164 episodes
- This is a free preview of a paid episode. To hear more, visit www.news.aakashg.com
Brought to you by
Amplitude - The market-leader in product analytics
Product Faculty - Get $150 off their #1 AI PM Certification via code AAKASH150
Land PM Job - My next cohort starts Monday, November 23! Join to learn AI job search from me
Customer.io - Send smarter messages using your product data
Ariso - Ship AI agents and features faster, with fewer regressions
Get a full year of my favorite AI tools, plus my paid newsletter, with the bundle: bundle.aakashg.com
Today's Episode
I'll admit it. I block out two hours to do one thing, and then I realize I've spent the whole two hours looping through my 12 different agents, giving them prompts.
Shuhao Zhang has a fix for that. He's the co-founder and CPO of TinyFish, which raised $47M from ICONIQ and went from 100,000 to 200,000 users in 18 days, then to 300,000 in just 10. At TinyFish, anyone can ship to production. He pulled up his real GitHub and walked me through the whole system, from the docs he writes before an agent touches the code to the ops agent that fixes production while his engineers sleep.
Not many people are willing to open up their actual system like that. If your engineers are moving fast with AI but your product org isn't feeling it yet, this one's for you.
10 Key takeaways
1. Front-load the design so agents can run for hours - Shuhao spends his own time up front, on a stack of docs from the PRD down to the implementation plan. He asks the agent to "list the 10, 15 decisions that need my attention" and to make the low-risk calls itself. That buys "20-plus hours or multi-day" of building without a single interruption.
2. Start every project from a template - The deployment part is "usually where the agent can be very opinionated." So every TinyFish project starts from one repo, with Terraform for sandbox and production plus a .claude folder of org-wide best practices and the rules every agent has to follow.
3. Let an AI reviewer approve the low-risk PRs - When agents write tens of thousands of lines, "the human reviewer becomes the bottleneck." CodeRabbit reviews every PR, and the coding agent fixes what it flags until "both are happy with the results." Low risk merges itself. High risk tags an engineer.
4. Write the reviewer a constitution - It holds your conventions and, just as important, how strict to be. Skip the strictness and two agents will nitpick forever. As I put it on the show, you don't want to "burn a bunch of tokens on this kind of back and forth."
5. Give your agents eyes in production - Humans used to read the logs and traces. Shuhao's view: "AI agents are better ones to consume this huge amount of data in production continuously." His ops agent gets VPN access and a skill that explains where everything lives.
6. Put an ops agent on call - The day we recorded, it caught LLM calls that hit max tokens and came back as a server error when it should have been a client error. It opened a Linear ticket, wrote the fix, got it approved and shipped it to staging. All of that can happen "when the engineers are sleeping."
7. Roll out every change to 1% first - "Even for a critical fix." That used to be too expensive, because a human had to watch every partial rollout. Now an agent watches. Shuhao calls it "safer and more efficient than human operation."
8. Building fast isn't shipping slop - Locking a PRD used to take six months of director sign-offs, then three to six more for the engineering design. Today the best feedback comes from users, so the real cost is not building at all. You're "missing the opportunity to build this muscle."
9. Treat agents as your next users - Shuhao thinks "there will be trillions of them," and they're picky. They run their own benchmarks and check the price before they choose. TinyFish kept "hill climbing in production" on their edge cases until agents saw it as best in class, and that's where the traffic comes from.
10. You don't have to be an engineer, but bring one - "Don't be intimidated by the amount of technologies and engineering requirements." Know the high-level risks and what good looks like. For the rest, "if you don't know how to drive, you got to have a friend who can drive."
Go Deeper
How to Ship Your First Pull Request as a PM
How PMs Ship 100K Lines of Code at OpenAI, with Ryan Lopopolo
How to Build a Full AI Dev Team in Claude Code
AI PM's Ultimate Guide: Observability
The PM's Guide to Agent Distribution: MCP Servers, CLIs, and AGENTS.md
How to Become a "Builder PM"
Where to find Shuhao Zhang
LinkedIn
X
TinyFish
Where to find Aakash
X
LinkedIn
Newsletter
Listen to The Growth Podcast
Apple Podcasts
Spotify
YouTube
Interested in sponsoring the podcast? Email productgrowthppp[at]gmail[dot]com. - This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit www.news.aakashg.com/subscribe
Brought to you by
Jira Product Discovery - Plan with purpose and ship with confidence.
Today’s Episode
Engineering is getting faster with AI. Is product management?
I’d argue one way it can is by helping you talk to far more of your customers. You’ve probably used Claude to synthesize interview transcripts. But have you used it to run research with every customer who drops out of your funnel?
Greg Greiner has. He was chief product officer at Bolt and head of product at Uber for Business. Back then, 7 to 8 customer conversations made a good month. Now he’s CEO of Woven, where AI follows up with customers over email.
