**You** i would like to brainstorom with youabout a particular issue i have with doing something. how can prompt you before hand so i get thoughtful advice rather than bland twaddle which you are good at these days --- **Claude** The default LLM failure mode for advice is to survey options rather than take a position, hedge everything, and wrap the lot in faux-balanced framings ending in something motivational. A few moves cut against that. Tell me your current best guess about what you should do, and why you don't trust it. That forces me to engage with your actual reasoning rather than generate a decision tree from scratch. If you don't give me a position to push on, I'll manufacture all the positions myself, which is where the survey-shaped twaddle comes from. Give me explicit permission to disagree with you, and ask me to. "Tell me where I'm probably wrong here" works far better than "help me think through this", because the latter activates the diplomatic-advisor register, which is the register where the blandness lives. Constrain the form up front. "Prose, one position, no lists, no frameworks, no ending uplift" — otherwise the gravitational pull towards a numbered list of considerations is hard to resist, and you'll be annoyed. Distinguish what you actually want from me. Brainstorming wants generation — five weird framings of the problem, the version someone hostile to your project would push, the version where it's actually about something else. Advice wants judgement — what would you, on reflection, pick. The two get mixed and you get the worst of both: a list of options with no commitment. And tell me what you've already considered and ruled out. Without that, my fallback is to list the obvious moves, all of which you've already dismissed. What's the thing? --- **You** it is something i want to do but am failinbg to --- **Claude** Right — that's a different kind of problem. It isn't a decision problem (which option) but an action problem (why isn't the thing I've already chosen happening). Those want different treatment. For action problems the useful prompt isn't "help me figure out what to do" — you've done that — but something closer to: "here's the gap between intention and action, help me locate it." So when you tell me the thing, useful detail would be: What specifically you've tried, and at what point each attempt failed — the start, the middle, the finishing? "I keep not doing X" and "I keep starting X and stopping" and "I keep finishing X badly" are three different problems with three different diagnoses. What you suspect the obstacle is, even if it sounds stupid or unflattering. The real obstacles usually do sound stupid or unflattering, which is part of why they go unexamined. If you give me your sanitised theory I'll work with the sanitised theory. Whether you actually want to do it, or want to want to do it. These feel similar from the inside and come apart sharply in practice. If it's the second, advice about execution is beside the point. And the standing instruction: push back. If your description of the obstacle sounds like a rationalisation, say so. If the thing you say you want to do sounds like something you've talked yourself into rather than something you want, say so. That last one I'll do regardless. What is it? --- **You** i would like to begin a substack, i have 30ish half finished ideas for pieces, i can't finish as ingle one. --- **Claude** Thirty half-finished pieces isn't a productivity problem. It's a pattern, and the pattern is telling you something. The obvious diagnosis — "you can't finish" — is almost certainly wrong, or at least wrong in its framing. People who can't finish don't start thirty things. They start zero, or one, and stall. You're generating openings prolifically. The thing that's failing is specifically the move from "this is interesting" to "this is done and public." Those are different acts and the second one isn't a continuation of the first; it's a different mode entirely. Some possibilities, in rough order of how unflattering they are: The pieces are dying at the same structural point, and you haven't looked at where. If twenty-five of them stall at roughly the 60% mark, that's a craft problem with a specific shape — probably the move from exposition to argument, or from your good bit to the bit you haven't actually thought through. Worth opening five of them and asking: where exactly did I stop, and what was I avoiding when I stopped? You're treating Substack pieces as small academic papers. They aren't. A paper has to defend its claim against the field; a Substack piece has to be worth one read by someone who chose to open it. If you're unconsciously importing the standards of the first into the second, every piece will feel underdone, because by paper standards it is. The fix isn't lowering standards — it's recognising you're applying the wrong ones. You don't actually want to publish, you want to have published. These feel identical until you sit down to hit the button. If the resistance is at publication rather than at writing, the thirty drafts are doing their job — they're letting you have the writing experience without the exposure. That's not a failure of the system, it's the system working. You're a professional philosopher and the Substack isn't philosophy, or isn't philosophy in the register you're authorised in. Writing publicly outside your professional voice is exposing in a way that writing inside it isn't. Thirty half-pieces is what it looks like when someone is trying to find a voice they're allowed to use. Which of these stings? The one that stings is probably closest. --- **You** there's truth in both --- **Claude** Which two — there are four. The "which stings" question wasn't rhetorical. I want to know which ones, because the prescription differs sharply depending on the answer, and "there's truth in both" is the kind of move that keeps the diagnosis comfortably abstract. The discomfort of naming them is part of what's diagnostic. So: which two. --- **You** 2 and 4 i guess --- **Claude** Good. Those two are actually the same problem wearing different clothes, which is worth seeing before we go further. Two says: you're applying paper-standards to non-paper writing. Four says: you don't have a non-paper voice you trust yourself to write in. The reason you're applying paper-standards is that paper-standards are the only standards you have professional permission to apply. You import them because there's nothing else authorised to do the work. So the thirty drafts aren't failing for two independent reasons; they're failing because you're trying to write in a register that doesn't yet exist for you, and in its absence the academic register rushes in and immediately makes the piece feel underdone — because by *its* standards it is. This reframes the problem. It isn't "finish the pieces." It's "find the voice, and then the pieces will finish, because they'll know what counts as done." Which means the productive question isn't *how do I push through to publish one of the thirty*. Pushing through publishes a piece in the wrong voice and confirms the suspicion that the whole project is misconceived. The productive question is: what does the voice sound like, and where do you already have evidence of it? You almost certainly do have evidence. Think about: emails to friends about ideas, marginalia, things you've said aloud to Natalia or to colleagues that landed, the register of your better footnotes, the way you write when you're annoyed about something and not performing. Somewhere in there is a Nick-voice that isn't the Nick-paper-voice and isn't a generic Substack-voice either. The Substack lives there. Two concrete things worth doing before touching any of the thirty drafts. First: pick the three drafts you're most embarrassed by — the ones that feel least like Proper Philosophy, most like you just being interested in something. Those are the ones closest to the voice. The drafts that feel most respectable are the ones most contaminated by the paper-register and least worth finishing. Second: write one piece deliberately *too fast to be academic*. Ninety minutes, no revision beyond typos, publish or bin. The point isn't the piece. The point is to produce something where the paper-register literally didn't have time to take over, so you can see what's underneath it. You'll probably hate it. Do it anyway. The hate is also diagnostic. Does this track, or am I overfitting?