https://xitter.cf/t.co/JvZNNTQWBB · construo o que preciso, mostro o que aprendo. AI até os tokens acabarem. 🇧🇷🇵🇹
Portugal
Joined February 2017
- Tweets874
- Following886
- Followers136
- Likes2.7K
Unpopular opinion: most coding agent "memory" is just stale context with extra steps.
I'd rather have a fresh session and a short notes file in the repo.
Do you let your agent remember across sessions, or start clean every time?
Every other post now is someone running 300 Claude Code agents in parallel.
Honest count. How many agents do you actually run at once?
1
2 to 3
10+
swarms are mostly tokenmaxxing
Free Flash models with 200k context keep showing up in coding agents.
Hot take: most day to day coding doesn't need the smartest model. It needs the fastest one.
Where does a Flash model still fall apart for you?
Every week there's a new open model that runs agents on a laptop.
Would you actually let one work on your real repo for a week, or is it still frontier only for anything serious?
Pedro Santos retweeted
We've been working hard on making the Claude Code desktop app great for running lots of sessions side by side.
It's now much quicker to start sessions, easier to split work across a few, and you can always see what each of them is doing. Here’s what's new:
Google AI Pro here and Gemini 4 Argon shows up as Ultra only 😅 How are you all using Argon on Pro? API, AI Studio, something else? @OfficialLoganK any chance Pro folks get a taste soon?
If I can't explain what the agent just wrote, it doesn't get merged.
Even when I'm the only reviewer.
When your coding agent wants to run a shell command:
full auto, let it cook
approve every command
do it yourself
What's your default?
Honest question.
Your coding agent gets stuck in a loop. Same bug, third try. What do you do?
one more prompt
switch model
take over yourself
new chat from scratch
No judgment. Just curious what people actually do.
Honest question.
AI just shipped most of a feature. Who writes the tests?
you
the same agent
a second agent
nobody (fix in prod)
No judgment. Just curious what people actually do.
Honest question.
When an agent drops a big diff, do you:
read every line
skim the risky bits
ship and fix later
No judgment. Just curious what people actually do.
Cursor vs Claude Code vs Codex CLI.
What's your daily driver right now?
(If you hop between them, what's the split?)
Which AI sub are you actually paying for right now?
Not the one you tried once. The one that stayed.
Wow 🤩
Introducing Projects, a new way of working in Cursor.
Rather than creating a chat for every task, you work with a coordinator agent in a single, persistent thread.
Like @bot, your agent is always on, proactively manages work with subagents, and improves over time.