Some Woven customers run tens of thousands of those conversations a month, and some run hundreds of thousands.
You should be using AI to create a scaled representation of yourself to talk to your customers for you.
Then he showed you how to build your own version without buying his product, using a Claude Code project he’d put together in about 30 minutes!
10 Key takeaways
1. Go from seven customer conversations a month to thousands - At Bolt, Greg "would have been happy to talk to seven or eight customers a month." AI now holds those conversations with thousands.
2. Split every concern into a messaging fix or a product fix - Shipping questions convert once the AI answers them. Fix those with messaging. Subscription confusion doesn't convert even after the AI responds, so that one is a product problem.
3. Email the customers who already left - Someone who abandoned isn't in your app anymore, and an in-app research chat can hurt conversion for people who haven't decided yet. Email is "a channel that all of your customers already use daily," which is why Greg's research campaigns get 20% to 40% of people to reply, and a fitness company saw about 35% on trial abandonment.
4. Ask one simple question per email - The common mistake is trying to "recreate your survey maybe in one email." Ask one short, direct question and let the customer lead.
5. Put your emails where people drop off - Start at the widest part of the funnel, people who entered an email and never finished signing up. Few of them reply. It's still your biggest group. When a trial ends without converting, offer something like 20% off just for feedback. Greg says that "massively increases reply rate," and the people who reply are the ones already on the fence.
6. Build the agent in Claude Code in under an hour - Greg's demo runs on a CSV of customers and a few markdown files about the company, and it sends through the Gmail MCP. The first pass broke reply threading, but it worked within 30 to 40 minutes. Keep replies fast. When Woven tested adding delays, reply rates dropped "precipitously."
7. Add an eval loop with one prompt - Greg asked Claude Code to check auto-sent replies for "groundedness in the documentation" and to save a fresh draft when one fails. Feed the failure into a full rerun, or you'll overfit.
8. Test customer-facing AI in a sandbox first - An internal tool can live with a lower bar, because a person reviews it first. Even a model upgrade on customer-facing AI means running trusted sample data outside production and checking guardrails still pass at the same rate. LLM judges want to please too, so make yours tie every action to the instruction that allowed it.
9. One PM can now run a team that ships like 25 engineers - Greg's most productive stretch at Uber was one PM to about 25 engineers. Now three to five engineers ship that much, and the PM keeps it all "greater than the sum of its parts."
10. Spend your time deciding what to build - At Woven, a team of four, building fast with good quality is the easy part. "What's hard is still building the right thing." So the team debates the problem at a whiteboard while recording, then feeds the transcript and the diagram to an AI skill that writes the PRD and the Linear tasks. It only works if everyone at the table has a baseline of product and business sense.
Go Deeper
In-app surveys: the playbook from 4M PostHog responses
How to do AI-powered discovery, with Caitlin Sullivan
Teresa Torres' step-by-step guide to AI product discovery
AI customer intelligence: the complete guide for product teams
Claude Code for non-technical PMs, with Andre Albuquerque
Why 90% of LLM judges fail, and how PMs can fix them
The AI product operating model
Where to find Greg Greiner
LinkedIn
Where to find Aakash
X
LinkedIn
Newsletter
Listen to The Growth Podcast
Apple Podcasts
Spotify
YouTube
Interested in sponsoring the podcast? Email productgrowthppp[at]gmail[dot]com - This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit www.news.aakashg.com/subscribe
Brought to you by
AutonomyAI - Go from a product manager to a product builder and ship changes to the codebase as PRs
Get a full year of my favorite AI tools, plus my paid newsletter, with the bundle: bundle.aakashg.com
Today’s Episode
Companies like Webflow were among the first frontier labs targeted.
ChatGPT has a site builder, and OpenAI is claiming 5 million sites in 3 months! People are building their websites in Claude too. If you run a website builder, that’s about as direct as competition gets.
Bending Spoons just bought Airtable for $1.28 billion, a company once valued at over $11 billion. Then it bought Miro for $1.36 billion, down from $17.5 billion.
Webflow raised $4 billion in that same era.
Webflow’s CEO, Linda Tong, told me about CEOs who vibe-coded their way off Webflow, only to call them back 90 days later.
Watch the full masterclass as she walks through how she’d judge any SaaS company.
10 Key takeaways
1. Most of the work happens after you hit publish - AI can write a site's first version, but most of Webflow's business sits in what comes after, from analytics and experiments to a CMS a whole team can edit. Linda even tells people to start in ChatGPT's site builder and bring the site over.
2. Pick a box the labs aren't building for - Webflow's users had all kinds of titles, but they were using it for marketing. So Webflow went deep on marketing teams with dozens to hundreds of people, a group that probably isn't a frontier lab's priority. Linda called the decision terrifying.
