Poorvi Shrivastav 👊🏽

Honored and excited to be in good company. Join me and other product leaders at clubhouse on April 6th.

Request to @meetgranola make it easier to subscribe events from multiple calendars

Falling in love with @lovable it's 11:30 PM and I am redesigning the learning app for my 6 years old. What have we unlocked in terms of true personalization.

@nikitabier This is a great strategy and a differentiator. I am seeing an increase of folks from builder background, fitness beginners and parents who are 1st gen immigrants in my feed. It seems to be working.

The goal for your X timeline is to get out of the mainstream algo and the political crusades and find your niche. You should be able to post about your interests and have friendly, relevant people chime in. If you’re seeing gas station fight videos, your account is not ramped up yet. We are working everyday to fix this.

@bhalligan a 100%. Saw 2 pitches last week - same issue - both co-founders playing agent-coach during the pitch.

Dear Founders, The CEO should captain/lead the vc pitch, regardless of how much experience the co-founder has. Love, Brian.

@levie @levie agree on the list and adding that with an agents PM also needs to think about 'operational visibility' in a different way. Move beyond real-time proactive insights dashboards. These agents can be managed by an agent manager or a human manager.

When doing product management for AI agents, the biggest shift is that the user you think about making successful is the agent. They are effectively your new “customer”. And the whole game is to figure out how to get the agent the most relevant context possible to execute their task, which is why context engineering is such a big deal. By default, AI agents know absolutely nothing about the work that they're supposed to do. You have to take this raw form of intelligence and mold it directly to the workflow that you're trying to automate. Imagine telling a super smart person (with no domain expertise by default) that just joined your company that you want them to review a contract in line with all your other contracts, or write code for a new product, or come up with a medical diagnosis about a patient, or estimate risk about a particular initiative, etc.  That new person has near infinite permutations of things they *could* go off and do to solve that task. That person would likely need instructions about how to perform the task, they would need a clear set of goals and “rules” that you expect the output for, and then they would need to talk to colleagues, review internal documentation, review past work in a similar area, and so on. Well, the issue is the same for the AI agent. Except they're at a huge disadvantage: they can't "easily" talk to colleagues, they often don't have access to the same resources a person does by default, and they by default don't know which part of their knowledge to tap into for a task.  But most importantly, they have an inherent limit of the information they can consume in each step of the process (size of the context window). And when they get too much data they perform worse at the task (context rot).  Entire companies will win or lose based on their ability to get the *right* context to the agent. Thus, AI agent PMs will require a very different type of expertise: * Deeply understand the domain that you're building agents for. In an ideal world you actually have worked in that space, but if you haven't then a good chunk of your time should be studying the actual people that work in that space. Understanding every single step of their work is going to be critical. * You have to think through essentially what would a human need to know to do the task across instructions, rules, existing data sources, best practices, etc. And then figure out how you actually get this data to the model (context engineering). * Figure out what end-user UX and features are needed to supply the agent with the right context to perform the task. Things like task queues, connecting to data sources, how to re-prompt the user during a long running task, how the user can review the output work and make modifications, and so on. * Doing evals on the AI agents with every tweak of the agent's instructions, model improvements, and any other variable that changes in the process. Especially figuring out how these changes affect the real-world customer environments is critical. Lots of work changes for engineers and PMs in a world of building AI agents and we’re just at the start of thinking through what this looks like.

@akothari @Tesla We are interested. We have a 7 years old but she has shown interest in understanding both operations and technology.

I took my daughter to the @Tesla factory in Fremont yesterday. The 3 hours we spent there were both awe-inspiring and mind-bending. It made me realize how much I want to take her to more factories, offices, farms, and beyond. I’m thinking of starting a WhatsApp group for SF parents (ideally with kids around 8–12 years old) to coordinate these trips. If you’re interested, reach out via DM!

So excited about @Spotify launching comments and responses especially on podcasts. Should have happened earlier. They have converted one dimensional signals to multi-dimensional engagement data that will unlock value for both customers and company.

@anjuls Haha.. that happened to us on last Durga Puja!

Went for Ganesh pandal tour but ended up with a food tour. #indore

One feature request - once I analyze call summaries and want to better understand a specific point, I need an 'explain' feature in-line in 'Ask Granola' chat.

@meetgranola is a truly user's user tool. Proves you don't have to have complicated workflows to tackle operational complexity. Record, analyze and produce insights in a way that makes your users life easier.

@lennysan Lovable and/ or Claude code - even used it for some validation test on my Stanford CS exam.

Do you use Claude Code (or some other CLI) for a non-coding use case? I'd love to know how. Please share in the comments. If it's awesome, I'll feature you in the newsletter 👀

@OpenAI Agent mode is been harder to work with. Was hoping to create google calendar events but gave up after 3 levels of google trust checks only ending up with request to change passwords.. agent mode + connectors should balance trust and efficiency.

@OpenAI ChatGPT bringing back GPT 4o based on user feedback is both a lesson in packaging nuances and customer centricity.

@OpenAI study mode is acting weird, giving me answers but not leading to answers. This time I am researching complex topics like 'DPO' and 'PPO' instead of K12 curriculum. Surprisingly, Gemini Learn LM model is much better with socratic answering.. hmm.

@adamnash @OpenAI @kevinweil Increase voice limits too in this plan

Fairly sure @OpenAI would make more revenue if they released a family plan for up to 6 members for $50/mo. Add in some vertical tutoring features or parental supervision and charge even more. cc: @kevinweil

Does @OpenAI have a feature request log? Analyzing a long document or response (e.g. strategic insights) and storing comments/ questions as I read through the document. If I use 'Ask GPT', it starts a new thread making me lose context of earlier document/ response.

Hail the product that's easy to buy, easy to sell and easy to use. Simplistic packaging for the win - says my dad.

If there is one course you should consider taking this year, it's this one @AnthropicAI

Introducing AI agents alongside humans on the Internet is like introducing cars alongside humans on physical streets. (1) Some spaces will be AI-only, like bots trading against each other, similar to dedicated highways for cars. (2) Some will be human-only, like proof-of-human social networks, similar to walkable communities offline. (3) And some will be a mix, like Claude Code, similar to city streets with human and machine mixed in close proximity. But the whole digital world will need to be rearchitected.

@thesamparr 'Leading from the middle' book

Someone in my company just asked a good question: Does anyone have good recommendations for how to get better at managing up? Could be a book, a course, visual, article, anything. I'm all ears! Anything good?