Today’s Episode

90% of teams have adopted AI, yet 56% of CEOs say they saw no major financial benefit. Both metrics are accurate, and I’m pretty sure you fall somewhere between them.

With new AI tools flooding the market every week, it is safe to say that every team has its own AI setup by now. But are any of those setups in sync with each other?

I’ve been building this argument in stages. First, Carl Vellotti showed you how to build a personal OS, and Hannah layered it with building a team OS. Jiaona and Mikhail vouch for a full-blown company OS as well.

So, today you get a screen share.

My guests are the product team at Together AI that raised $800M at an $8.3B valuation and sells inference and fine-tuning to developers. Charles Zedlewski is their CPO who brought Necoline, Pavneet, and Hassan on the call to talk about how to build a shared context repo.

Their customers are already agents and their engineers are already agent-first. Their product team had no choice but to catch up.

Brought to you by

Arize - Ship AI agents and features faster with fewer regressions.

Get a full year of Arize, Bolt.new, Airtable, Speechify, Descript, Magic Patterns, Linear, Dovetail and Mobbin at bundle.aakashg.com.

10 Key takeaways

1. Individual productivity can move a company backwards - The team's starting question was not how to make each person faster. It was whether everyone generating unlimited code and content actually added up to progress. Charles called the failure mode flooding your coworkers' context windows, where everyone launches slop at each other.

2. The shared repo holds context and skills, not code - Markdown and YAML files covering customer intelligence, sandboxes, and the output of strategy meetings broken down by mission and milestone. Anything tied to a specific codebase stays out of it. The point is that a PM can read another team's context and draft a real proposal before taking up that PM's time.

3. Skills live closest to the work they touch - If a skill references code inside one team's repo, it stays colocated there. Everything else goes to a personal or shared repo. Test it on a branch, use it a few times, and only push to main once it proves repeatable. Niche ones never get pushed.

4. Shared context is a hierarchy, not a flat pool - The team abandoned the idea that everyone should carry everyone's context. Most people have no motivation to learn the depth of someone else's area. They want the one answer they came for. Some people live at the bottom of the hierarchy, most just traverse the top.

5. The PRD stopped being a gate - Historically it was the document everyone aligned on before building started. Together treats it as a trigger for ideation and problem solving instead. One to two pages, defining the customer problem, a few solution options, and a sample user journey. That is enough to argue about whether the thing is worth building.

6. A prototype replaces the bulk of the long document - A separate skill takes the one pager and produces a prompt for a design tool, and that visual is where the sharpest feedback shows up, from engineering and marketing alike.

7. Discovery collapsed from half a day to five minutes - The research agent pulls from the support platform, the project tracker, and internal docs at once. It surfaced 19 tickets filed in two months, flagged that the feature had been partially built and abandoned, and gave verbatim quotes with sources. The value is not the summary. It is not duplicating work someone already started.

8. Automate execution, keep decisions human - Defining the feature, the API surface area, and the abstraction layer stay hands on. Code writing is the part that runs on its own. The PRD skill is explicitly instructed to challenge the PM's assumptions rather than accept them.

9. Agents are already the majority user, so validate for them - Agent evals spins up a sandbox, gives an agent a real task against the product, and watches. It caught that agents could not find the fine-tunable models page because it was not linked from the quick start. Dozens of docs fixes came out of this. Charles calls agent success the new bar for UX.

10. They refused to oversell the gains - No story points, so no proof, but velocity is up more than 5%. Charles finds 3x claims suspicious, because discovery, debate, and coordination do not get magically better with AI. Costs stayed sane partly through open weight models, partly because optimizing for collective output never produced the runaway token budgets others report.

Go Deeper

As promised, I put together this section to help you become the best version in your product management journey. If you’re starting from zero on this, begin with How to build a Team OS in Claude Code with Hannah Stulberg, then scale it up with How to build a Company Operating System with Hermes and OpenClaw. For the skills layer specifically, 3x CPO Oji Udezue on the Essential Claude Skills for PMs is the one to read. And if section 3 sold you on agent evals, go to How to Build Frontier-Lab Quality Evals with Daniel McKinnon. Finally, Pavneet’s take on the one-page PRD lines up almost exactly with what Srini Raghavan showed from Freshworks.

You can get my podcasts on Apple, Spotify, and YouTube.

If you want to advertise, please 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

Today’s episode

Most PMs have automated something with AI by now. A PRD review. A weekly status update. The problem is that every session starts from scratch.

Loops are what fix that. A skill that runs the same way forever is just a skill. A skill that takes the log of what happened and rewrites itself is a loop. Everyone keeps saying loops are the new prompts. Almost nobody shows you how to build one.

Tyler Folkman is Chief AI Officer and Head of Product at JobNimbus. In this episode he builds a loop live on screen. He runs a preflight check on his own podcast recording, spins up three prototype variants while answering questions, writes a decision skill from scratch, and closes the loop on camera.

He also covers the loops every PM should be running, the hooks that stop Claude from doing real damage, and the point where vibe PMing breaks.

Brought to you by

Customer.io - Send smarter messages using your product data

Ariso - Ship AI agents and features faster with fewer regressions

Viktor - AI employee connected to 3k+ tools with every action approved by your team

Bolt - Build a complete design system from your code

Product Faculty - Get $150 off their #1 AI PM Certification with code AAKASH150

10 Key takeaways

1. A skill becomes a loop when you feed the learning back. A static skill sits there and stays the same forever. The loop closes when you hand the AI the log of the whole session and ask what to improve. Skip that step and you are just running a skill.2. The gate is the most important part of the loop. Fetch inputs, do work, pass a gate, write the artifact. The gate is where you validate the work was correct, and making it deterministic matters more than anything else in the chain.