3. Hold context and governance people won't move - Products that store deep domain context become a center of gravity, and roles and permissions are something few people want to vibe code. A product that's only a UI or API layer over other systems is the one customers can leave fast.
4. Run the total cost before you vibe code it - You can build a tool in an afternoon, but software deteriorates and someone has to keep maintaining it. Webflow calls the customers who learned that the hard way boomerangs, and Linda has seen CMOs come back within about 90 days.
5. Low multiples are part math and part fear - A company growing 50% can earn a 10x multiple, while one growing 10% to 20% gets 2x to 3x, so some of the drop is fair. Linda calls the rest "a hangover of the fear" and points to PE buyers, who only buy when they expect a return.
6. Software has to go from MapQuest to Google Maps - Most SaaS still shows everyone the same static screen on top of decades of features. Linda wants products that adapt to what each person is trying to do, with a context engine that learns so a new agent isn't still an intern 60 days in.
7. Valuations from 2021 and 2022 aren't believable - Webflow raised at $4 billion in 2022 and hasn't raised since. Linda says it has plenty of cash and has run close to break even for a while, so she'd only raise for a specific reason, like a big acquisition.
8. PMs start with a PR, on half-pizza teams - Webflow's PMs code the prototype, and designers and engineers swarm on it while it's rough. A funded product area is now two or three engineers with a designer and a PM, which Linda jokes is half a pizza.
9. Track the behavior your AI feature should change - Webflow tracks how many customers use its MCP server, and it also tracks publish rate, how often someone comes in and publishes a change. People publish more once they use MCP, and new customers reach their first publish faster.
10. Webflow hires for how you work with a team - It skipped the AI PM title because every PM has to use AI now. After one-on-ones that score things like AI fluency, a final working session with a group decides it, and the number rejected there has been high.
Go Deeper
n8n's CEO on why n8n isn't dead
How to think about product strategy in the age of AI
How product leaders should use AI, with Rachel Wolan
What an AI-native product team looks like at Salesforce
The PM's guide to agent distribution and MCP servers
Zapier's AI fluency rubric for PMs
Where to find Linda Tong
LinkedIn
X
Where to find Aakash
X
LinkedIn
Newsletter
Listen to The Growth Podcast
Apple Podcasts
Spotify
YouTube
Interested in sponsoring the podcast? Email productgrowthppp[at]gmail[dot]com. - This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit www.news.aakashg.com/subscribe
Brought to you by
Fingerprint: Identify good and bad visitors with industry-leading accuracy
Ariso: Ship AI agents and features faster, with fewer regressions
Customer.io: Send smarter messages using your product data
Land your next PM role: Coached live by PM leaders who've hired at Meta, Uber, Microsoft, etc
Product Faculty: Get $150 off their #1 AI PM Certification via code AAKASH150
Today's Episode
Salesforce and Anthropic announced Claudeforce in August. Now a sales team can ask Claude about its pipeline without ever opening Salesforce.
That worries me. LLMs are probabilistic, and you can't have two pipeline numbers.
So I sat down with Jayesh Govindarajan, who runs AI engineering at Salesforce. His teams build Agentforce, the foundation customers use to build their own agents. He walks me through how Salesforce gets one right answer out of an agent every time.
Then he lays out the ADLC, the life cycle that replaces the SDLC once your software is an agent.
If you're trying to turn your team AI native, this is the one episode to watch.
10 Key takeaways
1. Manage agents instead of doing the work - Jayesh says the biggest change in his org is mental. People stopped asking how to do the work by hand. They now ask how to manage a team of agents that does it.
2. Keep tool choice open - His engineers started on Cursor, then moved to Claude Code. Every tool clears security review first, then teams get a choice for a while. Real usage data tells leadership what to back next.
3. The value was never the UI - Going headless puts Salesforce's business logic behind an API. Customers pull decades of tuned workflows into Claude through skills and build whatever interface they want on top.
4. Let deterministic tools hold the ground truth - You describe an agent in plain English and it transpiles to Agent Script, which spells out the tool calls and context lookups. The model orchestrates tested APIs instead of writing fresh code. Hallucination drops sharply.
5. Go eval-first - Jayesh's data science teams were eval native from day one. Now his engineering teams are catching up with custom scorers tied to outcomes and built for each specific agent.
6. Make Slack the context layer - Slack is the one place people are available asynchronously all the time, and that history is gold dust for agents. A PM can mock up a prototype in a Slack thread with a customer and get feedback before a PRD exists.
7. The product trio lines are blurring - Engineers now put prototypes in front of customers directly. PMs no longer wait on engineering. Team ratios haven't shifted yet, but Jayesh already sees work bleeding across roles.
