reading the papers and using Claude/GPT to ask for references etc is good enough for me. I did watch a few, ones on GFS and Raft but didn’t see the ROI, lectures are too long. Reading the papers multiple times and reading blogs, references and asking Q’s to GPT is what I am doing.

Obviously for the assignments themselves, I use AI sparingly. Mostly as a sounding board when I have doubts or to know how to do something in golang, since it isn’t my primary programming language.

Note, I am not asking for solutions like “implement Append Entries RPC”, the solutions are public so AI could probably implement everything. My workflow is to implement it first, debug test failures, go through each individual test and then come up with clarifying Q’s, such as the one below,

In Raft paper by Ongaro et al, for the below condition, If there exists an N such that N > commitIndex, a majority of matchIndex[i] ≥ N, and log[N].term == currentTerm: set commitIndex = N (§5.3, §5.4).

Is commitIndex changed only when we receive a successful reply for Append Entries ?

Lastly I have an AGENTS.md that prohibits GPT from giving solutions directly, that would defeat the purpose of doing the assignments. It was used in one of Stanford’s GenAI course, look it up.

+1 to this. I am on Assignment 3D (dealing with log compaction in Raft). I have learnt more about Distributed Systems just by "auditing" this course.

For Paxos, I also recommend reading Paxos v Raft, Have we reached consensus on distributed consensus ? by Heidi Howard et al. It explains Multi-paxos by using the style and abstractions from the Raft paper, which makes it much more understandable. Paxos can be intimidating but this was the paper that made it click for me.

I highly recommend doing the labs from 6.824.

I know the article is AI slop just by the structure, my last job at MSFT had me producing artefacts like this, which is why I quit ;)

The issue with buying land outside major metro areas like BLR are the following,

1. Irregular power supply. I lived 2 hours outside of BLR in a village and the power would regularly go off for minimum of 3 hours, sometimes more. The current KA government made a promise to voters to supply subsidised power and the downstream effect of this is that they must mandatorily supply power to BLR which is the only economic engine of KA. Consequently they end up shafting everyone else.

2. Poor education quality. The best teachers and smartest students are concentrated in BLR, unless you are willing to homeschool or are fine with your child not getting the best, this is a difficult tradeoff to live with.

3. Narrow minded people in villages. Unless you live in touristy areas like Coorg, CKM etc you will not find the diversity that you find in a big city like BLR.

4. Less than ideal medical care. Most villages don’t have the sort of hospitals and doctors you would find in BLR. The best doctors who would be the best paid won’t come to villages. Although I did live near a teaching hospital and the medical care there was average. If you’re a healthy individual, you might ignore this.

Surprisingly the roads were decent, a result of being near a highway and having very low population density.

Apart from these, I was living a “slow life”, had great AQI (most days), wide open spaces, great and cheap food. If you’re fine with the shortcomings, buying land in and around major cities is affordable but the shortcomings I mentioned are real and I don’t see them going away anytime soon.

As to the efforts to decongest BLR, there are quite a few like Bidadi township but they are are more like bringing BLR style haphazard development to the outskirts, but these things are still very much in the nascent stage now.

Meta is a cash generating behemoth. They have such a well oiled ad machine that it’s going to be a long time before they destroy themselves. Case in point, they look fine financially even after burning mountains of cash on the Metaverse.

I would worry about Meta dying only when people stop using IG, which they clearly are NOT. IG is firmly entrenched in modern society and culture, they make insane amounts of money off of that.

One thing is becoming more and more clear though, their AI efforts are going nowhere. I would be interested in where they end up in the AI race, they are never going to get as much mindshare as Claude, Codex or Gemini.

Excess verbiage in communication. Design docs, root cause debriefs etc all are clearly AI generated with little thought to whether they are helpful to the reader. They are certainly helpful to the writer as they can just offload it to the AI.

Its hard to discern the kernel of truth that the author is trying to communicate. So much of documentation is now AI generated and rarely does it help in understanding code bases or diagnosing issues in production.

This was my experience at MSFT for the past year or so.

So you and OP believe that, these systems are good indicators of skills, so much that the last statement sounds normal?

Op here, yes I do believe, the levels aren’t perfect though. I believe in them because I was once L60 and L61, I knew intimately what was expected of me then and I can assure you the work I was doing as an L63 could be done by someone few levels down, which is to say I was underutilised. The team was also not the right fit for me.