3. Agents make decisions, prompts wait for you. The difference is not the model. An agent gets a task plus a way to verify itself and executes as far as it can alone. Prompting keeps you in the loop, which caps how far you can scale yourself.

4. Write the first version of a skill by hand. Human authored skills tend to beat AI authored ones because you know more about what you actually want. Once you are on the AI loop it moves fast enough that injecting your own thinking gets hard, like promising you will still pedal on an ebike.

5. In product, the gate is a customer, and that breaks the loop. Code loops run fast because the gate is a test. You cannot lock customers in a room and iterate on them. The workaround is mining the research calls and transcripts you already have to build a cheap first filter.

6. Prototypes are free now, so generate variants instead of one answer. The internal standard is at least three variants per idea. One minimal, one full featured, one creative. Then narrow a hundred ideas down to five before anyone talks to a real customer.

7. Synthetic customers filter, real customers decide. Customer research transcripts loaded into a warehouse let AI inspect a prototype as your customer. It is not a high bar, but it catches low hanging failures fast, which is the whole point of a gate.

8. Write docs for AI, keep human docs to three pages. Part of onboarding is not written for people to read. It is context so the AI can answer questions. Anything a human is expected to read should be one to three pages, visual, and cut down by hand.

9. Hooks add determinism that a prompt cannot. Telling Claude never to delete everything or never to share credentials only works if it reads that instruction today. A hook fires on the bash command itself. Session close hooks can also force the improvement step you would otherwise forget.

10. Ship AI output you have not thought about and it costs you. Passing unreviewed AI work up the chain just moves the effort to someone busier than you. Answering a question in a meeting with what Claude said is the failure mode. Use AI to push your thinking, not to replace the part where you do it.

Related Content

If you don't know where to start on Claude loops, I got you covered with my ultimate guide on Loops for PMs. And then you can follow it up with The Complete PM Guide to /goal in Claude Code. Tyler mentioned the evolving role of PMs and how it is shifting into a Product Builder role, so you can read all about it in my How to Become a Builder PM deep dive.

If you want to advertise, please email productgrowthppp[at]gmail[dot]com. 

P.S. Please subscribe on YouTube and follow on Apple and Spotify. It helps!

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

Today’s episode

Today I’m showing you how to build a company operating system, with Mikhail Shcheglov, CPO at OLX Classifieds.

After 5 months of continuous building on OpenClaw and Hermes, his entire product team now runs on it.

His knowledge graph covers 54% of the company's product, business and customer context. That is a number he tracks as a personal KPI. At that level the agent already makes backlog decisions. Stakeholders pitch feature requests to it before they’re allowed near a PM. It runs his email, his calendar, his recruiting funnel and his design system.

He opens his IDE on camera and shows all of it. 2 findings cut against everything you’ve been told. Summarizing your meeting transcripts costs you 20-25% recall, so he stores every single one raw. And letting Hermes write its own skills off repeated tasks produced a 31% accuracy lift in controlled testing.

We also get into something heavier than architecture.

What happens to PM headcount when one PM covers four domains? And what does a CPO actually screen for now when hiring?

Don't miss....

Brought to you by

Bolt.new - Ship AI-powered products 10x faster

Product Faculty - Get $150 off their #1 AI PM Certification: code AAKASH150

Customer.io - Send smarter messages using your product data

Land PM Job - 12-week live course to master the PM job search

Viktor - Use $100 in starting credits to get 5x more done with this AI employee

If you want access to my AI tool stack including Airtable, Speechify, Descript, Magic Patterns, Linear, Dovetail, Arize and Mobbin, that's $27,000 of value for $150, grab https://bundle.aakashg.com/. 

Key Takeaways

Context coverage is a CPO level KPI - Mikhail tracks what percentage of the company's industry, business model and customer knowledge his agent actually holds. It sits at 54%. That is enough for it to operate like a junior to mid PM and make backlog calls. At 70 to 90% he expects strategy level work.

The real problem AI solves is knowledge leakage - A domain expert leaves and takes five years of context with them. Every company has this hole and almost nobody measures it. One store of business, customer, product and technical knowledge closes it, and the better your AI knows that context the more you can hand it.

Do not summarize your transcripts - Summarization cost them 20 to 25% recall. You lose the granular detail where the answer usually lives, and you force every conversation into a template it was never shaped like. Store everything raw.

Memory needs three layers, not one - A knowledge graph for structure, a vector database for fuzzy retrieval, and raw daily transcripts in MD files. Exact keyword matching fails on most real queries because real questions are ambiguous. The vector layer carries the load.

Auto generated skills lifted recall by 31% - Hermes watches what you keep asking for and decides on its own that a skill is worth writing. Tested across five core topics with ten questions each, control group against treatment group. Plus 31% accuracy.

Imperatives matter more than prompts - Their rules file runs 700 lines. No fabrications. Think before you act. Facts over guesswork. And a ban on what he calls fake helpful, where the agent can't do the thing so it explains how you could do it yourself.

CLAUDE.md stays short, SOUL.md goes long - CLAUDE.md holds under 100 lines and carries the highest priority. SOUL.md runs 800 and sits second. Some of those 800 lines contradict each other and it still produces his most accurate output, because every imperative gets tested against real queries.

