Andrea D'Ambrosio retweeted
Dude, what's going on with Private Mail?
I'm spending the whole afternoon setting up my email
Andrea D'Ambrosio retweeted
I want to be honest about why I disappeared
Building something alone means there are periods
where you have to go quiet
not because you're failing
but because the work demands everything
The last few months were that kind of period
I was heads down on DuckCode
A coding agent that runs in the terminal
Multi-model, MCP native, three distinct modes
But more than the features
I was trying to answer a question that kept bothering me:
Why do most AI coding tools make you worse at your job over time?
You stop thinking about the problem
You start thinking about the prompt
You become a reviewer of AI output
instead of an engineer who solves problems
I've felt this myself
And I didn't want to build another tool that does this
So DuckCode has Duck mode
It's a rubber-duck debugging protocol
where the AI doesn't write a single line of code
Instead it interviews you
asks you the right questions in the right order and guides you to the answer yourself
At the end you have a spec document saved in your project
And more importantly you understand your own code better than before
This is the thing I'm most proud of
Not the build mode
Not the MCP integrations
Not the multi-model support
The mode that makes you a better developer by refusing to do the work for you
I'm back
And for the first time in a while
I'm building something I'd use every single day myself
More tomorrow
Andrea D'Ambrosio retweeted
I'll fail a thousand times and start over a thousand times.
I've built so many products that have failed, do you really think I'll stop?!
Most AI coding tools are built around one idea:
do the work for you
And it works
Until it doesn't
Until you're six months in
and you realize you don't actually understand
the codebase you've been "building"
You understand the prompts
You understand how to ask
You don't understand the system
I've been that developer
It's a uncomfortable place to be
DuckCode is built around a different idea
There's a mode called Duck mode
Named after rubber-duck debugging
The practice of explaining your problem out loud
until the solution becomes obvious
In Duck mode the agent doesn't write a single line of code
It reads your files
It understands your context
And then it asks you questions
Not helpful suggestions
Not "here's what I'd do"
Just the right question
at the right moment
until you find the answer yourself
At the end of every session
a spec document gets saved in your project
Written by you
In your own words
Because you did the thinking
I know this sounds slower
It is slower
The first time
The tenth time you're faster than any developer
who outsourced their thinking to an AI
This is what I've been building
Not the fastest coding agent
Not the one that writes the most code
The one that makes you dangerous
even when you close the terminal
While I was gone I asked myself one question
What do I actually need when I'm stuck on a problem?
Not another AI chat interface
Not a copilot that autocompletes my mediocre code
I need something that lives where I work
Reads the actual files
Understands the actual context
And either builds with me or makes me think harder
So I built it
DuckCode: a coding agent that runs in your terminal
Three modes: Plan, Build, and Duck
Duck mode is the one nobody else is building
Instead of writing code for you
It asks you the right questions
Until the answer becomes obvious on your own
Rubber duck debugging
But the duck is a senior developer
who has read your entire codebase
3
8
471
npm install -g duckcode-cli
That's all it takes
A senior developer in your terminal
ready to read your code, search the web,
edit files and run commands
No setup
No API keys
Just install and build
More on this soon
I want to be honest about why I disappeared
Building something alone means there are periods
where you have to go quiet
not because you're failing
but because the work demands everything
The last few months were that kind of period
I was heads down on DuckCode
A coding agent that runs in the terminal
Multi-model, MCP native, three distinct modes
But more than the features
I was trying to answer a question that kept bothering me:
Why do most AI coding tools make you worse at your job over time?
You stop thinking about the problem
You start thinking about the prompt
You become a reviewer of AI output
instead of an engineer who solves problems
I've felt this myself
And I didn't want to build another tool that does this
So DuckCode has Duck mode
It's a rubber-duck debugging protocol
where the AI doesn't write a single line of code
Instead it interviews you
asks you the right questions in the right order and guides you to the answer yourself
At the end you have a spec document saved in your project
And more importantly you understand your own code better than before
This is the thing I'm most proud of
Not the build mode
Not the MCP integrations
Not the multi-model support
The mode that makes you a better developer by refusing to do the work for you
I'm back
And for the first time in a while
I'm building something I'd use every single day myself
More tomorrow
I've been quiet for a while
Not because I had nothing to say
Because I was building something
that needed all of me
Back now
7
319
I'll fail a thousand times and start over a thousand times.
I've built so many products that have failed, do you really think I'll stop?!
People haven't realized that by 2026, software development will account for just 1%.
The moat is what they should be focusing on.
2
9
351
Andrea D'Ambrosio retweeted
Most founders launch too early or too late. The real skill is knowing which product needs speed and which needs patience.
Sunday night
Two products live
Both moving forward
Both imperfect
Both real
TuBoost: creators are using it daily
Clips being generated from podcasts, vlogs, interviews
Real content going live on real channels
Cassandra: Telegram integration in development
Product Hunt launch waiting on that one feature
Because launching twice on the same product
means the second launch has to mean something
I could launch Cassandra on Product Hunt tomorrow
But the right moment is when the product
is exactly what I want people to find
when they search for it six months from now
Patience is a product decision too
Sunday done
New week tomorrow
Night
Andrea D'Ambrosio retweeted
Transcript-only clipping is like judging a song by reading the lyrics. You miss the emotion. We analyze the audio. That’s the difference.
Andrea D'Ambrosio retweeted
Every support conversation is a sales opportunity. Most tools only solve one. We solve both. Every question = every lead
Most founders launch too early or too late. The real skill is knowing which product needs speed and which needs patience.
Sunday night
Two products live
Both moving forward
Both imperfect
Both real
TuBoost: creators are using it daily
Clips being generated from podcasts, vlogs, interviews
Real content going live on real channels
Cassandra: Telegram integration in development
Product Hunt launch waiting on that one feature
Because launching twice on the same product
means the second launch has to mean something
I could launch Cassandra on Product Hunt tomorrow
But the right moment is when the product
is exactly what I want people to find
when they search for it six months from now
Patience is a product decision too
Sunday done
New week tomorrow
Night
Andrea D'Ambrosio retweeted
Every question unanswered at midnight is revenue walking away. You're not losing customers to competitors. You're losing them to silence.
The businesses losing customers aren't losing them at 2pm
They're losing them at 11pm
When someone lands on the site
Has one question
And nobody is there to answer it
The question doesn't go away
The customer does
This is the problem Cassandra was built for
Not to replace human support
To cover the hours humans don't
Andrea D'Ambrosio retweeted
Every AI clipper can find moments. Few can predict what your audience will actually click. That’s the difference









