Harris

Most PMs don't have a discovery problem. They have a time problem. Recruiting, interviewing, and synthesizing user feedback takes forever. I've recently been using Trooly.ai to shortcut that. Here's how a FIFA World Cup 2026 study played out. 🧵

If research keeps getting pushed down the sprint because there's no time. This is worth trying 👇 app.trooly.ai (sponsored, but this is genuinely the workflow I used.)

Trooly organized everything by recurring pattern. Navigation issues in one section, auth friction in another, schedule problems in a third. What normally takes days of recruiting participants, conducting interviews, and organizing notes became one structured workflow.

Most PMs don't have a discovery problem. They have a time problem. Recruiting, interviewing, and synthesizing user feedback takes forever. I've recently been using Trooly.ai to shortcut that. Here's how a FIFA World Cup 2026 study played out. 🧵

I have recently stopped asking "Which AI is best?" Instead, I give every AI one job. That one change eliminated hours of unnecessary work every week. Here's the workflow I use almost every day and so shall you: ↓

@Ninaa967 Good to know that you like the stuff

@HarrisDecodes Thanks for the info

Solid take.NotebookLM’s real superpower isn’t the viral podcast, it’s the hard grounding. By forcing every answer to come only from the documents you upload (with clickable citations), it removes the biggest risk in research AI: hallucinated facts or blended in training data.For anything where accuracy and verifiability matter, that constraint is huge. The podcast is just the fun wrapper around it.

Upload your documents. Get back a podcast of two AI hosts discussing them. That is the feature that made NotebookLM go viral. It is not even the most useful thing it does. NotebookLM is Google's free research tool. You give it sources PDFs, Google Docs, web pages, YouTube videos, pasted text and it becomes an expert on those specific documents. Only those. Nothing else. That last part is the whole point. And it is what makes NotebookLM different from every other AI tool you use. Here is the problem with ChatGPT, Claude, and Gemini for research. They know everything. Which means they will confidently answer from their training data, blend in facts you did not give them, and occasionally invent a citation that sounds real and does not exist. When you are trying to understand a specific set of documents, that is a liability. You can never be fully sure whether an answer came from your source or from the model's memory. NotebookLM closes that door. It only answers from the documents you upload. Every answer comes with inline citations pointing to the exact passage in your exact source. Click the citation, jump to the sentence. No hallucinated facts. No outside noise. No made-up references. For research, that single constraint changes everything. Now here is what people actually do with it, beyond the famous podcast feature. The Audio Overview. This is the viral one. Upload your sources and NotebookLM generates a podcast, two AI hosts discussing your material in a natural, conversational back-and-forth. Students turn lecture notes into something they can listen to on a commute. Researchers turn a stack of papers into a discussion they can absorb while walking. It is genuinely good. And it is free. The interrogation. Upload 40 research papers on a topic. Then ask: "What do these papers agree on, and where do they directly contradict each other?" NotebookLM reads across all of them and maps the consensus and the conflicts with citations to which paper said what. The study guide. Upload a textbook chapter and ask it to generate a study guide, a timeline, a glossary, or a set of practice questions. It builds them from your material, not from a generic template. The synthesis. Upload your own messy notes alongside the source material and ask: "Where am I misunderstanding this compared to what the sources actually say?" It finds the gaps between what you think you know and what the documents actually state. Here is why the document-grounding matters more than the features. The best way to understand a complex subject is not to ask an AI what it knows. It is to give the AI exactly the material you need to understand and then question it relentlessly, knowing every answer is anchored to a source you can verify. NotebookLM is the only major AI tool built entirely around that principle. It is not trying to know everything. It is trying to understand your documents and help you do the same. The podcast feature got it the attention. The grounding is what makes it the most trustworthy AI tool you can put your research into. And all of it is free.

@nrqa__ From weeks of grinding to one Slack message. This is the real power of agentic AI. Full pipeline in one chat. Wild.

POV: making a video ad 2024: research the product, study competitors, script it, film it, edit it 2026: type it in Slack, tag Claude, and Arcads does all five inside the chat😳

@rubenhassid Respect for the honest update. Files & folders were holding us back. Skills & Projects is clearly the better system, especially for teams.

@nikola_mr64990 Excellent thread! These prompts are a smart way to turn Perplexity into a proper research assistant for dividend stocks.

BREAKING: AI can now create dividend portfolios that can generate $100,000 in passive income a year — for free. Here are 12 powerful Perplexity prompts With which you will find safe + growing dividend stocks. Save this thread. 🧵

@cyrilXBT In Benchmarks test solo stats the real work is team ball. Opus plans & directs, Sonnet 5 executes fast & cheap. That's how you actually ship without burning cash on every token.

EVERYONE IS DUNKING ON SONNET 5 FOR LOSING TO OPUS 4.8 ON BENCHMARKS. They are all making the same mistake: judging a teammate by a solo test. Sonnet 5 was never built to win alone. It was built to execute fast and cheap while a smarter model directs it. Set Opus 4.8 to ultracode: - it plans the workflow - spawns the agents - executes - verifies its own work automatically Then hand the implementation work to Sonnet 5. When Fable 5 comes back, the setup gets genuinely good: - Fable 5 decides what to build - Sonnet 5 actually builds it You stop paying flagship prices to do grunt work. Bookmark this. Follow @cyrilXBT

@Shipon_40AI You got the point

@HarrisDecodes Sharing the workflow could save everyone hours.

@alakara06 @shedntcare_ Exactly! Give it a strong reasoning brain (like Opus) plus these hands and imagination becomes the only limit. Great point.

@HarrisDecodes @shedntcare_ aslında bir katman daha eklenmeli beyin lazım opus gpt gibi bu eller olmalı nereye nasıl tıklayacağını çok iyi biliyor ama karar merkezi şart o zaman sınır hayal gücü olur

China just released a desktop automation agent that runs 100% locally. It can run any desktop app, open files, browse websites, and automate tasks without needing an internet connection. 100% Open-Source.

@aaliya_va Very true. Stream lining workflow is the door to productivity

@HarrisDecodes Simple workflows usually save the most time

@Code_SantrexAI Exactly 💯. Train each tool for specialized task

@HarrisDecodes Nice move. Focusing on one job per AI to cut clutter is a game changer. Sharing your workflow could save a lot of us time too.

10/ Takeaway The future doesn't belong to people using the best AI. It belongs to people building the best AI systems. That's the mindset shift that changed everything for me. If this helped, Follow @HarrisDecodes Like the thread, RT so that your audience also read it and comment your thoughts Every week I share practical AI systems that save hours—not just prompts.

9/ The Real Secret The best AI workflows aren't built around tools. They're built around roles. Once every model has a clear responsibility... Everything becomes faster, better, and far less frustrating.