Make the agent the gatekeeper - Stakeholders are trained to pitch the agent first. It asks clarifying questions, checks the request against priorities already set, declines politely if it doesn't clear the bar, and routes it to the right PM if it does. The org chart is mapped internally so it knows who owns what.

Half of PM time is process, not thinking - Weekly reports, stakeholder updates, demos. Delegate that layer and one PM does the work of two, pointed entirely at discovery. He now runs one PM across three or four customer facing domains, and only keeps dedicated owners on monetization and search.

Own the agent yourself or lose the advantage - Feedback arrives daily and he pushes changes from his phone straight into the repo. Hand it to an AI ops hire or an engineering team and you keep the tool but lose the speed. Nobody without skin in the game iterates fast enough.

Related content

I’ve already created all of the resources Mikhail mentioned in this episode. But I never mapped them all to one place. That changes today in this edition. PFA the map sorted by the parts you want to build.

Memory first, I Built You Memory for Claude Code, Hermes and OpenClaw

Once you've got a memory layer, put it in version control. GitHub for PMs has the 3-repo setup I use for skills and eval

The Hermes Agent Guide for PMs is a 20-minute setup and a 30-day rollout with persona and skill templates included

For OpenClaw, start with Mahesh Yadav on becoming a Builder PM

For the org-wide version, watch my episode with Jiaona Zhang, CPO at Laurel, on building a Company OS

My conversation with Hannah Stulberg of DoorDash on building a Team OS is the same thing one floor down, with a free starter kit

👨‍💻 Where to find Mikhail Shcheglov:

LinkedIn: https://www.linkedin.com/in/scheglovm1/

Substack: https://corpwaters.substack.com/

👨‍💻 Where to find Aakash:

Twitter: https://x.com/aakashgupta

LinkedIn: https://www.linkedin.com/in/aagupta/

Newsletter: https://www.news.aakashg.com/

If you want to advertise, email [email protected].

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

Check out the conversation on Apple, Spotify and YouTube.

Brought to you by

* Customer.io - Send smarter messages using your product data

* Ariso - The AI operating partner for every manager and team

* Product Faculty - Get $150 off their #1 AI Builder Fellowship. Code AAKASH150

* Land PM Job - Join me for Cohort 5 starting November 16th

* Amplitude - Custom agents that monitor your funnel and file the ticket

Today’s Episode

Freshworks is a $3.4B giant of SaaS. They’ve been around since 2010. They have over 75,000 customers and 4,000 employees. Their 2026 revenue will be $960M.

They’re a colossus.

When Srini Raghavan joined them as Chief Product Officer, they had a 6 month release process. Under his tenure, they moved to a 2 week release cycle.

They embraced a new way of working powered by AI.

In today’s episode, he breaks down everything:

* The AI PDLC they embraced

* The AI Harness they created in Cursor, so it can run on any models

* How the PM role shifted to Product Builder

If you’re a product leader, this is a great example of how to become AI-native.

If you’re a PM, it’s a great harness (using Grok models!) to learn from.

I hope you enjoy it as much as I did:

Apple | Spotify. | YouTube

I’ve written up the key takeaways for newsletter subscribers as well.

1. The AI PDLC they Embraced

How do you go from shipping every 6 months to every 2 weeks?

It’s not about becoming AI first! Most teams try that: they open Cursor or Figma Make and start prompting.

Srini made the point that it actually all begins with being Data First:

When you’re data first, you create the right foundation to actually be AI first. At Freshworks scale, with 300 million end users, you can’t afford a hallucination.

For Freshworks, that was three things:

* A design system structured so agents can parse it

* Coding standards written down explicitly

* A single repo as the source of truth

They wrapped that all into a system they call Prism. It’s a knowledge hub that knows the product and its dependencies inside out, plus a context hub that passes feature context between phases, and a central library of what they call AI builder artifacts. These are the skills, rules, commands, and agents that describe how Freshworks specifically builds.

The AI PDLC sits atop all of that. And it looks like this:

It’s the same lifecycle every SaaS company runs from discovery through release, with two changes. First, there’s a governed AI agent working inside each. Second, there’s an evals phase at the end.

And that’s how their release cycle went from 6 months to 2 weeks.

But even in the episode, there were some cracks. Figma Make skipped a few design system components. Srini noted that those are the places where humans still have a role.

2. The AI Harness in Cursor they created

I began in engineering. After 14 years of not touching code, now I’m spending a lot of time in Cursor.

Srini is now spending lots of time in Cursor, and he showed us. Everything starts with a slash command, /fw-innit. Here’s what that kicks off:

The harness asks for the business unit, Epic ID, and feature team. Then it runs 12 phases, all the way from idea brief through to prototyping and QA.

Srini did the whole demo on Grok models! He explained:

I used to use Claude, but Grok seems to be working really well, and it’s really really fast. Nothing takes more than 10-15 seconds.

And indeed, that’s what we found. If you really need speed, consider Grok.

He live built a performance dashboard for their new EX Agent Studio. After he described the feature in plain English, the agent came back with questions a good PM would ask:

* How deep should the drill down go?

* Who is the primary persona?

* What does success look like in 6 months?

It then wrote its own SQL against Bel, their Databricks data lake with usage from all 75,000 customers. It assembled this into quantitative evidence in the PRD:

It looked across all 4,358 active ITSM accounts and showed its work.

