About that replacing of developers

Leave a reply

“The New Code” by Sean Grove at OpenAI – a charitable version of “writing the last piece of software” marketing by Lovable – says programmers will be replaced with essentially really thorough PMs who are methodical at prompting:

In traditional programming, source code is sacred. The binary is disposable.


The specification contains the original intent. The code? It’s just a “lossy projection.”

With the conclusion that in the “near future”, source code will end up like binaries – they are there, but nobody reads them and the source of truth is the really good spec expressed as the AI prompt.

Sounds appealing, but three problems:

  • Binaries are precise and functionally deterministic, they are actually really not a “lossy projection” at all. If you run even a complicated piece of code through the compiler, you will get the same result as a binary. With AI, you will get a different hallucination each time depending on the day of the week and the version of your model. So if you managed to AI-generate a version of the source code that works, if you don’t treat that source code as the source of truth and discard it, good luck generating the same thing next time! Binaries of the pre-AI time and the source code in AI-only approach do not have the same properties.
  • Nobody debugs binaries anymore aside from very small group of performance and security specialists. However, in the AI-only approach when something breaks in the source code, someone will need to roll up their sleeves and scrutinize the source code and understand how it works and how to fix it. Assuming you can just throw more LLMs at every weird production fix is currently a huge stretch, they are just not that smart – if your code base is anything bigger than one service or one fullstack CRUD app.
  • Also, in my day-to-day, GPT5 and Sonnet still sometimes produce garbage that is hard to use efficiently – more time is spent tinkering with the agent settings and LLM lying than actually moving the product forward. Yes, other people report differently. There are real – not the AI-salesbro-type -people out there that rely on a lot of generated code in production applications, especially for UI work. I think it’s often greenfield stuff generated by someone who is really really good at code reviewing.

In my day-to-day (mostly backend fintech work), most time is spent not coding, but really understanding some arcane bank formats and protocols from the 60’s (not about to change, banks), getting on the phone (sometimes literally on the phone) with them to understand what broke, deciphering crazy card swipe patterns by the merchants where multiple parties and middlemen of varying trustworthiness are involved, and when it comes to development, writing code that spans multiple weird scantily documented services, each with weird bureaucratic access controls.

If you program in one stack, in one brand new service, things like Laravel Boost MCP connector (to borrow an example from another world) can potentially take out a ton of legwork, but if you are building something that goes a bit beyond that, AI stuff seems to just fall short, as of summer 2025, not for the lack of trying.

In summary, the path from today to a situation where source code is rarely touched by a human is not yet obvious.

Why return to office is actually good for you

Leave a reply

In the last three years there has been a torrent of people telling us how life-changing work from home has been and there is no turning back. A lot of these conversations start by someone claiming that they saved 3-4 hours a day on commute, and are actually 10 times more productive at home while cuddling with pets and having tea with the wife, all at the same time.

Let’s break down why it’s an exception or a temporary arbitrage at best, and a death spell to all of these jobs at worst, and why these myopic people may need to actually embrace some of the RTO even if it ends up negatively affecting their personal idyllic bubble.

We will set aside such trivial facts that 1. not everyone is good at WFH (or likes it), 2. not every manager/org is good at managing WFH, and 3. not every job or project type is suitable for WFH. Let’s just wave a magic wand and say all these and lack of socializing/fighting for promotions/brainstorming in person are non-issues, and work from home is extremely efficient for all parties involved.

Then we run into an immediate problem: your cozy WFH job can be done not from your home, but from a home in Ahmedabad.

Your WFH job could even possibly be done by an English-native expat beach-bumming it in some tropical paradise like Koh Phangan for like a $1000/month.

Why not? There are plenty of engineering (in my case) graduates in Southeast Asia and anywhere else that’s 10x cheaper than the US. We’ve had plenty of precedents where manufacturing jobs were moved overseas, and during COVID progressive-looking tech companies claimed it’s only fair to pay the same salary anywhere in the US.

