@makowskidi
iAccount based inSingapore
About this account
- Account based in
- Singapore
- Connected via
- Thailand Android App
Account-level information from X, not a live location or the device used for a specific post.
I help companies automate their processes with AI 👉 https://nitter.cf/t.co/S9gXnsLcVM https://nitter.cf/t.co/YOfaPXz9G8 https://nitter.cf/t.co/Z6WCfYCc7o https://nitter.cf/t.co/SJduUCeiwQ
Planet Earth
Joined September 2011
- Tweets7.4K
- Following391
- Followers1.4K
- Likes44.3K
Dawid Makowski retweeted
This week a friend of my, also an experienced coder, testing my newest product told me it was his first experience with a product fully written by AI agents, and that it was much better than he expected.
The numbers: first commit on 17 July, 600 commits since, about 2.5 months of work. Still buggy in places. Written by agents, but guided at a very low level.
He asked how I coordinate the agents, feed them tasks and verify their work. The baseline:
- Playwright with browser access, so the agent can click through what it just built
- The agent runs its own tests all the time: smoke tests along the way, the full suite before anything bigger
- It has to follow the framework's guidelines, plus my own written guidelines on top that grow every day
No secret sauce. The devil is in the details, and for me the biggest detail is templates.
An LLM is a probabilistic text generator built on patterns from its training data. The more templated your input, the closer the output lands to the template. So I very often point the agent at another project of mine and tell it: implement it exactly like this.
My templates are often 10 to 20 times bigger than the output I expect. On top of the template go detailed instructions. It's "Don't make me think", applied to the agent instead of the user.
The rest comes from hundreds of hours of talking to Claude, until you know instinctively what works and what doesn't. It's like any other muscle in that sense.
Dawid Makowski retweeted
Hi!
I'm 44,
CTO, coder, SaaS builder from Thailand/Singapore. Originally from Poland.
Looking to connect with more builders, digital nomads & indie hackers!
Let's surf this connection algo trend as much as X algos allow!
Cheers!
Hi!
I'm 44,
CTO, coder, SaaS builder from Thailand/Singapore. Originally from Poland.
Looking to connect with more builders, digital nomads & indie hackers!
Let's surf this connection algo trend as much as X algos allow!
Cheers!
Dawid Makowski retweeted
Most rewrites aren't caused by bad code.
They happen when the business changes direction and the architecture can't follow. A product built for one country goes multi-currency. A B2C app suddenly needs company accounts. A one-off feature becomes the core.
a
When I review an early-stage system, I care less about how clean it is and more about how many of those turns it could survive without starting over.
Most rewrites aren't caused by bad code.
They happen when the business changes direction and the architecture can't follow. A product built for one country goes multi-currency. A B2C app suddenly needs company accounts. A one-off feature becomes the core.
a
When I review an early-stage system, I care less about how clean it is and more about how many of those turns it could survive without starting over.
Something I didn't expect from working with coding agents: technical debt now has a second cost.
Agents copy whatever patterns they find in your codebase. If there are three ways of doing the same thing, you get a fourth. If there is one consistent way, you get that way again, almost every time.
A boring, consistent codebase has become the best prompt you can give an agent.
Love it :)
Usage-based APIs should have hard spending caps by default.
A midnight “budget exceeded” email is useless if a service keeps burning money until morning.
When the budget runs out, should requests fail or should the provider keep serving them and charging?🤔
I’d take the error over a surprise $15,000 bill.
This week a friend of my, also an experienced coder, testing my newest product told me it was his first experience with a product fully written by AI agents, and that it was much better than he expected.
The numbers: first commit on 17 July, 600 commits since, about 2.5 months of work. Still buggy in places. Written by agents, but guided at a very low level.
He asked how I coordinate the agents, feed them tasks and verify their work. The baseline:
- Playwright with browser access, so the agent can click through what it just built
- The agent runs its own tests all the time: smoke tests along the way, the full suite before anything bigger
- It has to follow the framework's guidelines, plus my own written guidelines on top that grow every day
No secret sauce. The devil is in the details, and for me the biggest detail is templates.
