We can't find the internet
Attempting to reconnect
Something went wrong!
Hang in there while we get back on track
AI sped up feature teams without fixing outcomes
In a conversation with Joe Murray, Marty Cagan argues that companies running a stakeholder-driven feature roadmap used AI to "turbocharge that same process." Building got fast, but they are "going faster than ever, but not seeing improved results" . His diagnosis: the real problem has always been figuring out what to build, and the cost of development hid that . In the product operating model, leaders assign problems with outcomes, and empowered teams find solutions that are valuable, usable, feasible and viable. AI helps those teams with discovery as well as building .
Points PMs can use:
- Value and viability are the risks that sink efforts. Value means solving the problem better than competitors do, so customers actually switch. Viability has to hold up for legal, compliance and sales, not only for the stakeholder who asked .
- Treat synthetic users with care. AI lets teams prototype and test faster than before. But synthetic-user results are "just as good as the source," and Cagan says there's "a lot of garbage out there" .
- He doesn't believe in the "SaaSpocalypse." Serious SaaS apps contain thousands of business rules, often undocumented, and he hasn't seen a single predicted AI-built ERP replacement in three years . Instead, vendors are opening their platforms to agents, which he thinks makes them more valuable . Customers weighing a build-it-yourself option often count only development cost and ignore operating cost .
Hiring data favors PMs; UX research keeps falling
Lenny Rachitsky's fifth twice-yearly job market report finds that PMs are "winning in the AI era." It also finds that engineering demand has stalled, design openings are at a record low, every role now requires AI skills, and remote roles are at a record low . A follow-up chart shows data science openings more than doubled from their 2023 lows, while user research openings are down 85% from 2022, a record low . Chamath Palihapitiya reads the data as showing that narrow skills are in less demand than "the ability to solve more generalized, abstract problems" .
Aakash Gupta shared Levels.fyi submission data from Zuhayeer Musa as a directional proxy, not a headcount. It shows about 30 engineers per PM at Apple, 12 at Google, 6.6 at Microsoft, 3.6 at Walmart, and 2.6 and 1.8 at Amex and Mastercard . Gupta guesses that coordination and compliance work pushes the payments companies toward more PMs. He calls AI "the big variable" in where the ratio goes from here .
Discovery: ask about the struggle, not "why"
A PM building an opportunity solution tree for a dev-tools product found that all 12 interviews produced feature requests like "needs bulk export." Their plan was to ask "why" on every sticky note . A reply pushed back: the tree exists to find underserved needs, not to reword requests until they sound like problems. Instead, have users walk you through what they were trying to do, what got in the way and how they worked around it. Keep the tree going as long as the outcome matters, and prune it as you learn . This month, Teresa Torres's group read of Continuous Discovery Habits covers Chapter 11, on choosing analytics and tying them back to the outcome .
Roadmap access is a trust problem
A PM whose VP wants an internal roadmap that can't be shared or screenshotted got little sympathy on r/ProductManagement. One commenter said a roadmap is a communication tool and asked what the VP actually wants to achieve . Another called it fear winning over trust. Their fix: label any display with the date it was made and how often it gets reviewed .
AI output that's almost good enough
After 20+ years of launching software, Hiten Shah says he nearly shipped an AI-made launch video that barely explained the product. He then studied 8 recent launches and wrote a playbook . A B2B cybersecurity PMM described the same risk in messaging: "almost good" AI assets look fine to non-experts, so they get used. Fast production also leads to sprawling specialized messaging, and the long central guide now mostly serves as a source for AI tools rather than something people read .
Briefly
- Muse: The original team was 10 people, kept small on purpose so communication and decisions stayed tight . Sarantakos says she can now go from idea to built prototype to real discussion "in a 24-hour cycle" .
- Agent coordination: On a16z, Alex Imas cited benchmarks where one agent beats two agents splitting the same task, while two people outperform one person with double the resources. Today's agent swarms mostly run tasks in parallel rather than coordinating .
- AI can make stakeholder-led feature-roadmap teams ship faster without improving outcomes; Cagan contrasts this with a product operating model where leaders assign problems and outcomes, and empowered teams find solutions that are valuable, usable, feasible, and viable.
- Product discovery should establish that a solution addresses the problem better than alternatives—not merely confirm that the problem exists—and assess customer value and business viability alongside usability and feasibility. AI speeds prototyping and testing, but synthetic-user tests are only as good as their source data and generation.
- Treat internal and behind-the-scenes software, including systems used by agents, as products when they affect customer experience. For difficult client problems, forward-deployed empowered teams need direct access to end users, relevant data, and stakeholders; the team can include product managers, designers, and engineers.
- On SaaS replacement, Cagan argues that serious applications are difficult to recreate because they contain thousands of often-undocumented business rules, and he says he has not seen the predicted AI-built replacements materialize. SaaS vendors are instead opening functionality to agents, which he says can make their services more valuable; vendors still face pressure to demonstrate unique value, while buyers should consider operating costs and feature bloat—not just development costs—when weighing in-house alternatives.
- Product practitioners can configure an LLM as an always-available coach for product methods, technique selection, and outcome measurement; configure it to challenge assumptions rather than provide affirmation.