At the end of the process is a step called CPO check which reviews the draft the way Srini would. His team named it after him.

From there, the PRD goes straight into Figma Make. Because of being data-first, they have a design system built into it. And after a few prompts we had a working prototype:

That’s the end-to-end process. No Claude Code or Codex. Just Cursor + any model.

3. How the PM role shifted to product builder

Which brings us back to where we started: what roles will execute work like this in the picture?

Srini manages 200+ people with titles like PM, UX researcher, and designer. His prediction is:

The titles Product Manager, Product Designer, and Engineer will go away in the next 5 years.

He didn’t say that to get clicks on Social Media. That’s how he’s running Freshworks. When he joined, Freshworks ran 1 PM and 1 designer for every 10 to 20 engineers.

Now, their new workflow is 1 PM to 1 engineer. Some teams have no dedicated designer at all.

Here’s why: those titles marked stations on an assembly line. The PM would write the PRD, hand it to design, and design would hand it to engineering. Now, like in section 2, one person can do the whole assembly line.

If you want to prepare for this future, I’ve put together the roadmap:

Pay close attention to step 8. Since one person can now run this assembly line, that person also owns whether the thing made any money.

In the full episode, he also covers how Freshworks’ agent builder works, how CPOs can replicate his playbook more. Check it out: Apple, Spotify., YouTube.

Get more of Srini

Freshworks is hiring and you can find Srini on LinkedIn. He also generously shared his deck.

Since the episode was recorded, Srini actually left Freshworks. With his tenure as CPO at RingCentral, SVP at Five9, and Director at Cisco, I’d expect you see him at another high-powered role soon.

Go Deeper

* Hear from more CPOs on becoming AI native: Rachel Wolan, Jiaona Zhang

* Learn more about becoming a Builder PM: Mahesh Yadav

* Explore other Coding Harnesses: Codex, Claude Code

PS. Please subscribe on YouTube and follow on Apple & Spotify. It helps!

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

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

Whether you’re using a PM OS, or using regular Claude, or using another AI harness (like ChatGPT Work or Codex)…

One of the most important parts of any PM’s AI setup is the skills you have.

Today’s guest Oji Udezue, CPO at Typeform, Calendly, and Parsable, and he goes into a masterclass of what skills PMs should have, how to build them, and how to use them.

Brought to you by:

Bolt.new - Ship AI-powered products 10x faster

Product Faculty - Get $150 off their #1 AI PM Certification: code AAKASH150

Customer.io - Send smarter messages using your product data

Ariso - Ship AI agents and features faster, with fewer regressions

Land PM Job - 12-week live course to master the PM job search

Get More of Oji

LinkedIn

ProductMind

The open-source skills repo

Shipyard

Building Rocketships

Go Deeper

I’ve done extensive testing on what makes a good skill and how to build a PM OS for paid newsletter subscribers (that go much deeper than this podcast).

→ Get the founding plan to get access.

I also have several free podcasts on Claude Code that you may find helpful: my 3 part series with Carl Vellotti, PM OS with Dave Killeen, Team OS with Hannah Stullberg, Company OS with Jiaona Zhang, and Claude Code for CPOs.

→ To never miss an episode, subscribe on YouTube and follow on Apple & Spotify.

Finally, in my live course, I teach you how to do extremely advanced PM automation Claude Code.

→ Join us.

Today’s Episode

A developer posted this workflow in March, and it is the clearest picture of where PM is heading that I’ve seen all year.

Rasty Turek spent the past year building with coding agents, and he mapped how his process changed over that time. He reckons he now spends around 90% of his time on evals. His eval started as QA, and then it became the spec.

Great, now everyone agrees evals are important and will become indispensable for PMs going forward. But there is very little on how to write one.

That changes today.

I’ve now done 6 episodes on evals, and all of them start with an agent that is running and failing. So what do you do on day 0?

Daniel McKinnon was a PM on the Llama models at Meta, a boomerang who spent around 7 years there in total. He sat on Facebook’s central AI team for the entirety of its existence. He wrote enterprise evals for Gemini, Llama, and Ray-Ban Meta.

His first job at Meta was on the speech recognition team. He had to figure out how to check whether the models were any good. They weren’t called evals back then. But he’s been writing them for his entire career anyway.

In this episode you’ll learn:

* How to build an eval set from nothing

* The floor-and-ceiling method for calibrating

* How to score it and make the shipping call

Check it out:

Please fill out this short survey on PM salaries.

🆓 I’m doing a free webinar Thursday on getting AI PM interviews. Join me:

The next cohort of my LandPMJob program starts in August. If you want my 1:1 coaching, sign up.

----

Check out the conversation on Apple, Spotify, and YouTube.

Brought to you by:

* SerpApi - Get started with SerpApi using 250 free credits.

* Product Faculty - Get $550 off their #1 AI PM Certification with code AAKASH550C7

* Ariso - Ship AI agents and features faster, with fewer regressions

* Land PM Job - 12-week experience to master getting a PM job

* Pendo - The #1 software experience management platform

----

Key Takeaways:

1. An eval is a trivia question for the model - At its core, an eval is a prompt with a correct or plausibly correct answer plus a way to score whether the output is good. It is the clearest way to communicate what your product should do in the AI era.

2. Offline evals catch problems before you ship - Test the model offline against a fixed prompt set before pushing to production. If it fails, you change the model, the prompt, or the approach before real users ever see it.