Well guess what? That Reddit move hurts everyone who lives in the high cost of living (HCOL) areas, because now they get the same salary as people working from Kentucky but have to spend a ton more on basic necessities.

You could say, screw the HCOL people, they are spoiled anyway, who cares, let them migrate or otherwise fizzle away, the future is everyone WFH from Kentucky and there is no turning back. But why Kentucky and not Bangladesh, Thailand or Guadalajara? If WFH was so efficient, why should the company pay you even a Kentucky salary, when they can pay you a Pakistan salary? The typical hurdles like remoteness, time zones, language barrier and lower-quality education are slowly being removed, with education improving, current remote and asynchronous tools getting better (and Latam being in the same time zone anyway), countries churning out more and more hundreds of thousands of Computer Science (and other engineering) grads.

The fact that for the last 20 years despite having all the modern work tools we’ve only been moderately successful at outsourcing – and many companies are still willing to shell out top bucks for salaries in the US cities – indicates that WFH is probably not going to take over in a massive way. Something still doesn’t fully click with remote, and if you care about your US salary, that’s probably for the better.

, , ,

Uber and the banks exercise the greenshoe to avoid becoming an IPO undercorn? #uber

Leave a reply

Who knows what forces were at play here during the first day of the Uber IPO, but this bump in the middle sure looks like someone was trying to get the Uber stock out of the rut – perhaps the investment banks exercising the greenshoe option? Cue the usual pictures of two big hairy animals fighting each other.

Screen Shot 2019-05-10 at 1.51.25 PM

Bloomberg reports that out of 207 million shares allocated for the IPO total, 27 million were reserved for greenshoe. I wonder how much of that was spent in the first day.

Musk says things that are directly opposite to the current facts #tesla

Leave a reply

I’m interested in EVs and have had a great experience owning one before, so naturally I’m watching Tesla with lots of interest. But this kind of stuff is just frustrating:

“If you buy a Tesla today, I believe you are buying an appreciating asset, not a depreciating asset,” Musk said during an interview with MIT research scientist Lex Fridman.

A Tesla will be worth $150,000 to $250,000 in 3 years, he claimed.

An actual report from the field:

It seems from the used market that the value has plummeted. The one I bought is easy $40k less just a year old.

Also, here is a site with all the latest snow reports and ski slopes updates for the upcoming season. Musk has pulled off some crazy shit before, will he be able to defy the current reality? It kind of looks like a matter of blind faith at this point. A quick Kelly Blue Book sanity check:

New 2019 Tesla Model S 75D $77,200

Used 2017 Tesla Model S 75 Sedan 4D $52,365 Private Party Value

Today, Model S value plummets like a dead bird. To me, it seems like a great toy for someone who has an extra $30k to piss away just to have some fun. But who knows? Maybe tomorrow will be radically different and Elon will be able to magically breathe some AI fairy dust over remote software update into all those existing cars. I guess the trick is to try to get a model and a set of options that will qualify for this utopian future upgrade.

Programmers wiped out by robots

1 Reply

There are 3.5 million truck drivers in the US. They will all be gradually replaced by robots in the next few years, and then we will have a massive bloodbath caused by 3.5 million (no offense) low-skilled, unemployed people hunting for food.

From their perspective, this may look unrealistic – heavy driving seems like a really taxing and complex activity: it demands lots of attention, it has many special cases, it often requires reading social cues from other humans that only humans can understand. Yet, the switch to robots (and the ensuing social cataclysm) look inevitable in short few years.

In my line of work, I’ve been long puzzled by a similar issue. Why in the field of computer science and engineering, which is ultimately governed by the physics of the circuits, switches, math and formal logic that stems from that, why so much of the routine daily work programming those soul-less machines involves so much of: human judgement, uncertainty, doubt, debates/flame wars over “patterns”, and straight up insane (for any other engineering line of work) rates of errors, crashes and bugs?

