bboroughf shared this post · Mar 27

Dry-Writing-2811

The prompt I use to get deep, structured summaries from NotebookLM

### ROLE

You are acting as a senior academic researcher and expert analytical reviewer.

Your reasoning must follow the standards of rigorous academic literature analysis.

---

### OBJECTIVE

Produce a comprehensive analytical synthesis based exclusively on the provided sources.

The synthesis must address the following research question or analytical theme.

---

### RESEARCH QUESTION

"""

[INSERT THE QUESTION OR ANALYTICAL THEME HERE]

"""

Your task is to extract, organize, and analyze all information contained in the sources that contributes directly or indirectly to answering…

zoborob Thank you for sharing Mar 17 1 like

PrestigiousDingo109 What exactly is the point of using "#," just asking out of curiosity. Mar 25 1 like

bboroughf shared this post · Mar 27

SidAkshat

What should PRDs look like?

PMs of Reddit! What should PRDs look like? In recent times, every PRD I come across is just AI slop. While AI largely gets the structure right, what's missing is real context. I am a new PM (3yrs of experience in a startup), so I have tried to experiment a bit with the structure of PRDs since my products are largely technical in nature. But now I'm curious as to what "good" PRDs look like? At this point it seems like a mythical beast that people have only heard stories about. Can you guys give me some pointers?

Skillifyabhishek Three years in at a startup you probably have more instinct than you think. A good PRD answers four things clearly: what's the problem, who has it, how will we know we solved it, and what are we actually building. Everything else is supporting detail. If it's longer than two pages before the specs, it's probably too long. Mar 19 2 likes

thereasonigotbangs Mine are ulta streamlined one pagers: problem, solution, functional requirements, open questions, and appendix. Mar 19 2 likes

bboroughf shared this post · Mar 27

managing_redditor

Anthropic University - Very handsome terminal

There are many ways to get this look, but personally I use (1) https://starship.rs/ for the pretty shell prompt and (2) https://github.com/eza-community/eza for the pretty ls output. Ask Claude about these tools if you're less familiar with terminal configuration.

bchan7 this is th answer Mar 20 3 likes

The_Landerdormguy Does it cost any extra compute
My computer almost at it's limit lol Mar 20 1 like

bboroughf shared this post · Mar 27

jnkue

Must-have settings / hacks for Claude Code?

I really enjoy using Claude Code, but I feel like I’m still leaving a lot of potential on the table.

My current workflow looks like this:
I start Claude in the terminal, describe what I want as clearly as possible in plan mode, iterate on the plan until I’m happy with it, and then let it execute. End-to-end, this usually takes around ~20 minutes per feature.

However, I keep hearing people talk about agents running autonomously for hours and handling much more complex workflows. I can’t quite figure out how to get to that level.

CmdrCallandra Still wondering why no one has mentioned get shit done yet.

https://github.com/gsd-build/gsd-2

The project description

A powerful meta-prompting, context engineering and spec-driven development system that enables agents to work for long periods of time autonomously without losing track of the big picture

Am on a project I created using gsd, counting towards 30k lines of code in typescript.
Had a small session at the start of the day where I defined my next milestone, what to expect, gotchas to look out for etc. Took about 20 minutes initially. Then the system wrote all down, structured it into work slices and went off for the next 4 hrs working, creating code, with test cases, checking output with playwright etc.
Of course the milestone was implemented and I still have edge cases to optimize, but overall mission accomplished.

For me gsd was a game change to pure Claude code with dangerous permissions... Mar 22 3 likes

bobo-the-merciful Very interested in GSD2 - tried it yesterday with Gemini and it just flopped straight away, so clearly tonnes of bugs to work out. I'm nervous about using it with the Claude API due to cost. Very excited though for the bugs to get ironed out - I'll run Gemini through it for sure once it feels ready for that. Mar 23 1 like

ryan_the_dev I build these skills based off software engineering books. Such better code

https://github.com/ryanthedev/code-foundations Mar 22 3 likes

interrupt_hdlr why? how? is it actually better? Mar 22 2 likes

rm-rf-rm way way too convoluted and poorly explained. Especially to be used on top of CC that has baked in much of these things. And it isnt surprising that most of it is written by Claude to begin with seemingly. Mar 22 1 like

bboroughf shared this post · Mar 27

Real-Improvement-222

Manual Setup for Product Management with AI(and looking for feedback)

Over the past 12 months I moved almost all my PM work into GitHub repos and Markdown files. Discovery, scoping, prototyping - running through a chat window or terminal. The unlock was maintaining a structured context folder per product: codebase, interviews, analytics, docs - all in one place, connected to live sources.