3. The best eval sits between too easy and too hard - An eval that scores 100% gives your engineering team nothing to optimize. An eval that scores 0% is equally useless. Aim for a 25% to 50% success rate so there is room to run.

4. Old benchmarks are already saturated - MMLU, HellaSwag, ARC and the rest were built for a simpler question-and-answer world. Frontier models now score effectively 100% on them, which is why you have to keep building new evals and throwing away old ones.

5. Writing an eval is mechanical once you understand the problem - Come up with roughly 100 prompts that match the real distribution of tasks. The hard part is not the writing. It is deeply understanding the domain first.

6. Subject matter expertise drives everything - The cystic fibrosis and congenital heart disease evals worked because Daniel understood the genetics, not because of any template or tool. There is no eval template the way there is a PRD template.

7. Modern evals are agentic, not just Q&A - The genetics eval hands the agent a file with billions of variants and asks it to find the cause of a disease. This is a task, not a lookup, and it mirrors how real AI products now work.

8. Find the model ceiling on purpose - The easy cystic fibrosis case gets solved by most models. The harder digenic congenital heart disease case exposes where even strong models fail. Knowing the ceiling is the point of the exercise.

9. Sample multiple times before you trust a result - Models are non-deterministic. Run the same task several times so you understand the real distribution of outcomes rather than a single lucky or unlucky pass.

10. Meta and Google build products very differently - Google is seen as more engineering-led, Meta as more product-led and far more aggressive culturally. Daniel worked on both Gemini and Llama and saw everything from Llama 3 highs to Llama 4 lows.

----

Where to find Daniel McKinnon

* LinkedIn

* X

* Gamow Labs

Related content

Podcasts:

* AI Evals with Hamel Husain and Shreya Shankar

* How to Run Evals in Claude Code with Aparna Dhinakaran

* Evals are the New PRD with Ankur Goyal

Newsletters:

* AI Evals for PMs: Everything You Need to Know to Get Started in 2026

* AI PM’s Guide to LLM Judges

* AI Evals Explained Simply

----

PS. Please subscribe on YouTube and follow on Apple & Spotify. It helps!

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

Check out the conversation on Apple, Spotify, and YouTube.

Brought to you by

* Land PM Job - 12-week experience to master getting a PM job

* Jira Product Discovery - Plan with purpose, ship with confidence

* Amplitude - The market-leader in product analytics

* Bolt - Ship AI-powered products 10x faster

* Product Faculty - Get $550 off their #1 AI PM Certification with code AAKASH550C7

----

Today’s episode

Every startup founder picks up Zero to One. Reads the chapter on Delaware C-Corps. Files the paperwork. Moves on.

That paperwork will outlast every product decision they ever make.

I sat down with Eric Ries, the man who created Build, Measure, Learn. NYT bestselling author of The Lean Startup. Co-founder of Answer.AI with Jeremy Howard. Founder of the Long-Term Stock Exchange. Dario Amodei called him before Anthropic’s seed round.

His new book Incorruptible drops May 26. It is the blueprint for building a company that the financial system cannot capture.

He also demoed live how he wrote the book using Solve It, the AI platform from Answer.AI. Not prompting. Not generating. Editing the model’s responses directly.

If you are building anything you want to outlast the next funding round, this is the one episode to watch.

Check out the episode on Apple Podcast and Spotify.

If you want access to my AI tool stack, grab Aakash’s bundle.

----

Key Takeaways:

1. Governance has four dimensions - Compliance is table stakes. Purpose, coherence, and integrity are the three most boards ignore. Companies that nail all four outperform the market over decades.

2. Financial gravity destroys good companies - The unconscious reflex to comply with the values of those who have more than you. Jim Senegal called it heroin. You compromise once and it gets baked into the forecast.

3. Costco's governance fortress is the blueprint - Staggered board terms, poison pills, fiduciary hierarchy. $10K at the Costco IPO is worth $8.7M today versus $151K in the S&P 500.

4. Stone does not enforce itself - Johnson & Johnson carved values into limestone. Asbestos ended up in the baby powder. $10B settlement. Structure protects ethos but does not create it.

5. Mission lock vehicles create 6x survival - A separate entity holding the for-profit board accountable. Novo Nordisk, IKEA, Patagonia, Hershey, Vanguard all use this structure. 60% survival to year 50 versus 10%.

6. Anthropic's LTBT took two years to defend - AI safety experts appoint board seats. The trust gains power as the company hits milestones. Structural protection is why Anthropic can afford to be courageous.

7. Public Benefit Corporations write mission into the charter - Legal permission to pursue purpose over shareholder value. Not the B-Corp certification sticker. A legal structure.

8. LLMs are conformity machines - They produce the center of the outcome distribution. For competitive advantage you must change how you use AI.

9. Solve It enables human-in-the-loop writing - Edit the model's responses directly. 600 test readers, 10K structured comments, Python scripts organizing feedback per chapter.

10. Build Measure Learn works at any timescale - Not about absolute speed. Relative velocity versus your industry convention. The AI labs that release more quickly create decisive trust advantages.

----

Where to find Eric Ries

* LinkedIn

* Incorruptible

* Answer.AI / Solve It

* Long-Term Stock Exchange

* Lean Startup

----

Related content

Podcasts:

* AI Product Strategy with Aman Khan

* The Marty Cagan Episode on Product Management

* How to Succeed as a Head of Growth with Dan Olsen

Newsletters:

* AI product strategy in 2026

* How Cursor grows

* PLG in 2026

----

PS. Please subscribe on YouTube and follow on Apple & Spotify. It helps!

