Problem solver, independent consultant, software architect, on a quest to make data friendly and helpful.
Internet
Joined July 2012
- Tweets2.9K
- Following654
- Followers1.1K
- Likes2K
Norbert Orzechowicz retweeted
In 1994, software was dead. People told me programming was for nerds and that I should become a doctor or a lawyer instead.
In 2004, software was dead. People told me that someone from India would take my job for $5/hour.
In 2014, software was dead. People told me that no-code was going to kill every job and doom my career.
In 2024, software was dead with Devin, Windsurf, Copilot, and OpenAI o1.
We kill software every 10 years, and yet the demand for good Software Engineers has never been higher.
Yep, absolutely needed, wanted and with lot of class 🤡
Tylko ktoś kto nigdy niczego nie utrzymywał, ani nie kierował zespołem ludzi, nie rozliczał się z bakclogu, nie miał klientów na karku może się ekscytować niekontrolowanym push streamem do repo
Wewnętrzne przyspieszenie w Anthropic, które zepchnęło ich infrastrukturę deweloperską do granic wytrzymałości.
AI tak mocno przyspiesza produkcję kodu (w tym własnego oprogramowania Anthropic), że firma musi przebudowywać fundamentalne narzędzia, żeby móc iść jeszcze szybciej. Claude był zarówno przyczyną problemu, jak i głównym narzędziem do zbudowania Sciona.
Publicznie pace the frontier, wewnętrznie wymieńmy gita, bo Claude generuje za szybko :)
Well said! In the last couple months I tackled issues that were on my pile of shame for years. Yet, I don't feel like I have more time, opposite! I could use some help in fact as solving those old issues, opened new possibilities!
Unpopular opinion: In 5 years, we'll need more software engineers than we do today. Not fewer.
Everyone is calculating how many developers AI can replace. If a team of 3 can now do the work of 10, surely the other 7 are out of a job?
I think that completely misses what's about to happen.
For decades, the biggest constraint on software wasn't the lack of problems worth solving. It was the cost of solving them.
Millions of businesses still run on spreadsheets, emails, WhatsApp messages, and outdated software. Not because they're happy with it, but because spending $200,000 on a custom system never made financial sense.
Now imagine that same system costs $20,000. Suddenly, millions of businesses that couldn't afford software development become potential customers. Projects that were never viable become viable. Companies that already build software can afford to do ten times more.
And here's the part people underestimate: the cheaper software becomes, the more reasons we'll find to build it.
Of course, some jobs will disappear. Certain tasks that once required entire teams will be handled by a single engineer. But assuming the amount of software the world needs is fixed is a pretty strange assumption, especially in an industry where we've barely scratched the surface of what can be automated.
I could be wrong, but I think we're heading toward an explosion in software demand, not the death of software engineering.
We're confusing the declining cost of writing code with the declining demand for software engineers. They're not the same thing.
Claude might have all the answers but you need to ask the right question first - thats why you need to X years of experience.
- we removed unit tests in favor integration/functional tests
- ci/cd invoice exploded
- we removed cicd, it was slowing down our process
- now our agents deploys our code
- agent decided that database is going to be much faster without users
future looks awesome ☠️
We deleted 800k lines of unit tests from the @sazabi monorepo earlier this week.
Three hypotheses:
1- Unit tests "lock-in" slop code by making it hard to change or remove
2- Unit tests slow down dev cycles by making your agent do more work locally and by increasing CI times
3- Unit tests waste money by making your agent generate more tokens and by adding useless CI runs
Will report back soon with the results!
> more developers → more hours → more revenue
that was also part of the reason why IT salaries were so inflated, why juniors were sold as experienced seniors and why most of the modern world systems are 💩
what can go wrong this time?
I sold two software agencies. I wouldn’t build a third one the same way today.
AI is breaking the economics of selling developer hours.
For 20+ years, the formula worked:
more developers → more hours → more revenue.
Now one developer with AI can do the work that used to take a team.
I believe the next agency model is:
IP + AI + responsibility.
Build on existing software. Use AI to deliver faster. Charge for the problem you solve, not the hours you spend.
If you run a software agency: what happens to your business when one developer can do the work of three?
My substack is dedicated to this topic.
open.substack.com/pub/tomasz…
Your LLM slop generator vomited out some slop unit tests?
Of course the solution is to blame the testing methodology not your slop factory, prompted with "test coverage 100%, make no miasteks" 🤡
Fun times
> The best part is no part
Every component you add is something that can fail, cost money, or need maintenance.
Elon used to say it, and I fully agree, because
> Perfection is achieved not when there is nothing more to add, but when there is nothing left to take away.
here's how i shipped 2,500 PRs last month to production
this was originally supposed to be for Cursor Compile in London. i couldn't make it since i was livestreaming for Grok @Bot Galaxy so i'm making it available for free here on X! watch it on 2x speed, i talk slowly
Actually, AI is most beneficial once you can answer a strong "yes, I can" to all of those points.
Nowadays individuals can scale like never before, but only when they have solid foundations.
Otherwise it's a giant with feet of clay - grows fast, falls loud.
No, patterns, principles, and good practices are meant to keep code in good health, self-descriptive, maintainable, scalable, and predictable. Don't listen to ignorants; codebase quality is now more important than ever.
yes, don't be a moron and learn how to code, you cant be a manager, even agents manager if you don't understand what you are managing
Before we get too excited, lets wait for the whole AI business to "stablize" and give us the true price of tokens. Tiny detail that breaks all predictions 🤷♂️
Right now they are 🔥 💰 like nothing else before
Next release @flowphp is going to be fully schema driven! Historically extractors could yield different schema per batch.
Now all extractors are going to be self descriptive, meaning they always yields rows with the same shape.
Thanks to this we can know the output schema 👇
When dataset shape is predictable at each step, and with a efficient binary serialization (Floe) data frame execution can be parallelized!
Executor, based on Optimizer output is going to be able to decide how to break pipeline into steps and which of them can be parallelized 👇