An LLM is a probabilistic text generator built on patterns from its training data. The more templated your input, the closer the output lands to the template. So I very often point the agent at another project of mine and tell it: implement it exactly like this.
My templates are often 10 to 20 times bigger than the output I expect. On top of the template go detailed instructions. It's "Don't make me think", applied to the agent instead of the user.
The rest comes from hundreds of hours of talking to Claude, until you know instinctively what works and what doesn't. It's like any other muscle in that sense.
Dawid Makowski retweeted
Hi!
I'm 44,
CTO, coder, SaaS builder from Thailand/Singapore. Originally from Poland.
Looking to connect with more builders, digital nomads & indie hackers!
If you have deployed any online product - let's get followed back together and track our progress.
Let's surf this connection algo trend boost as much as we can folks!
Cheers!
ps. shoutout to @compai - my SOC 2 cert partners :)
Hi!
I'm 44,
CTO, coder, SaaS builder from Thailand/Singapore. Originally from Poland.
Looking to connect with more builders, digital nomads & indie hackers!
If you have deployed any online product - let's get followed back together and track our progress.
Let's surf this connection algo trend boost as much as we can folks!
Cheers!
ps. shoutout to @compai - my SOC 2 cert partners :)
Dawid Makowski retweeted
On this day in 1608, Poles arrived at Jamestown, Virginia - just a year after the colony itself had been founded.
Think about how remarkable that is. More than four centuries ago, at the very beginning of permanent English settlement in North America, Polish craftsmen were already there. They had been recruited for their specialized skills and helped produce glass, pitch, tar, soap ash, and other valuable materials needed to make the struggling colony economically viable. Captain John Smith himself specifically praised the work ethic of the Polish and other continental craftsmen.
They were not simply passengers or bystanders in the story of early America. They were skilled workers helping build it. Historical records even describe Polish settlers coming to John Smith’s aid during a violent encounter, and later protesting when they were denied the same political rights enjoyed by English colonists. Their protest succeeded: colonial authorities recognized their rights, and the Poles returned to work and agreed to teach their crafts to others.
It is extraordinary to realize that the Polish-American story reaches back to 1608. Long before the great migrations of the nineteenth and twentieth centuries, long before Pulaski and Kościuszko became names associated with American independence, Poles were already crossing the Atlantic and contributing their knowledge, labor, and courage to a new society.
(Encyclopedia Virginia)
Polish history in America did not begin yesterday. It began almost at the beginning of America itself.
Dawid Makowski retweeted
The interview question that tells me the most: tell me about a project that went wrong and what part of it was your fault.
Some people describe the failure in detail and own their part. Some explain why it was everyone else.
I've rarely regretted hiring the first kind.
The interview question that tells me the most: tell me about a project that went wrong and what part of it was your fault.
Some people describe the failure in detail and own their part. Some explain why it was everyone else.
I've rarely regretted hiring the first kind.
Dawid Makowski retweeted
I do technical due diligence for investors before they put money into a company.
Founders expect me to look at the code. The code matters less than they think. What I really want to know is who could rebuild this if two key people left tomorrow, and how long it would take.
A lot of startups have great code that only one person understands. That's not an asset. It's a risk with a salary.
Dawid Makowski retweeted
I run a software company that builds for clients (A2Z Web). On the side I build my own products: @SharpAPI, BrandGEO, Legalnaut.
Every client project teaches me what people will pay for. My own products teach me what it feels like to be the one paying for every bad decision.
The second lesson has made me much better at the first job.
I do technical due diligence for investors before they put money into a company.
Founders expect me to look at the code. The code matters less than they think. What I really want to know is who could rebuild this if two key people left tomorrow, and how long it would take.
A lot of startups have great code that only one person understands. That's not an asset. It's a risk with a salary.
I run a software company that builds for clients (A2Z Web). On the side I build my own products: @SharpAPI, BrandGEO, Legalnaut.
Every client project teaches me what people will pay for. My own products teach me what it feels like to be the one paying for every bad decision.
The second lesson has made me much better at the first job.