If you want to advertise, email productgrowthppp at gmail.

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

Today’s episode

“This one’s too complex. I’m still stuck on ChatGPT.”

I get some version of that DM every week, usually right after I publish something on the PM OS or the Team OS. And every time, I feel it, because those guides do assume you’re already up and running.

So I made this episode for the person sending the message. Jyothi Nookula has been an AI PM since before that was a title, through Netflix, Meta, and Amazon, and I asked her to take a PM from zero to eighty on the entire Claude stack in one sitting.

She walks through:

The five-layer Claude stack, so you finally know which surface and which model to reach for and when

A chief of staff you build in Claude Code that reads your meetings and quietly learns your org, your people, and your politics

The self-improving agent loop she used to beat 30 engineering teams at an internal hackathon, as a PM

Everything she shows is something she runs at work, which is the only reason any of it holds up.

Send this to the PM in your life who keeps saying they’re behind. Two hours from now, they won’t be.

----

Brought to you by:

Hyper Agent: Turn your recurring PM work into reusable agents

----

If you want access to my AI tool stack - Dovetail, Arize, Linear, Descript, Reforge Build, Relay.app, Magic Patterns, Speechify, Bolt.new and Mobbin - become an annual subscriber ($150), and grab Aakash’s bundle.

If you want access to my AI PM customizations - PM OS, Job Search OS, and Prompt Library - become a founding subscriber ($250).

----

Key Takeaways:

1. Match the model to the job - Sonnet handles ninety percent of PM work at the best cost. Save Opus for genuinely hard reasoning, and hand fast bulk jobs to Haiku. Defaulting to the smartest model for everything just burns time and money.

2. Stop skipping the knowledge layer - Projects, skills, and memory are what make Claude know your actual work instead of guessing from a blank slate. Almost everyone underinvests here, and it is the difference between a chatbot and an assistant that knows you.

3. A skill beats a prompt - A skill is a saved playbook Claude picks up on its own when it fits the task. It only loads when needed, so it never clogs the context window. Build one once and stop re-explaining the same task forever.

4. Write your skills yourself - Human-written skill files consistently beat AI-written ones. Draft with Claude to move fast, then layer in the domain knowledge only you have. That last step is what makes it actually work.

5. Automate your time based work - A morning brief, a standup summary, and an end-of-day wrap can all run on a schedule while you sleep. You walk in already knowing what needs your attention. It clears the busywork that eats your mornings.

6. Give your automations guardrails - Cap the length, tell them to stick to facts, and never let them hallucinate. Left unchecked, an AI brief will pad itself and invent things. A few hard rules keep it sharp and trustworthy.

7. Build a chief of staff that learns your org - Point Claude at your meeting notes and let it build a picture of your people, your priorities, and your politics over time. Feed it transcripts first, since they carry the richest signal. It compounds into something no generic chatbot can match.

8. Keep that knowledge base on your own laptop - Your most personal work data does not belong in someone else's cloud. When you leave a company, it walks out with you. You keep full control of your most sensitive context.

9. The PM job is changing fast - The ratio is shifting from one PM per eight engineers toward two PMs per one. Building is becoming part of the role, and the PMs who can ship are pulling ahead.

10. Building is easy now, taste is scarce - When anyone can build, the edge moves to knowing what is worth building and what good actually looks like. That judgment is the one skill you cannot download.

----

Related content

Github repo: https://github.com/fibbonnaci/ai-builder-skills

Podcasts:

How to Become an AI PM - YouTube | Spotify | Apple

How a VP Uses Claude Without Producing Slop - YouTube | Spotify | Apple

This CPO Uses Claude Code to Run His Entire Work - YouTube | Spotify | Apple

Newsletters:

I Built You Memory for Claude Code, Hermes, and OpenClaw

I spent 100s of hours building a PM OS for you

How to build a Team OS in Claude Code

----

Where to find Jyothi Nookula:

LinkedIn: https://www.linkedin.com/in/jyothinookula/

NextGen Product Manager: https://nextgenproductmanager.com/

Where to find Aakash:

Twitter/X: https://x.com/aakashgupta

LinkedIn: https://www.linkedin.com/in/aagupta/

Newsletter: https://www.news.aakashg.com

---

PS. Please subscribe on YouTube and follow on Apple & Spotify. It helps!

If you want to advertise, email productgrowthppp at gmail.

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

Today’s episode

Today I’m showing you, advanced Codex workflows with Meng To, founder of Design+Code.

He told me early on that he’s barely touched Claude. He’s been living inside Codex every day since launch, running a setup most PMs have never seen up close.

Plan mode and a fleet of 20 agents running at once while he steps away from his desk entirely. Slides, charts, and full brainstorms generated inside the same chat window, reviewed, regenerated, and shipped, all without writing a line of code.

We also get into something heavier than workflows.

Why PMs are losing their jobs to layoffs right now? And how to future proof your career?

Don’t miss....

----

Brought to you by:

Arize: Trace, evaluate, and fix your AI agents before broken behavior ships to users

----

If you want access to my AI tool stack - Dovetail, Arize, Linear, Descript, Reforge Build, Relay.app, Magic Patterns, Speechify, Bolt.new and Mobbin - become an annual subscriber ($150), and grab Aakash’s bundle.

If you want access to my AI PM customizations - PM OS, Job Search OS, and Prompt Library - become a founding subscriber ($250).