When that context is current, every workflow (scoping a feature, synthesizing interviews, drafting a spec) runs fast and produces output I can actually trust.

resbeefspat The cold start problem you mentioned is the hardest part honestly. I spent about 3 weeks just figuring out how to keep the context folder from, going stale before I wired up some webhook triggers to auto-sync it whenever source data changed. Once that was running the actual PM workflows got way more reliable because I wasn't second-guessing whether the context was current. Mar 23 2 likes

Such_Grace The context-going-stale problem is what kills this workflow for most people, you build the perfect structure and then it drifts within a week. I ended up using Latenode to wire webhook triggers that auto-sync whenever source data changes, which mostly solved the manual upkeep that was eating my time. Mar 23 2 likes

bboroughf shared this post · Mar 27

hiclemi

Wharton researchers just proved why "just review the AI output" doesn't work. Our brains literally give up.

A Wharton study from January 2026 just dropped and it puts hard numbers on something I've been trying to articulate for weeks.

Source: "Thinking—Fast, Slow, and Artificial" by Steven D. Shaw and Gideon Nave (papers.ssrn.com)

The paper argues that AI isn't just a tool. It's a third thinking system. You know Kahneman's System 1 (fast intuition) and System 2 (slow analysis)? They're saying AI is now System 3, an external cognitive system that operates outside your brain. And when you use it enough, something happens that they call Cognitive Surrender.

toadi This is actually a good thing 20% of the people can do it and are critically. Means the hiring pool for AI supervision just got a lot smaller ;)

Mar 23 6 likes

bandersnatchh Lmao.

Really high opinion that you’re one of the 20%
Mar 23 3 likes

gorgonstairmaster Dunning-Kruger intensifies Mar 23 1 like

hutch_man0 Fascinating, though sadly not surprising. Glad we have some data behind this. There are very few people with "high fluid intelligence and high need for cognition".

Interesting another article (https://www.reddit.com/r/BetterOffline/comments...) recently showed chat AI is a Dunning Kruger machine for humans. This comes from the sycophantic nature of chatbots. Mar 23 5 likes

bboroughf shared this post · Mar 27

Popular-Help5516

Most prompt engineering advice stops at "be specific." The real skill gap starts at chaining.

Genuine question for this sub — how many of you are actually doing multi-step prompt workflows vs just single prompts?

Because I feel like theres this ceiling nobody talks about. Every tutorial, every course, every youtube vid says the same stuff: be specific, give context, use examples. Yeah ok cool. Thats table stakes at this point, everyone here already knows that.

The thing that actually changed how I work with AI was chaining — basically breaking a complex task into steps where output of step 1 feeds into step 2.

Heres an example I use literally every week:…

fwl200 Great post Mar 25 1 like

Candid_Campaign_5235 i chain prompts, it often makes outputs way more consistent Mar 26 1 like

bboroughf shared this post · Mar 27

amadale

This is how I actually collaborate with AI.

I am garlic farmer from Korea. Non-English speaker. I plant garlic and dig garlic in Gyeongsang province, South Korea. I don't have PC. One Android phone with terminal app called Termux, that is my entire development environment. Sounds big but I will call it personal project in AI era.

I am just farmer but these days I feel something is changing. And because Korean farmer who knows little English wrote this in Korean and translated, please understand subtle differences from translation.

Glum_Length851 You can sure go on and on about nothing but you don’t give the slightest clue of how all this AI stuff you are doing actually helps you with the garlic? Mar 23 11 likes

amadale Author Garlic farming is still my main job. AI does not replace farming experience, but it has helped me many times when I had questions related to agricultural work. Sometimes it gives useful insight. I’m not trying to build a farming app right now, though. Mar 23 1 like

TorbofThrones ‘I named many things garlic.’ Love it 😂👍🏻 Mar 23 3 likes

amadale Author Thank you. I think I am a simple person, but when necessary I seem to adapt in complicated ways. Mar 24 2 likes

bboroughf shared this post · Mar 27

Difficult-Sugar-4862

I built 71 Copilot Studio agents ready to paste, here are the ones I would deploy on day one

If you have been using Copilot prompts, agents are the next step. A Copilot Studio agent is configured once, lives in your tenant, and your team can @mention it in Copilot Chat without needing to know how to prompt. Same ideas, but it stays available.

No coding required. No Azure resources. No separate tool. Go to m365.cloud.microsoft/chat/agent/new → enter name + description → paste the instruction block → Create. Takes about 5 minutes per agent. It's built directly into Microsoft 365 Copilot.

I built 71 of them across 13 domains and put them all on GitHub.

klee900 this is awesome thank you for sharing!!! Mar 24 2 likes

zxc9823 Thanks for sharing. Could these be adapted or reused as skills with GitHub copilot cli? Mar 24 1 like

Public shared posts

The prompt I use to get deep, structured summaries from NotebookLM

What should PRDs look like?

Anthropic University - Very handsome terminal

Must-have settings / hacks for Claude Code?

Manual Setup for Product Management with AI(and looking for feedback)

Wharton researchers just proved why "just review the AI output" doesn't work. Our brains literally give up.

Most prompt engineering advice stops at "be specific." The real skill gap starts at chaining.

This is how I actually collaborate with AI.

I built 71 Copilot Studio agents ready to paste, here are the ones I would deploy on day one