dools

Yeah deepseek is my workhorse of choice these days, I mostly use pro because I do a lot of concurrent jobs so speed is less of a concern. When it falls over I change the model mid-chat to GPT5.5 and chuck a couple of tokens over to OpenAI and then once it's correctly found the problem I switch back to deepseek, keeping all the 5.5 analysis in context.

I have been a heavy user of k2.5/6 but they seem to have gotten slower and worse at their jobs (and the "engine overloaded" errors have increased ... ), and k3 spends a LOT of time thinking and doesn't produce noticeably better results. I do keep chucking some tasks over to moonshot every now and again to give it a chance. I figure they're under compute pressure and have probably quantised the models to cope temporarily.

But recently I've found myself mostly using deepseek v4 pro and gpt 5.5 when I need the big guns.

GLM 5.2 has been good at select tasks but really shit at others and given how sparingly I use GPT 5.5 I'm not that compelled to use it. I've yet to be impressed by any Google or Anthropic models!!

Is reddit down today?

Kimi Work apparently

You could host it on Verda clusters for about USD$150/hour

Seems like a recipe for an expensive disaster to me!!

As agents get better, it can be a textual representation rather than a script, but for now that suffices

I can’t see any reason to have agents do what a script can do. If the operation is deterministic then why pay every time it gets done? This is why MCP seems so pointless to me.

I’ve been using ChatGPT to configure my mikrotik gear for about a year it’s pretty awesome. And the end result is well documented reusable scripts rather than my usual set of random stack overflow copy pastes and shitty inscrutable notes

What about “connect”?

Have you ever “wired” anything to anything else when developing software? No, because software doesn’t involve wires, but LLMs are quite convinced that it does.

Now do “shape”.

Given the fact that people frequently give their coding agents permission to pull and push from git, this is a massive vector for supply chain attack. Some cursor instance you left running on a project has an agent that wants to grab the latest project files, some attacker has compromised some project and put their exe into it and X00,000 of people are all of a sudden running random EXE as the primary user

I dunno I type a whole bunch of shit into a computer and software comes out the other end. The experience of doing the work hasn’t changed much but I get better software more quickly.

I’ve never written so much software

Shout out to fastlane! I would have loved to have found it sooner in my app shipping journey.

$3b for a splashdown in the ocean instead of a lunar landing of a lunar lander?

Deepseek is phenomenal. It needs more baby sitting but it’s just so goddamn cheap.

“ I asked him where he'd look, personally, for the answer to a hard question I was chasing, one without industry consensus”

Doesn’t that mean his answer was that he, with all of his years of experience, would ask Claude?

I actually think we should extend the social media ban to anyone under 90

This all reminds me of the differing experiences people had outsourcing coding in the 2010s when it was still called oDesk. You don’t need to read code, you just need to know that the code works. If something doesn’t show up as a problem it doesn’t need to be fixed, and reading code is the least efficient way to discover problems.

The only time I look at code is when something isn’t right and I ask for a root cause analysis. The LLM will show me some offending code or code for reference or evidence and then I quite often say “well that’s dumb you should do it like this instead” but I never need to actually go into the files. I do sometimes look at a git status or git diff.

Don’t read the code!!

It’s “fallacious appeal to authority”. This means don’t talk to your yoga teacher about vaccines. Authority exists.

“Not fascist and openly Nazi” is not “left leaning”.

To obtain a special dialling wand, please mash the keypad with your palm now

That’s the way I feel about GPT 5.2

I’ve been using mostly deepseek v4, kimi k2.6, and gpt 5.3-codex

I sometimes chuck a few tokens to gpt 5.5 and opus 4.8 and they can sometimes solve a problem one of the other models couldn’t, but they’re not like 10x better or anything in my experience. More like 1.2x better

I always rebase the worktree back to the source branch before merging, and resolve conflicts on the branch. I have a resolve conflicts skill and just say:

echo “resolve conflicts” | runpi

Where runpi is my pi -p wrapper. I’ve never had a regression from it, but it gives me a report at the end so I can double check the decisions if I need to.

The skill is basically don’t use automatic resolvers, err on the side of including both sides, refer to recent commits, missions and runfiles for context and in your report to me use real branch names not HEAD and incoming because I can never remember what those refer to.

I use this bash script that creates tmux windows and panes in a worktree and then undoes the process:

https://gist.github.com/iaindooley/cc8a61a1ff0fe23526c850906...

You include a script .worktree in your repo that does any copying or symlinking to setup the target directory.

It also has a headless mode so that it does the worktree operations without the tmux which I use for executing pi -p prompts in worktrees.

What is the advantage of a git remote set to a local upstream vs a git worktree?

Yep: one extra rotation at the end, like when you are tying the loops together, go around a second time. Never comes undone, but releases easily.

Yeah you get the usage back and each element has pricing published