Levels tell you two things, the scope and autonomy you get in your work and more importantly compensation. From the compensation view point, it does tend to be “gamed” frequently by developers hired from outside who might be out earning devs promoted within.

Have no idea why people here are picking up on MSFT’s levelling system? I didn’t invent it.And it actually starts at L59.

The point I made was that as an SSE (L63), there’s a certain amount of scope and autonomy that is expected neither of which I was getting and hence I resigned. I am not trying to bully or denigrate anyone junior.

The levelling system specifies the output and the characteristics of the output expected out of an engineer, that’s it. Whether I believe in it or not is beside the point, I was in the system so I did believe it otherwise progressing through my career would have been impossible.

You are not alone. I was in this exact same position at MSFT and I put in my resignation. I am an L63 but the work I was doing, was something an L60-L61 could do and I frequently felt I was in one of those Bullshit jobs (courtesy of David Graeber). I was paid handsomely but once the sign on stock ran out, I saw that I was staying in the job just for security. I felt like one of those Hooli engineers who were sunbathing at the Hooli office terrace waiting for their stocks to vest. I am only 9 years into my career and I didn’t see that as the optimal thing for my career rn.

I didn’t have any major financial obligations like you though, so it was a much simpler decision for me.

Hang in there buddy and also thanks for the deeply human comment.

Here’s my take as an Indian having lived in cities like Mumbai, Hyderabad and Bengaluru

1. No manufacturing expertise in India, we aren’t a manufacturing powerhouse despite having similar levels of population as China. We have always prioritised services. That has an inherent limit.

2. Local governments here are powerless. I don’t have the source for this but only about 3% of public spending is done by local governments at the municipal level. Even for “rich” cities, local governments although are wealthy they are always controlled by whoever has power at the state level. It’s why Indian cities have such decrepit infrastructure vis a vis China.

3. Caste based politics and more recently freebies based politics. Recently, state governments all over India have been all out bribing voters with cash transfers if voted to power. This won’t bode well for long term financial stability. This is also short term thinking at its finest. They have run out of ideas.

4. Weak rule of law. There’s a huge backlog of cases in our justice system right from Supreme Court to the local courts. Doing business in India can backfire in spectacular ways (inordinate tax demands by union government etc).

Entire studies can be done but these are top 4 things keeping India where it is.

Is this course still available? What about the course materials? I know it will be dated but if so can someone pls share the links. Tried searching for it on google but couldn’t find it.

It’s more nuanced than that. There are lines of business where there is plenty of growth and places where growth has stagnated or business is decelerating.

In the former, you can see some of the products that MSFT is shipping such as MAIA chips, Azure Horizon DB, MSFT Fabric, Sovereign clouds etc. These businesses are seeing steady growth and there’s plenty of work to be done whereas for products like IoT, XBox and anything to do with gaming in general etc you will see that there isn’t much growth left and yet these products have a lot of headcount behind them.

So MSFT isn’t stagnating (revenue keeps increasing). Although there are quite a few businesses within MSFT that can use a lot of headcount reduction, this is my take as a low level L63 grunt.

This matches my experience at MSFT. Very little work for too many people, MSFT CFO also said in the latest earnings calls that MSFT will have fewer people next year than this year, they offered buyouts to tenured people, did layoffs in LinkedIn not sure if there’s more to come.

There’s a hierarchy amongst knowledge work and AI hasn’t yet been able to do the work that is rare and valuable.

Over the past two decades, there have been lot of solved problems like building boring scalable web apps, UX design etc and AI is fairly good at this, enough so that good prompting can get you very far. This shouldn’t be a surprise, there’s a lot of publicly available data for this (GitHub repos etc).

On the other hand, there are rarer Computer science problems like designing efficient Datacenters, GPUs, DL models. Think about problems that someone of Jeff Dean’s or James Hamilton’s (AWS SVP) or a skilled Computer Architecture researcher like David Paterson’s ability would solve. These are incredibly hard and rare problems and AI hasn’t been able to make much progress in these areas. That’s true for other sciences as well.

If you’re a regular Joe like me who builds boring CRUD apps, AI is coming for you.