----

Key Takeaways:

1. Codex is a fleet operator - Meng runs 20 agents at once while stepping away from his desk entirely. Each one works on something different, slides, charts, brainstorms, while he does something else completely.

2. Plan mode isn't optional - Skipping it means paying twice, once to build the wrong thing, once to undo it. Codex returns a full breakdown, architecture, steps, and open questions, before touching anything.

3. A screenshot beats a paragraph every time - It shows the AI what you mean instead of what you think you mean. A simple two-key shortcut drops any window straight into the chat as context.

4. The taste skill is the real differentiator - Without it, AI design defaults to generic. With it, the output looks like something a senior designer with years of experience actually made.

5. Trust is earned in tiers - Read only first, then supervised access, then full access. Skipping straight to full access before learning where the AI tends to fail is how people get burned.

6. HTML beats Figma for speed - Every extra tool is a login, a subscription, and a context switch the AI can't do for you. Keep your blast radius small.

7. UGC won because audiences are tired of corporate polish - A synthetic version of you, used honestly, reads as more human than a generic message. Ten old photos is all it takes to build a digital twin.

8. Technical PMs aren't surviving layoffs because they write code - Meng hasn't written a single line in six months. They're surviving because they're fluent enough to direct a fleet of agents and catch a wrong output before it ships.

9. Meng builds his own tools when nothing off the shelf fits - His own video editor, his own SaaS templates, his own design brainstorming app. The tool built for your exact workflow beats the popular one every time.

10. The bar isn't five star anymore - Five star is just the floor everyone clears by default now. The real question is what six, seven, all the way to eleven star looks like, because that ceiling rises exactly as fast as the floor does.

----

Related content

Podcasts:

How to Design with AI - YouTube | Spotify | Apple

How to Use Codex Like an OpenAI PM - YouTube | Spotify | Apple

The Ultimate Guide to ChatGPT Codex - YouTube | Spotify | Apple

Newsletters:

OpenAI’s Codex is the Best Way to Use ChatGPT

I spent 100s of hours building a PM OS for you

How to build product strategy in the age of AI

----

👨‍💻 Where to find Meng To:

LinkedIn: https://www.linkedin.com/in/mengto

Design+Code: https://designcode.io

Aura: https://aura.build

👨‍💻 Where to find Aakash:

Twitter/X: https://x.com/aakashgupta

LinkedIn: https://www.linkedin.com/in/aagupta/

Newsletter: https://www.news.aakashg.com

---

PS. Please subscribe on YouTube and follow on Apple & Spotify. It helps!

If you want to advertise, email productgrowthppp at gmail.

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

Today’s episode

Product teams have figured out AI for engineering. The PMs are using Claude. The engineers are in Cursor. If you’ve been reading this newsletter from the start, you’ve seen how the top 1% are using AI to 10x their output.

But there are still teams at companies like Adobe, teams in sales, customer success, finance, who don’t have access to any of these tools. You ask them what AI to use for their next task, and you get a blank stare.

That gap is the real problem. And it compounds every day.

Jiaona Zhang “JZ” has built the fix. She is the CPO at Laurel, which just raised $100M in Series C, and she has led product at Airbnb, Dropbox, Webflow, and WeWork. Today she runs a product team that ships frontend and backend features end to end, without any engineering handoff.

In this episode, she screen-shares everything. Laurel’s full Company OS built in GitHub, with skill files for every function from CS to legal to finance. The playbook to agent pipeline that turned 50-page docs into automated workflows. The daily Slack briefing that tells every person exactly what to do and which skill to use when. And a ton more.

This is one of the densest episodes I’ve ever recorded. The knowledge per minute is as high as it gets.

Don’t miss.....

----

Brought to you by:

Ariso - Ship AI agents and features faster, with fewer regressions

Bolt - Ship AI-powered products 10x faster

Pendo - The #1 software experience management platform

Product Faculty - Get $550 off their #1 AI PM Certification with code AAKASH550C7

Customer.io - Send smarter messages using your product data

----

If you want access to my AI tool stack - Dovetail, Arize, Linear, Descript, Reforge Build, Relay.app, Magic Patterns, Speechify, Bolt.new and Mobbin - become an annual subscriber ($150), and grab Aakash’s bundle.

If you want access to my AI PM customizations - PM OS, Job Search OS, and Prompt Library - become a founding subscriber ($250).

----

Key Takeaways:

1. Every company has a 1% who are AI-native and a 99% who do not know what to use when. The Company OS closes that gap by encoding the 1%'s workflows into skills that anyone can use when they open Claude.2. Build the ontology before you build the OS. Map every team's work to categories and tasks first. Color-code what should get more human time vs what gets automated. The OS is built from that work map.3. Even the friction of going to a different interface kills adoption. A separate agent tool in a new tab will not get used consistently. Deliver skills and automations inside Slack and email, where people already are.4. When AI adoption is everyone's responsibility, it is no one's responsibility. Dedicate one person full-time to AI Operations. Start with one person who demonstrates value. Every other function will want their own version within months.5. The Company OS turns a 50-page playbook into a set of agents. Write the playbook first. Then audit it. What requires a human? What can be automated? Build the skill files from what remains.6. The captain model replaces the handoff chain. Every feature has one owner end-to-end. The captain is whoever has the most critical skill for that feature's hardest problem.7. PMs at Laurel ship front-end and back-end features. Not just growth experiments or copy changes. Core product features deeply integrated with billing systems and time entry logic. One PM who identifies as a designer shipped one of these end-to-end last month.8. JZ went from hundreds of reports to 5 PMs and 4 designers. They ship more than ever. Adding people adds coordination cost. In a world where one PM can take a feature from discovery to production in a day, large teams cancel out their own capacity gains.9. The new PM interview is a screen share. JZ asks every candidate to show their actual screen. In 60 seconds she knows their level of AI skills.10. The PM fundamentals never changed. Problem space first. Know why and for whom you are building before you build. The speed changed dramatically. What you are supposed to be doing at the heart of it did not.

