Pim Elshoff * The Entrepreneurial Employee

Coding is hard. Save mental energy, run your tests often 👇 ❌ Type, type, type, read, stare, think, type, type, read, stare... ✅ Type, run, read, type, run, read ‼️ Compiling is automated; don't compile! ❗ Slow down when test failure is unexpected; read errors, not code

Update: it read to 75% session used, made two passes at the code, then started compacting (i.e. forgetting everything it loaded) and then raced right back to 98% trying to remember SOMETHING. And then 100%! I got two minor tasks done. Absolutely worthless. @claudeai

Two weeks ago Claude Sonnet 4.6 could do several tasks in one 5hr session budget in a brownfield, very imperfect repo. Today, Claude is spending 45%(!) of the budget just to read the project... that it started the night before. I'm at a loss here 🤷‍♂️

@NoahRevoy Thanks Noah, much appreciated!

@Pelshoff Wherever you go and wherever you land, you will bring tremendous value and be a major asset to those you work with.

After four years of various adventures, I'll be leaving Superscript. I'm working through April, will take most of May for other work, and then will be available for freelance work come June. Link in comment 👇

Code quality won't save you from bugs. The most beautiful code is still littered with coupling issues via over-investment in abstractions and DRY and virtually no OCP adherence. The solution to maintainability always has been at design quality - AI or not. Design quality and code quality are not the same thing as many are screaming at you to keep you convinced for many selfish reasons. If you fail to couple over an immutable source of truth and observe OCP for new features by copying then adjusting for perfect solution fit, then you haven't solved maintainability issues. You can have the worst looking code absolutely wipe the floor with the crispiest, cleanest, jaw-droping code you've ever seen. I'll take that bet every day. The industry is in a wild goose chase to make amazing code, fast; and that's what's burning all the power for AI, tax money for go no where death march projects and resource (people) Tetris for management in the dilbert land of enterprises. The cancer treatment pharmaceutical didn't want the tobacco manufacturer to be out of business much like your ide manufacturer didn't want you to stop DRY and chasing pretty code. @EventModeling #EventSourcing.

FFmpeg is moving to Rust 🦀 Our use of C and Assembly in FFmpeg has been an unacceptable violation of safety. FFmpeg will be running 10x slower - but we're doing it for your safety. All your videos will appear green - safety first, working software later.

From startup to shutdown, loading Windows 3.1, MS Word, and MS Excel, on a computer several orders of magnitude slower and with 1000x less RAM than a modern PC. Note how snappy everything is. iT's dOinG mOre No it isn't. We don't have to live like this!

My kids are four and one, and I already experience this. Cherish every little thing they do, before it's just a memory!

Nederland heeft in slechts één maand tijd 130 miljard euro aan ongerealiseerde winst gemaakt op het Groningse gasveld. Ik vind op zijn minst dat we daar 36% van mogen zien als grondstof dividend dit jaar. Toch mooi weer zo'n 2600 euro p.p. Box3 drama 🔂

@NoahRevoy There's never been a better time to have ideas

My wife created her first client website in 1998. It was a product catalog with several hundred products. It took her hundreds of hours because everything had to be hand coded in HTML, and all the images had to be resized in Adobe. She charged $150. Today she could build the same website with Claude CLI in about 30 minutes and charge $3,000.

oog can't sleep. walk outside cave. middle of night. papa already there. sitting. looking at nothing. oog: "papa?" papa: "sit." oog sit. long quiet. bug noise. sky big and close. oog: "papa can't sleep?" papa: "papa just thinking." oog: "about what?" papa: "cold season coming. need more hide. need more dried meat. fire pit cracking, need new stone. koom need new wrap, growing too fast. fru need new spear, old one splintering." papa say all this flat. like reading from wall scratching. not complaining. just... listing. oog: "papa think about this every night?" papa look at oog. little smile. tired one. papa: "papa think about this every MOMENT. just only say it out loud at night when nobody listening." oog: "oog listening." papa: "...yeah. oog is." quiet again. oog: "papa. when does it stop? the list?" papa laugh. real laugh. the kind that surprise him. papa: "oog. it don't stop. you just get better at carrying it while smiling at breakfast." oog sit with this. heavy thing disguised as a sentence. one day oog will have own list. own night sitting. own bug noise and big sky and a head full of everyone else's needs. and oog hope... oog hope someone sit next to oog and say "what papa thinking about." so the list can live outside oog head. even for one minute. that what papa needed tonight. not solution. just a witness. love, oog