What I mean is if you are working on incredibly hard and rare problems that require rare skills and also those problems don’t have publicly available data that LLMs can be trained on, you’re safe from being “automated” away. If not, you must plan accordingly. Also if you’re a skilled manager (in any field) AI cannot replace you, highly skilled managers that can get the best out of their teams have rare skills that aren’t easily replicable even amongst humans much less AI. Although, if going forward we need fewer developers we will need fewer managers too.

because the reader must now sift the synthetic context for whatever the document was originally about

time wasted using AI on tasks that did not need it, on artifacts no one will read, on processes that exist only because the tool made it cheap to construct them. On decks that spell out things that previously didn’t even need to be said or were assumed.

I work at MSFT and at-least in my org, this is happening at warp speed. Every document I read, my first thoughts are what is the kernel of the idea that the writer was trying to convey ? Because 95% of the content of the doc is just verbiage. You can always tell its verbiage, the em-dashes, the rhythmic text, the green check mark emoji etc. We are hoping that volume of output will make up for the quality or lack thereof. More markdown files, more AGENTS.md file but is that making us better developers ? It certainly is giving the illusion that we are faster but I don't know how management thinks this will lead to tangible impact on the top line or bottom line.

In my experience, some of the best writing (in design docs and PM specs) at MSFT have been human written. You can see the clarity of purpose from the writer, ithere is no need to read it again, it is equivalent to having a 1-on-1 with the writer themselves. But AI written slop, the less said the better.

This piece hits home, I wonder how the experience is at other Big Tech companies.

I work at MSFT. I can understand the incentives behind this change. Although I am not sure how different GitHub culture is from MSFT.

I am sure they are closely tracking this metric of Copilot authored PRs so that everyone down from Nadella to the dev and PM for this can use it to hype up GH Copilot. It’s also a simple and clean metric that goes well in your Connects (performance discussion), you could say the feature I worked on led to xx million copilot authored PRs and there is now an AI usage mandate and you need to mention how you used AI to do something more efficiently blah blah. It’s good old promotion theatre. I don’t think its unique to MSFT though and is probably common across Big Tech.

the Microsofties viewing their IBM colleagues as mired in pointless bureaucracy and the IBM folks viewing Microsofties as undisciplined hackers.

I work at MSFT, this made me chuckle hard. Microsoft must have been a very different company back then, because now I find myself and my colleagues mired in pointless bureaucracy via endless meetings, AI mandates, promotion theatre and the list goes on. I am decently paid but the bureaucracy is soul destroying.

I work at MSFT and can concur. It’s dystopian that this is industry wide. It’s just stat padding tbh, things that people can bring up during their performance discussions, oh hey I ran this AI initiative, wrote a bunch of markdown files, that had..what impact exactly ? Never mind.

Don’t even get me started on the meaningless meetings that make The Office seem like a serious drama.

Random reward mechanisms have been woven into the very fabric of the digital world, so doesn’t surprise me.

If you have the self-awareness to realise that your job is BS and is deleterious to your well being the path of least resistance is to simply quit and find another job hopefully that is more relevant to what you’re skilled at.

All this chicanery to escape work would only eat at my conscience. I don’t believe for one moment that I should be emotionally attached to my employer but at the same time I am a professional and am paid to do a job. I should hold up my end of the bargain. Being deceitful 24X7, 365 days a year will eventually wear me out because there’s always the chance I will be found out. There’s also the reputational risk of being found BS’ing which is non-trivial and could ruin chances of future employment.

The path of least resistance is simply to quit.

There is nothing wrong with using an LLM so long as a human takes ownership for the artefact (be that books, code etc).

I would rather the author automate the mundane and focus on conveying their ideas clearly.

As an aside, is there a Linux version for this ?

I work at MSFT and I feel burnt out too and am in a similar situation where I feel like resigning would be better for my mental health but AI isn’t a big contributing factor. I do have some arguments against speculative uses of AI though.

Experimenting with speculative uses is fine, technological breakthroughs require lot of iterations and some would naturally never make it but with the enormous amounts of capex that companies are investing, these have to impact the top line and eventually the bottom line as well. I just don’t see that happening now, I could be wrong.

1. To me speculative uses of AI like meeting notes summarisers seem to add little value if at all. First off, most meetings are performative work especially at big companies. Add to this, when someone just casually pastes the meeting notes from an AI summary and asks the meeting organiser to “pls check for correctness”, my blood just boils. Are we spending billions of dollars of capex for this ?

2. Every team builds their own “agent” for diagnosing incidents which is announced to huge fanfare but people rarely end up using it irl.

