GratefulPle₿ 🏴‍☠️⚡️

@webdevcody lol well if you don’t know the commands yes. Most ex-soydevs don’t for them the terminal is “new”. But if you know the command the reason to just run it is speed and typing less.

at this point there isn't a single terminal command you should be typing by hand

@dexhorthy @mattpocockuk well can' wait then!

@thegratefulpleb @mattpocockuk we talk about this on the great loops debate! youtube.com/watch?v=c35YoM…

Sometimes I experiment with going deep in the dumb zone to implement a feature Today it cost me 90 minutes of debugging to fix its stupid mistakes in a complex build Need to stop trying this

@dexhorthy @mattpocockuk So where does the dumb zone start with the latest models (gpt sol and opus 4.8)?

@mattpocockuk It’s good to keep in touch with the frontier…shape of the dumb zone changes on every new model. I still go to 300k+ just to confirm sometimes (and sometimes we can stay on track at 300k+ depending on the work!)

@mattpocockuk Yaya this all works great and is super cool and fine working alone for fun on some toy projects. And it ALL falls apart once you get a real job and work on code in a real corp that does anything slightly complicated.

1. Delete the docs you create to explain your code 2. Take the tokens you save on updating those docs 3. Spend them on making your code self-explanatory

@thdxr It’s all cool but all we’re waiting for is vim mode.

whenever we post any feature - x already does this, you copied it - i haven't heard of x - i look into x, it doesn't do it can you guys just enjoy the tools you use and stop being so lame with it

@thdxr Cool but does it have vim mode?

v2 has some nice stuff with background subagents and shell commands

@thdxr Does it have vim mode (even codex has one now)?

if you want to help beta test OpenCode 2.0 v2.opencode.ai - data in separate db which we might wipe - stuff will be broken - use /report to send us issues - v1 plugins won't work, v2 api not final there is a built in skill that you can ask for basically anything

@mattpocockuk Kind of interesting that setting ANTHROPIC_BASE_URL before running claude increases context by about 10k tokens, see /context all `System tools: 14.5k tokens (7.3%)` usually that's much less...

Here's a step-by-step process to kill all the bloat from your Claude Code system prompt: 1. Run a proxy so you can see exactly what gets sent to Claude Code (included in the article) 2. "Fuck, there is so much cruft in there" 3. Use my settings.json to kill all the bloat Down to a clean 13K tokens to start each session with. Nice. Full process here: aihero.dev/how-to-kill-th…

Remember that Theo writes react slop and has a team to clean up. If you write important software this doesn't apply.

you can really tell these people write nothing but crud react webapps "there's literally no need to read code anymore" meanwhile, "mythos-level intelligence" left on it's own:

@theo If you only merge such a small percentage of the written code but only review that much...how do you decide which 90% you throw away?

You should be reviewing a MUCH smaller % of your code today than you were 5 years ago. If your code is so important it needs every line verified, you better be writing a LOT of slop that verifies it too.

@webdevcody You’re a cool dude but did you read his comment…ya he meant you. 😅

to be honest, a lot of the code is not important to review at this point. a decent set of skills and periodic refactoring loops can make the code base clean, ai audits find more security issues than most devs can, performance can be iterated on using a loop.

@theo Why doesn’t this surprise me 🤔

I'll be honest, I barely even read the code back when I wrote it by hand...

@thdxr Great…so does it have vi-mode?

one of the reasons OpenCode 2.0 took so long was we redesigned it for hotreloading if you ask it to make a skill for itself (or make one manually) it'll get picked up immediately in a way that does not bust cache targeting public beta end of week, wish us luck!

@mattpocockuk yea it's crazy! Same with the ponytails-one...hype hype! Having said that I'm super curious what your global claude.md looks like ;)

The "X technique reduces tokens by Y%" fad is so old Can't believe people get taken in by this

@s4yonnara Nice you discovered that year old video. Welcome back. Let us know when there’s something new out there.

Andrej Karpathy spent 70 minutes breaking down how top AI users actually work with LLMs. The reality is simpler than people expect. You tell the model what you want in plain language and let it run. No 40-line system prompts. No secret tricks. By 2026 the engineer who writes off LLMs loses to the junior who just set one up properly. 70 minutes. Free. A rare straight look from an OpenAI co-founder. Bookmark it and watch.

@mattpocockuk Will this be part of grill me skill?

Two new leading words for planning work with AI: - Frontier - Fog of war "Don't plan past the fog of war. Let's resolve just the decisions at the frontier first." Great for getting aligned during a grilling session.

@mattpocockuk Nice! I think we need a video of you explaining the flow of this with an example. I kind of get it but also don’t really like how does it all come together at the end… 🤔

An absolute monster PRD, built out from a /decision-mapping session Love this new flow github.com/mattpocock/cou…

@mitchellh Interesting approach. Two things I’m not sure I grasped: 2. … Simultaneously, do this yourself. Means to split up the task myself? Why? 5. … repeat the "draw the owl" prompt. Why do that? Isn’t it done at that point (all sub tasks have been implemented)?

My heuristic is that any diff an agent generates over ~1500 lines is too big and is indicative that the problem needs to be decomposed. This is my general pattern now for feature work: 1. Try to implement the whole feature, loosely guided. I call this the "draw the owl" prompt in reference to the meme. Expect garbage, you're going to get garbage. 2. If the diff is less than 1500 lines, review it and iterate normally. If the diff is more than 1500 lines, prompt the agent to decompose the problem into atomic, incremental, reviewable tasks. Simultaneously, do this yourself. 3. Agents will very often make these tasks way too specific to the shape they solved. You need to massage it into the right general shape. Do that. 4. Kick off new agents to work on those incremental things (as parallelized as possible). Apply the same rules. 5. At a certain, point, repeat the "draw the owl" prompt. At some point, you will get beneath your review-ability threshold. This has been producing consistently high quality, maintainable, reviewable chunks of code that have a good handoff to either merge as-is or human refinement. And with the latest frontier models at xhigh thinking, these are all slow enough that you can usually have multiple going concurrently while you are actively reviewing others or working on your own tasks. HITL (human-in-the-loop) agents are still super important, especially for feature work. Features touch the human boundary in terms of UI, API, etc. And net new stuff can introduce pathologies in the architecture that violate desired invariants (these should be represented in specs or tests but we aren't perfect!). I know a lot of the leading edge agentic discourse is about "loops" and agents driving agents continuously. I do some of that (will report on that later). But, in terms of raw daily get-shit-done type of work, this is my most rewarding pattern at the moment.

@ThePrimeagen tell me more...

I think I have a new vim LLM idea I think this one might be the best. I know a lot of plugins, but this one is the best