The insane 36% unrealized gain tax may have passed in the house, but the law still has to pass the senate. This website makes it incredibly easy to e-mail a bunch of senators & make your voice heard. Do your duty, and send at least 5 today. 👇🏼 vermogensval.nl

@NoahRevoy Trust me plenty of programmers too 😂 one of the reasons why software grows too complex

I specified things like every possible client state and a fairly complex database schema so that the user experience would be smooth all the way through the program. That section alone was probably four or five pages long. A significant portion of it, maybe a quarter to a third, was devoted entirely to what I did not want the AI to do. That is something I probably would not have had to spell out for a human developer, but it turned out to be essential when working with an AI.

I built an entire SaaS product that is roughly comparable to Kajabi in less than a week using Claude Code in the terminal. It includes features Kajabi does not have that I specifically need, and it omits features Kajabi has that I do not. I finished it, ran it locally, and while reviewing it, realized I wanted a few additional features. I described them to Claude. Claude thought about it, proposed an implementation, I agreed, and it built the features. Fifteen minutes later, I was testing them. If I ask a traditional SaaS company for a feature I need, the odds of it being implemented are close to zero. The odds of it being implemented on a timeline that matters to me are effectively zero. In that model, my product has to adapt to the platform. Here, the platform adapts to my product. I do not know whether every reader could get the same results. I am not a programmer either. But I am very good at describing exactly what I want, and just as importantly, what I do not want. In this case, I gave Claude a governing document describing the business logic, followed by a nine step development plan. Each step was several pages long, and Claude was instructed to work through them one step at a time. If you can clearly describe what you want, how it will be used, and how your business logic works, this approach can produce extraordinary results. If you go to Claude and say, “Make me a website,” it will make you something. It will probably look good. But it will not necessarily be what you want. This method only works if you are good at thinking clearly and describing your intent precisely.

@NoahRevoy Btw the logic specification problem is mostly solved by the event modeling movement, which aims to turn software engineering into an actual engineering discipline and not a craft

@NoahRevoy That's an extensive specification! I've always seen my job as helping people discover the real spec; implementing is a necessary evil. Many programmers need to get real with what their added value is before it's too late..

Het box 3-plan is een Orwelliaans gedrocht. Vermogensaanwas is (deels) een eufemisme voor geldontwaarding. Overheid voert beleid om inflatie te veroorzaken. Logischerwijs stijgt de prijs van je bezittingen. Overheid: Aha! De ‘waarde’ van je vastgoed, goud, aandelen is gestegen, betalen! Ze doen alsof elke koersstijging vermogensaanwas is. Dat is het echter pas écht nadat de inflatie is goedgemaakt. (En dan niet de CPI-inflatie, want we beleggen niet voor potten pindakaas.)

A guy who worked for TI back in the day told me that after they created the first electronic hand calculator, they got a call from Bell Telephone. The folks from Bell sounded uncertain and worried. You see, when they were creating the first push-button phone, they did extensive tests to determine the exact best pattern for the numerical buttons - the pattern that would be easiest for people to tap in the dark or by muscle memory. But now they were concerned they’d completely blundered, because they saw that TI’s machine had the exact opposite layout. “So,” they asked. “What did your tests show?” The guys at TI just stared stupidly and said “Tests?”

Software engineers have negated every advancement in transistor density, computer architecture, compilers, and computer networking over the past 30 years.