----

Related content

Podcasts:

How a VP Uses Claude Without Producing Slop - YouTube | Spotify | Apple

How to Build a Team OS in Claude Code - YouTube | Spotify | Apple

How to Become a Builder PM - YouTube | Spotify | Apple

Newsletters:

I spent the last week building an OS in Claude Code

I spent 100s of hours building a PM OS for you

How to build product strategy in the age of AI

----

Where to find Jiaona Zhang

LinkedIn - https://www.linkedin.com/in/jiaona/

Reforge - https://www.reforge.com/profiles/jiaona-zhang

Laurel - https://www.laurel.ai/

Where to find Aakash:

X - https://x.com/aakashgupta

LinkedIn - https://www.linkedin.com/in/aagupta/

Newsletter - https://www.news.aakashg.com

---

PS. Please subscribe on YouTube and follow on Apple & Spotify. It helps!

If you want to advertise, email productgrowthppp at gmail.

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

Today’s episode

There are hundreds of guides on writing PRDs with Claude. Dozens on running user interviews. Almost nothing on how a VP of product actually uses it.

Matt Wensing is VP of Product and Design at Customer.io. They crossed $100M ARR, just shipped an AI agent, and are one of the fastest growing companies in B2B SaaS right now. I asked him to show me his actual documents, his actual Slack threads, and the exact sessions where Claude helped him produce leadership grade output.

What he showed me changed how I think about AI for leaders. Claude has the instincts of a brilliant new hire, it wants to deliver before it fully understands what you need, and at the VP level that gap shows up fast. Matt has spent months figuring out how to manage it, and in this episode he shows you everything.

----

Brought to you by:

LogRocket - Find the bugs killing your conversion before your users do. I ran a head-to-head eval to see if that's true.

----

If you want access to my AI tool stack - Dovetail, Arize, Linear, Descript, Reforge Build, Relay.app, Magic Patterns, Speechify, Bolt.new and Mobbin - become an annual subscriber ($150), and grab Aakash’s bundle.

If you want access to my AI PM customizations - PM OS, Job Search OS, and Prompt Library - become a founding subscriber ($250).

----

Key Takeaways:

1. Take inventory before you open Claude - Before building anything, list every piece of raw material you already have. Zoom recordings, strategy docs, past presentations. The quality of what you feed Claude determines the quality of what comes out.2. Pivot content, do not write from scratch - Claude's best use case is transformation, not creation. Give it two inputs and ask it to reorganize one into the shape of the other. Matt calls this matrix multiplication.3. Build slides first - Build the visual story first. Screenshot the finished slides and feed them back into the same Claude session. Ask it to write a talk track that adds depth using all the context it already has, not one that just repeats the slide.4. Kill eager suggestions immediately - The moment Claude asks if you want it to generate the next thing, say stop. You control the pace. A 200-iteration session with a great deliverable beats saying yes to the first draft every time.5. Start sessions in the abstract - If you reveal the domain too early, Claude pattern matches to the nearest template. Keep it abstract. Build a clean mental model first. Reveal the domain only when the framework holds up on its own.6. Layer complexity in slowly - Start with the simplest version of the framework. Let Claude stabilize on the basics before you add exceptions. Dumping everything in at once produces a lost in the woods experience for both of you.7. AI alignment decks always backfire - When you one-shot an alignment deck, you flatten the problem. Senior executives have spent months living with the real complexity. They feel the thinness immediately, even when they cannot say why.8. Decompose the problem before building anything - Challenge yourself to explode a nasty problem into all its pieces before you touch Claude. Put those observations into the context window first. Then assemble the solution.9. The Slack scanner keeps leaders close to the ground - Customer.io built an AI scanner that monitors dozens of Slack channels and surfaces threads where a product person should be involved. It runs continuously without overwhelming. 10. Chiefys audits your strategy docs automatically - Chiefys is a Slack bot that holds Customer.io's ratified company documents and checks new work against all of them. It flags contradictions and stale documents so nothing goes invisible after you ship something new.

----

Related content

Podcasts

PM’s Guide to Claude - YouTube | Spotify | Apple

How to Become a Builder PM - YouTube | Spotify | Apple

We Built an AI Product Manager in 58 mins - YouTube | Spotify | Apple

Newsletters

How to Use Claude for Work

How to Build Product Strategy with Claude Code

The PM OS

----

Where to find Matthew Wensing:

1:1 Video Consultation: https://intro.co/MattWensing

LinkedIn: https://www.linkedin.com/in/wensing/

X: https://x.com/mattwensing

Where to find Aakash:

X: https://x.com/aakashgupta

LinkedIn: https://www.linkedin.com/in/aagupta/

Newsletter: https://www.news.aakashg.com

---

PS. Please subscribe on YouTube and follow on Apple & Spotify. It helps!

If you want to advertise, email productgrowthppp at gmail.

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