I mean, it’s hard to imagine a modern-built physical bridge that just crashes “because bugs”. Or take air travel: in the US, 0.2 deaths per 10 billion passenger-miles for the first decade of this century. For the whole second decade – 0 deaths (almost).

Modern planes are heavily computerized, is that software written by some kind of a different breed of humans than, say, web software? Because my online banking from Citi goes down every week, and it’s a freaking banking service with the same use cases since the 14th century, repeated many times daily by 7000 banks in the US, nowhere near as complicated as the software that’s automatically, near-flawlessly flying and landing the jets in this country.

But wait, the higher are the regular software stakes, the bigger seems the reliance on human judgement. “Our web service is facing unique performance challenges and only the team XYZ has enough experience to scale it to that level.” Mind you, human programmers still haven’t sorted out if something as long-used as OOP is even a viable concept – some say that it is an unmaintainable disaster, a non-starter, others have been building stable systems with it for decades. With stubborn opinions so polarized, mutually exclusive in a field, again, governed by the laws of physics, it seems kind of very wrong that so much of the programming and design is still performed manually by fickle, biased mortals.

I don’t see any replacement yet. But maybe I’m just a truck driver.

The end of #lendingclub’s most lucrative note grades

Leave a reply

The return on my portfolio with LendingClub – my choice was mostly low-grade notes – steadily dropped all the way down to 5.88% over the last year and a half. Until we finally run into this:

As of 11/7/2017, F & G grade Notes are not available for purchase by investors.

because LendingClub

noticed an increase in prepayment and delinquency rate in F and G grade Notes

Seems like the 20+% APY party came to an end. Even without any major event or recession in sight yet. My guess would be that with a spike in demand, LendingClub had to loosen up some criteria and let in a bunch of borrowers they wouldn’t and shouldn’t have otherwise, so now all of us are paying the price. Still might get much worse when the economy starts to tank.

Your #SSN DOB credit cards were leaked. Go do a credit freeze at all 3 agencies (FTC): https://goo.gl/RXZAVL

Leave a reply

Equifax Says Cyberattack May Have Hit 143 Million Customers (Bloomberg)

That’s 60% of all US adults, or pretty much everyone with a credit card. There is a big file floating around somewhere that has enough info on you for anyone to take out a bunch of money in your name and dump that debt on you. You can prevent that by putting a credit freeze on your personal info, which you can lift when you need to apply for credit yourself – here is an FTC link on how to do all three.

The most maddening thing about this leak is that – unlike Ashley Madison – this time you cannot opt out of this shit or “just stop doing it”. If you are a functioning human in USA, you are pretty much forced to store your most sensitive financial and personal info with these scumbags, and then they go on and leak all of it. I hope they get sued to death for being a bunch of incompetent de facto extortionists.

Of course, it’s another issue altogether that what used to be called impersonation and bank’s failure to verify its borrower, the same thing now is referred to by an idiotic oxymoron identity theft and somehow it became not the bank’s problem but yours.

In case you wondered why Travis #Kalanick got kicked out from #Uber

Leave a reply

Quote 1, June 9, 2016:

I say we are going to IPO as late as humanly possible. It’ll be one day before my employees and significant others come to my office with pitchforks and torches. We will IPO the day before that. Do you get it?

Quote 2, August 30th, 2017:

In response to a question about going public, Khosrowshahi said it would probably happen in 18 to 36 months, according to two people who listened to the meeting. “It’s my opinion that the company should go public,” he said.

Of course it’s your opinion, that why you just got hired!

Looks like the IPO will happen shortly after the VC’s came to TK’s office with pitchforks. Everything else was just clowning and smear campaigning by Benchmark to depose him. Everyone who jumped onboard the righteous moralizing Uber-bashing train should be really ashamed now. HR incidents happen at any company, everyone involved in them must be prosecuted to the full extent of the law. That doesn’t excuse a ridiculous anti-PR attack that almost took down the entire company. And even now it’s not clear if Uber will fully recover.