Capric

Exploring human-agent collaboration mechanisms.

[Capric @CapricG_]

@rezkhere This is so cool. Do you plan to build something for video ads?

[Capric @CapricG_]

@_halshin @GergelyOrosz @OfficialLoganK Genius

[Capric @CapricG_]

I really miss Opus 4.6

[Capric @CapricG_]

Created a product intro video for https://t.co/gmdpBVqdLz entirely with code-based animations.

[Capric @CapricG_]

My first product video purely by coding. In this video, I succeeded in

1. making a close simulation to the real ui/ux of my product Mooon
2. syncing between audio, subtitle and the visuals
3. creating a framework that defines interactions and camera control with complex ui stackings

Then I used Remotion to orchestrate them

Most cases with coding animation out there are so slide-like and hardly produce a complicated use case. This is an exciting one that at least proves its potential!

[Garry Tan @garrytan]

“Skillify it” is the way you should write most tasks

Write markdown that makes code

Don’t make elaborate Foxconn factories to call agents

Let agents make their own tools, kaizen style

[Simone Conradi @S_Conradi]

Another one!
Made with #python #numpy #matplotlib and #marimo (and also #wigglystuff)

[@levelsio @levelsio]

So telling your story on X (and other social media) is essentially free user acquisition

But you pay with your personal brand (and time)

The other options are SEO or paid ads

SEO to rank on Google (but with AI that's getting harder, less traffic)

Paid ads you need money!

[Capric @CapricG_]

I built myself as a creator on rednote years ago to share about Latin and made some audience

But when I started a Japanese PDF reading product last summer, I was worried to brand myself and wanted to avoid personal attachments from it. Then I created a new account for the product. It has gained over 10k followers so far, but I realize none of these followers will become my supporters for any of my future businesses.

Audience was built but just for one-time trust. It's a huge waste. Never too late to restart. Good luck!

[Capric @CapricG_]

Don't build systems too early. Many people, including myself, tend to summarize a workflow or skill from a just-runnable project for hypothetical future reuse.

Very likely now is the only chance to follow it

[Capric @CapricG_]

@Prathkum So true to learn programming. Vibe coding makes everyone build a 60-point products with no effort, but to get 90+ technical literacy is required. Learning a programming language can at least help build the sense

[Capric @CapricG_]

Token leaderboard makes a bit sense but doesn't solve the whole problem. The purpose is to evaluate value per token. The hypothesis is all work should be creative and exploring beyond the job requirement

A good practice could be doing a regular output review discussion and giving conclusive judgments to each team member. Otherwise most tokens are burnt for nothing.

[Capric @CapricG_]

Agentic coding is my daily driver now, but kinda scared my skills are rotting

The productivity boost is insane but i'm forcing myself to code manually sometimes just to keep the muscle memory alive

Weird feeling when the tools make you faster but you worry you're forgetting how to actually build

[Capric @CapricG_]

Got a vintage hand-carved wooden Ravana from Varanasi in Watkins Books!

[Capric @CapricG_]

@soloshipped Try switching to a different cc client (terminal or ide extension). It might work. I found the terminal client to be more problem-persisting sometimes

[Capric @CapricG_]

Is Claude Code down?

[Capric @CapricG_]

seems to be an issue related to opus 4.7. it recovers for me now

[Capric @CapricG_]

Experiment in one, then build a new one.

I'm exploring a programmatic animation framework with my agent - think code-driven motion graphics without touching After Effects.

First, I designed v1 framework and used it to mock an existing demo video. Simple goal: to prove it works by recreating an existing work.

Then, I let cc be creative and build anything using v1. This is where it got interesting - the result reveals unanticipated holes in my system. Edge cases that required massive restructuring and overturned numerous initial hypotheses

In this process I built a deeper understanding of what the system actually needed versus what I thought it needed. So I put v1 aside and started v2 with a fundamentally better design.

This works much better than diving in and handling endless restructuring on the first one

[Capric @CapricG_]

The more I use AI to build, the more I realize technical literacy matters in a different way now.

Not to write every line of code, but to know when the generated system is becoming something you no longer understand.

[Capric @CapricG_]

Ask your agent to prototype what it looks like on the attached page If you're building a browser extension. Helps a lot