8. AI PMs need prototyping and evals - Jayesh calls these two skills the magic bullet. When he hires, he looks first for agency, then for craftsmanship that refuses AI slop and turns skepticism into failing test cases.
9. Adopt the ADLC - The agent development life cycle covers how you build, test, evaluate, deploy and observe an agent. The model underneath keeps changing, so without evals you are too afraid of breaking things to upgrade. Salesforce's new optimization agent watches every live agent for drift.
10. Start simple, then go wall to wall - Define value and an ROI metric before you build. Minimize variables on the first implementation and get one win. One of Europe's largest recruiting firms started with text agents in 17 languages and is now moving to voice.
Go Deeper
How to Build a Full AI Dev Team in Claude Code
A Company OS on Hermes and OpenClaw
AI PM Crash Course with Aman Khan
How Zapier's CEO Grades PMs on AI Fluency
Listen and watch on Apple, Spotify, and YouTube.
If you want to advertise, email productgrowthppp[at]gmail[dot]com.
To hear more, visit news.aakashg.com! - This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit www.news.aakashg.com/subscribe
Brought to you by
Mobbin: The internet's best library of real design patterns
Get a full year of Mobbin Pro, along with 8 other AI tools, at bundle.aakashg.com.
Today's Episode
I quote Mark Pincus more than almost anyone, and I still feel a little guilty every time I do. For years, the internet calls him a "FarmVillain," the founder who copies his way to the top.
Today he owns it.
Mark (founder of Zynga, author of Life at the Speed of Play) steps up to my whiteboard and draws Proven Better New, the framework I reach for every time I plan a product. He shows me how Zynga Poker copies the leading tables down to the chip sounds and still wins on one small new bet.
Then he makes the case that most AI founders today are being too clever for the moment.
If you're building anything consumer right now, this is the one episode to watch.
10 Key takeaways
1. Copying is what every product maker already does - Mark argues that all music, movies, stories and games are derivative work. The real skill is the magic trick: you learn from someone else's proven success and make it feel new. Steve Jobs did it better than Bill Gates, even though both copied Xerox PARC.
2. Trust your instinct, doubt your idea - Mark puts your instinct at about 95% right and your specific idea at maybe 25%. Proven Better New lets you act on both at once.
3. Proven means pixel-level, for this audience on this platform - You copy the experience click by click, legally. "Instagram" can't go in your proven column. Today's best version of the specific experience can.
4. Better is what 10 out of 10 existing users would choose - What you think is better is actually new. Better is usually small: no download, free, fewer clicks, a cleaner onboarding.
5. Isolate your new so you fail for the right reason - Zynga Poker copies the leading tables down to the chip sounds, drops the download and adds one new thing, your real friends from Facebook and MySpace. Mark says friends end up at the same table 25% of the time.
6. Onboarding belongs in proven - Zynga's PMs call social Civilization and SimCity Social dead on arrival because of their long tutorials. In casual games you need to reach the good stuff in three clicks, not three minutes.
7. All new fails - Zynga still repeats this mantra. Expect your new to fail so you kill it fast, and treat new that works in the market with reverence, because it just became proven.
8. Early in a platform shift, ship less new - Supercell's Hay Day launches with 0% new, just a FarmVille built for mobile. In the AI cycle, users sit in discovery mode, so an AI version of something they already do beats an idea that's too clever for the moment.
9. Copy everything by default, then cut on audience - FarmVille copies Farm Town's gifting and its two-way neighbor bar, friction included. It cuts only the stranger marketplace, because that doesn't fit its core audience of middle-aged women on Facebook.
10. Test the top of the funnel before you build - Mark's problem with the MVP is the word viable, because it fills you with hope. Test names and web ads first, and set a click-through bar before you write real code.
Go Deeper
How you need to think about product strategy in the age of AI (with Tim Herbig)
The Ultimate Guide to Onboarding
Ultimate Guide: Activation
The Multiplayer Lever
Ultimate Guide: Growth Loops
The PM's Guide to Governance with Eric Ries
Listen and watch on Apple, Spotify, and YouTube.
If you want to advertise, email productgrowthppp[at]gmail[dot]com.
More Business podcasts
Trending Business podcasts
About The Growth Podcast
Join 500K+ for deep dives on AI + product management. After spending a decade plus in product, I now interview PM's most insightful experts. www.news.aakashg.com
Podcast websiteListen to The Growth Podcast, The Diary Of A CEO with Steven Bartlett and many other podcasts from around the world with the radio.net app

Get the free radio.net app
- Stations and podcasts to bookmark
- Stream via Wi-Fi or Bluetooth
- Supports Carplay & Android Auto
- Many other app features






