3. Devs and PMs chasing “volume” of work. You prompt GPT for an issue and it is bound to give you pages of text that you can use to show how much of output you can churn. I have seen excessively verbose design docs that only the writer (and prompter) could understand and all this was accepted because “Hey, I used AI for this and it must be good”.

There are legit uses of AI and I do have a 20$ Claude subscription which I like and use but at big companies they are shoving AI into every nook and cranny hoping it shows up in the top line and bottom line and so far it doesn’t add up.

Lot of these uses are driven by fear, by repeated exhortations from upper management about shoving AI into every nook and cranny when they are just as much clueless as us. People’s mortgages, their children’s education and their retirement, in short their whole livelihoods are at stake even more so when companies will happily lay off workers without a second thought. So people have to use AI even when it adds questionable value, if at all.

I am not resistant to change and am not an AI Luddite. I am happy to use AI to become a better developer but most current use cases seem to add questionable value.

At MSFT, Product Managers were Technical Program Managers. Yes, a good PM is a joy to work with.

No, once you spend that much time at a Big co, you start treating it like your own little fiefdom. Your allegiance becomes to how do I increase my own scope and not how do I build great products for customers ? Also “coordination tax” is heavy at big companies, it takes monumental effort to ship even incremental improvements. You have to inject new blood and move fast with the times. Your tenure doesn’t give you a birthright to stay at the company forever. This is not to devalue people who have contributed immensely to MSFT before, there are tenured people who aren’t afraid of change and adapt with the times. But someone close to their retirement might not want to rock the boat too much.

There’s a reason why despite investing so much in building AI tools, MSFT has a bad rep and no one uses their tools. People use CC, Codex, Cursor when MSFT already has supposedly competitive tools.

Note: I work at MSFT.

In the mature service I worked on, adding new code was “templatized”, you had to add feature flags, logs etc which didn’t vary much no matter which feature it was. The business logic was also not that complex, I can see AI tools one-shotting that and it indeed is a productivity boost. You would be surprised to know that most work was exactly this, writing rather mundane code. Majority of the time was spent coordinating with “stakeholders” (actually more like gate keepers) and testing code (our testing infrastructure was laborious). This was at MSFT. There are lot of teams that are innovating at the frontier (mine wasn’t, at-least not technically), I don’t know how AI tools work in those situations.

My experience working at Big tech companies is that people with roles like “agile coaches", "technical project managers", UX testers add questionable value. And the QA is usually outsourced to service companies like MindTree, TCS etc anyway.

Lot of these companies are bloated from having way too many Engineers anyway. Once you have mature software that brings in bagfuls of money, you don’t need that many people to keep the ship steady. I have seen this first hand at MSFT, we started a new team back in 2019 and it probably had ~40 people full time across US and India. By 2024 when I left, we had about 20 people in India who could easily run the service, the US team was dissolved and they moved to other teams in MSFT. The fact was that new features were few and the team was in KTLO mode. I have seen the reverse happen too, the team I was working on was dissolved and we were moved to different teams and everything moved to the US last year, managers were converted to ICs and a few folks were probably fired but it was a ~10 year old service that didn’t need that many people to run, even more so after AI tools became big last year.

I loved High rise and Concrete Island. He was prescient about what modern and current day society looks like.

I will give his other works a read. I tried reading Crash multiple times but it was a bit too gory for me.

I am guessing this is Salesforce and the celebrity is Matthew McConaughey who’s real chummy with Benioff.

Brave of the developer to bring it up. This cult of personality is pervasive throughout the tech industry.

My mistake, yes indeed it’s 6.828. On a side note, for distributed systems 6.824 is on my list after I finish 15-445 from cmu :)

To anyone wanting to learn about OS development, nothing beats MIT 6.824.

I finished the assignments in that course and that covers all the important aspects like processes, context switching, CPU modes, page tables and virtual memory and many other relevant topics like file systems, device drivers etc. And also it’s free.

From the table of contents this course gets too involved in ancillary matters like bootloaders or the Rust language itself whereas the focus of any OS development tutorial should be on core concepts like how processes are implemented, how context switching works, how paging and consequently multi level page tables (actually, in code) work etc.

I work at MSFT. I understand why they migrated to LibreOffice. Outside of work, I use none of MSFT products.

I do have some burning questions though, 1. How are they saving their work to the cloud if they use LibreOffice ? I don’t think it offers the same functionality that M365 suite does. 2. How are they handling IT security? Are they using a different vendor ?