Ben Freed

hi frens! I published my Claude Code workflow system as a package called Kata! github.com/codevibesmatte… This system integrates fully with CC's native task system, no beads (though its an amazing project) or other external task managers needed. Also, it relies heavily on hooks and has a pretty ingenious stop hook mechanism that is fully flexible and changes based on what mode you're in. The number and type of modes and individual mode instructions are fully customizable and the package has examples and a good set of starter modes. I'll post more when I get a chance but the main reason i built this is because a lot of systems felt too heavy handed or black-boxy for my liking. Anyway check it out and let me know what you think!

@Trace_Cohen Flushing

What’s your favorite NYC food secret or cheap eat!?

@sinasanm They didn't run it on their cursor bench

is composer 2.5 10% or 10x better than kimi k2.6?

New Gemini pro app is so bad. I had it create a cheat sheet in Google docs with all my upcoming travel itinerary then when I asked it to edit it it said that it can't create Google docs and claimed that it had hallucinated the doc that it actually did create.

Ok the Theo podcast is really good

@robinebers I'll be trying the SDK out. Not particularly optimistic

@codevibesmatter 🤣 I always try them gotta give them a fair chance but this ain't it

Anigravity 2 is dead-on-arrival instantly uninstalling

@hunvreus @enjoyingthewind You find a happy middle. Spec it with a few questions answered and then get to coding fast

Why on earth would you want to revert to Spec-Driven Development? Yes, agents are way faster at writing code. And (some) humans are better at system thinking. But we also suck at planning. Any experienced engineer knows you simply cannot sit down, write the specs and then write the software that matches it. At least not if you plan on writing something "good". You need to work through the problem to understand its boundaries and shape a solution that makes sense. Just leverage the fact that writing code is cheap: 1. Prototype, 2. Document learnings, 3. Rewrite based on learnings, 4. Document solution, 5. Refactor, 6. Document changes. Even if you have to repeat parts or all of this, you'll get to a good solution faster than with SDD.

@toddsaunders @conductor_build Wait multimodal and multi-model are also different

I met someone tonight at an AI event that told me they were bilingual. I then went on a tangent about how I've been using @conductor_build and it's made multimodal. And how I've been using Codex and Opus together a ton. Turns out.. multimodal and bilingual are not the same.

WHAT

@kunchenguid @leo_linsky synesthesia is the unlock!

@leo_linsky hard to say sometimes i feel thinking in words is really inefficient great thinking can happen directly in the latent space - very often when we suddenly feel inspired with a great idea, it didn’t really come from a big monologue

Since thinking tokens led to a step-function improvement in AI reasoning, does that mean people with no internal monologues are worse at reasoning?

@toddsaunders Brilliant. Signals ingestion highly underrated

A close friend just showed me the best AI workflow I've ever seen. His vertical SaaS went from $1.2M to $5.5M ARR this year. He says this is why. It all runs on Facebook. Facebook has no newsfeed API, they use Cloudflare to stop scrapers. So he runs a browser harness, signed in with his own account, and sweeps four competitor groups and two industry groups every week. Every new post gets screenshotted. - GPT 5.5 reads the post, categorizes it, and puts all of the information into a table. - Sonnet 4.6 does the triage and the labeling. Things like competitor type and signal (complaining, feature request, churn, pricing). - Opus 4.7 does the synthesis and the weekly cron job that fetches the post. It then finds patterns in the posts and writes briefs that go to Slack every Monday morning. Then, every post is automatically turned into a markdown file, and is automatically action against. Feature request / product complaint - Opus creates that bespoke feature in real time, and takes a video recording of final product for the product team to post. Questions / industry insights - Opus creates a SEO optimized blog post, publishes it, and gives the link for the marketing team to post. His competitors' customers are writing his roadmap. His competitors' weaknesses are writing his content calendar. It's absolutely wild.

@kunchenguid Yassss

i’ve been in the front row seat of tech companies’ AI adoption and layoffs should i make a post / video to explain what’s happening? if so, reply and let me know what you are most interested in hearing about

@businessbarista Taste issue

AI slop should be called Human slop. It is almost always a human skill issue, resulting from bad direction, bad context, and bad taste.

@kunchenguid I dunno man I've been to some where they insist on cooking for you

you can outsource your thinking you can even outsource understanding you can’t outsource Korean BBQ

@quorralyne Cool I'll check it out!

Agreed. This is why we created Meko. If you use hooks with Claude Code, the entire: 1. conversation raw text 2. graph memory 3. flag for what is important to remember in the future Automatically added into Meko for agents to access. mekodata.ai

@GeeksMirage @trq212 Then it would be much better to use reasoning traces for that

@codevibesmatter @trq212 You’re assuming that this is to track progress, which it is not. It is used to track historical decision flow, so that you have a log of WHY the LLM made decisions, so that you understand what happened and can improve it if needed in the future.

a prompt I've been using a lot recently: implement <SPEC> and while you do, keep a running implementation-notes.html file (or markdown) with decisions you had to make weren't in the spec, things you had to change, tradeoffs you had to make or anything else I should know

@aashutosh_01 @trq212 github.com/codevibesmatte…

@codevibesmatter @trq212 Curious, what flow are you using to make thia happen?

@oussemadb @trq212 It doesn't have to be perfect but it should be more than a prayer

@codevibesmatter @trq212 I dont agree tbh cuz waiting for the perfect harness means waiting forever. A simple notes file you can read and act on today beats a theoretically elegant system that doesn't exist yet. Pragmatic always beats ideal in a moving field.

@kupolov @trq212 Definitely the latter. you need more of muh harness

@trq212 @codevibesmatter You guys should decide: is AGI coming soon, or is the current iteration of AI just the next level of abstraction -like the move from machine code to compiled languages like C? Dario is claiming devs aren't needed anymore.

@southphxceleb No it's been the way llms work since 3.5. it's just not the focus of the harness makers because it requires a very opinionated structure

I feel like as the models get smarter they’re making more nebulous decisions, introducing a “worklog” is an understandable iteration on this *brand new* way of building