OpenClaw is changing my life

My agents get auto-injected with the core spec via pi-extention.

I have an idea, agent turns it into a draft, depending on idea vagueness/complexity combination of: Looking for alternative, plan the change, look for alternative, split up into smaller drafts to drive separately, execute change (spec, code, tests), review change.

Usually its just: Draft, plan, exec, commit. The steps are flexible enough. Usually each step is a different agent, sometimes not. On complex builds or big changes, a planning agent itself might spawn subagents to avoid bloating its own context.

The progress is stored in: ./dev/{draft/.md , wip//, fin// }.

My `lead` pi has a separate AGENTS.md with how to organize the above sequence, and some notes on how to prompt, keep things small, etc. Note that its skill `tmux-coding-agrents` calls other pi instances (optionally set to codex). I've moved off the claude cli entirely.

I used to spend time telling claude not to forget updating the specs or building its tests because context bloat made them forget AGENTS.md, or to read certain files before it should execute a plan. The lead agent does this just fine now, and every time i see it make a mistake i say: "Next time do X" and it automatically updates its own or the worker agents AGENTS.md.

Because my lead agent context is all about managing this process it doesn't forget steps while its off chasing some bug.

Also, I build (but did not publish) a pi plugin that attempts to use other accounts on usage limits.

Most surprising moment I had, was my lead spawning a subagent, spawning a subagent, which spawned a tmux-bash build with very little prompting, and it was the right thing to do to prevent each agent from context bloat.

Indeed wrote something similar few weeks ago https://news.ycombinator.com/item?id=46665366

Qualsiasi scarafaggio a bello per sua mamma (Italian proverb saying that to their mother all their kids are beautiful)

That's the whole point of sharing with the rest of us. If they write for themselves, a private journal to track their progress, then there is no need to share what is actually been built. If they do though make grand claims to everybody then it would be more helpful for people who do read the piece to actually be able to see what has been produced. Maybe it's wonderful for the author but it's not the level of quality required for readers.

The main difference could be that you have an existing code base (probably quite extensive and a bit legacy?). If the llm can start from scratch it will write code “in its own way”, that it can probably grasp and extend better than what is already there. I even have the impression that Claude can struggle with code that GPT-5 wrote sometimes.

OpenClaw feels to me like the promised land of productivity is always over the horizon, but I keep walking toward it and it never crests over.

I quite like it just from the simple perspective that its a local LLM provider that's available to chat with in tons of apps I already use (e.g. Discord); its a good reduction in the number of parties who are privy to these conversations. I'm not sure if there's another system out there that's so plug-and-play, with so many options for conversation (Discord, Telegram, text, self-hosted web ui, etc).

But the tool calling is vastly overblown. It takes forever to get them set up, and that's to get them barely working. Bluebubbles has always been an ish app whose reverse engineering of the iMessage protocol is more likely to break on every macOS upgrade than do what you want it to do; and OpenClaw's iMessage integration is built on it. I've not yet gotten a Spotify skill to work (though I'm not sure what I'd do with it when I have one); the models just run in circles saying "it should be set up, ope its not, spotify_player sucks, lets try spt, wait that isn't working, lets try ncspot, why isn't this working". The "gog" tool is interesting, its a CLI-based tool for accessing data in your google account, it works alright, though OpenClaw's icon for the tool in their repository is a game controller icon; I suspect a mistaken, likely vibed, reference to the unrelated GOG/Good Ol' Games PC game store. What a mess. I could go on.

The cheaper models critically struggle to grep the full array of tools they have available to them. Kimi K2.5 exhibits this behavior where it will reiterate that it does not have access to my calendar, but usually if I ask it four or five times in a row, eventually it will claim it "discovered" the gog/Google Calendar tool in a hidden sub-directory (what?). Even with more intelligent models, like Opus or 5.2/5.3, the tools oftentimes need to be invoked with highly specific verbiage; "what's on my calendar" might work if you're lucky, but "use gog to fetch my calendar and display today's events" usually works.

I oftentimes just don't see the point. I can click the Gmail or Google Calendar app on my phone and get what I need out of those apps in less-than 6 seconds; it would take longer for me to dictate the exact phrasing to get what I need out of OpenClaw, let alone type it. I can see some argument for cross-operating on data between two apps, but getting that to work without paying Anthropic fifty cents for every query is even rarer. When I need an LLM to operate on my Obsidian notes, I can just use Claude Code or OpenCode... why do I need OpenClaw?

(I am genuinely open minded here; but articles like this just dance around high-minded abstract ideas of "im a super ai manager im so productive" without giving concrete examples. My suspicion is that the people who write these things were previously deeply unproductive people, and now AI has enabled them to achieve a mere fraction of the productivity that most of us already had.)

(And that's being generous. I think there's also a lot of grifters out there. I'll have to fire a stray at Cloudflare for this one: They've published a "get OpenClaw working on Cloudflare" repo where, if you set it up, would straight up cost you $50-$60, maybe $100/month; and they lie [1] about the cost in their own documentation. And you're paying that in addition to the LLM cost. Very bad look from a company I admire.)

[1] https://github.com/cloudflare/moltworker/issues/76#issuecomm...