Dragoș Mușetescu

I openly share my journey and lessons learned in building and growing businesses, with the hope of supporting and inspiring fellow entrepreneurs. #buildinpublic

[Dragoș Mușetescu @dragosdm]

I'm thrilled by the positive feedback on my efforts to enhance our hiring process. Today, I finally wrote an article about it. What are your thoughts? Any areas you think I should delve into more deeply?

https://t.co/mAc6FZ0gYm

[Dragoș Mușetescu @dragosdm]

@snajper47 Have you tried pi+gema? A friend of mine got 60tps on a m1 max mpb with 64gb ram

[Dragoș Mușetescu @dragosdm]

I love pi and after I saw @lucasmeijer love letter to pi I switched to html as output whenever I needed it.

[Lucas Meijer @lucasmeijer]

Been screaming this from the rooftops :)

[Dragoș Mușetescu @dragosdm]

@mitsuhiko @trq212 @lucasmeijer Yes! Seeing @lucasmeijer talk made me use HTML as well! ❤️

[Dragoș Mușetescu @dragosdm]

Have you tried pi? I was a long term Claude user and switched a few months ago. I find myself opening Claude less and less.

How’s the in-terminal browser? I’m constantly switching from https://t.co/oZzHizzY6G to ghostty 😅. I have disabled all ai features in warp and since they’ve added side tabs it made a huge difference for my workflow

[Dragoș Mușetescu @dragosdm]

@businessbarista AI-native

[Dragoș Mușetescu @dragosdm]

@amix3k @mitsuhiko Thank you, so much, Amir!

[Dragoș Mușetescu @dragosdm]

@amix3k @mitsuhiko @amix3k do you have any plans on open sourcing the handbook or parts of it?

[Dragoș Mușetescu @dragosdm]

@Jason I’m in

[Dragoș Mușetescu @dragosdm]

@amix3k I think it is very important to make sure that we do not outsource our thinking so that we can make sure we maintain and build: taste, judgment and trust.

[Dragoș Mușetescu @dragosdm]

Success with AI will not come from simply picking the best model. It will come from building the best workflow around the model.

Day 2 had a lot of great talks, but the common thread was clear: AI coding is becoming a systems problem, not just a model problem.

[Dragoș Mușetescu @dragosdm]

Day 2 at @AIEMiami left me with one big takeaway: AI coding is no longer just about model quality. We’re moving from “LLMs that help write code” to “agent-driven software systems”.

[Dragoș Mușetescu @dragosdm]

Our products now need to work for agents too.

That means:
- APIs
- auth
- docs
- CLIs
- interfaces

All need to become more agent-friendly.

In many cases, your next user may not be a person clicking a dashboard.

[Dragoș Mușetescu @dragosdm]

@mitsuhiko shortcut to copy the last model message to clipboard

[Dragoș Mușetescu @dragosdm]

Summary of Day 1:
Agents scale execution. They do not scale judgment. Which means the winning teams will be the ones that rebuild structure on purpose: pipelines, guardrails, review, persistence, and restraint.

[Dragoș Mușetescu @dragosdm]

@aiDotEngineer Miami Day 1 had one clear message: The future of agents is boundaries, not authority.

Best ideas I heard:
- code > tool calls
- pipelines > mega-prompts
- skills/markdown > bloated abstractions
- agents should author artifacts, not touch prod
- secrets should never reach the model

Agents scale execution. They do not scale judgment. That’s now our job.

[Dragoș Mușetescu @dragosdm]

The product lesson was maybe the most important:
AI removed the engineering friction that used to kill bad ideas early. Teams can ship bad ideas much faster.

Execution scaled. Judgment didn’t.