Product taste is now the most important PM skill, according to Muse’s head of product, who says the bar for PM has changed. Muse’s team spent seven months building a product “safe and simple enough for everyone” before launch; the effort followed Mona Sarantakos’s experience of an agent pulling together her children’s school activities from email and calendar after she granted it access, despite her husband’s safety concern.
Lenny’s 2026 tech-job-market report says PMs are “winning in the AI era” and AI skills are now required across roles; it also flags stalled engineering demand and record-low design openings. Data science openings more than doubled from their 2023 lows, while user-research openings fell 85% from 2022 to a record low.
- Nir Eyal argues that consumer psychology can differentiate a product: understand why customers behave as they do and serve them better than competitors who simply copy one another.
- Use behavioral design to help users do things they already want to do—such as exercise or learn a language—and make it possible to disengage from products they do not want to form habits around. Eyal distinguishes this persuasion from coercion, which he calls unethical and says can drive customers away and damage a business.
- Muse waited seven months to launch: after building a working end-to-end product in a few months, the team spent the remaining time on reliability, safety, and security, guided by “Trust is a must.”
- Dogfooding overturned the team’s expectations about key use cases: internal testers made shopping a prominent use case, and their bug reports prompted the team to build shopping-specific evaluations.
- To analyze long, multi-turn conversations, Muse used custom agent swarms on anonymized conversations and feedback to find what worked and where users got blocked; people still reviewed the results, and the system prompt had to be written by a person.
- Muse’s original team had 10 people, kept deliberately small to make communication and decisions tighter.
- Head of Product Mona Sarantakos says product taste is now the first quality she looks for in a PM: represent customers, hold a high bar, and be willing to challenge an engineer’s favored idea. She says PMs can now build an idea themselves and discuss it within a 24-hour cycle.
Lenny shared a setup pattern for having an agent use Claude through your own Anthropic API account: collect the API key (and optional workspace ID) through a secure field, store it locally, list available models and test a call; then have the selected model draft while you review and fact-check, with a fallback to self-drafting if API calls fail.
Mona Sarantakos first tried OpenClaw on a Raspberry Pi with access to her email and calendar; despite a family member’s safety concern, it quickly proved useful by pulling together her children’s school events for the week. The team then spent seven months building Muse to make the experience safe and simple before launch, even though it had started ahead of competitors.
The Academy SF argues that dropping out or attending college should not be the only options, framing a third-path education proposition. Andrew Chen expressed excitement for Gagan Biyani, Erik Torenberg, and the HAA team.
Teresa Torres is organizing a group read of Continuous Discovery Habits to help practitioners turn reading into practice, with one section per month, discussion questions, exercises, and resources . Chapter 11 focuses on measuring the impact of what you build: choosing analytics to collect, tying them to the outcome, and avoiding common measurement mistakes .
Idealab launched idealab.ai, a free AI tool that uses a corpus of 17 million documents—including patents, SEC filings, product recalls, online complaints and news—to surface business opportunities; the speaker says it produced 4.4 million opportunities and more than a million business ideas after filtering and checks such as whether the timing was right. Users can explore ideas by their interests or profile, submit an existing idea or business plan, and companies can look for uses of their patents and other IP.
The planned product goes beyond idea discovery: its proposed “periodic table of business functions” has 118 functions, starting with an idea and demand validation, with agent capabilities to be developed alongside partners. The tool is at an early stage, and its creators are asking users to test it, break it and report what is missing.
- For retail-tech discovery, one commenter argued that lacking retail credentials matters less than gaining customer access: secure recurring store-owner conversations and access to real operating data, observe work on-site, and defer hiring a retail cofounder until that access is established.
- A suggested discovery-to-pilot workflow is to list target companies, reach out, pitch, gather feedback, revise, and ask for an in-store trial. The builder reported speaking with 30+ companies and said two agreed to test the prototype in their stores.
- Commenters differed on how to assess founder-market fit: one cautioned that free feedback is weaker evidence than a pilot converting to a paying customer, while another called founder-market fit important to investors and cited relevant prior experience or paying-customer traction as evidence.
After 20+ years launching software, Hiten Shah nearly approved an AI-made launch video that barely explained the product; he then studied eight recent launches and wrote a playbook, highlighting the need to scrutinize AI-generated launch assets for clear product explanation. He also announced a free AI video-editing session covering prompts, first renders, revisions, and lessons from editing footage and building a launch video with AI.
- AI agent value depends on coordination, not just task competence: the speakers argue that organizations and teams—not individual workers—are the relevant economic comparison, and that current agent swarms often parallelize work rather than coordinate effectively. They describe coordination benchmarks where one agent can outperform two agents splitting a task, unlike human pairs that divide labor .
- Agent adoption faces both reliability and integration barriers: agents can make poor choices on human-oriented websites, while platforms may resist third-party shopping agents that bypass their upsell design; some users also prefer manual calendar entry to keep their schedules in mind .
- Usage growth does not guarantee revenue growth: the speakers say agentic AI consumes more tokens, but cite exponentially rising token use alongside flat revenue as token prices fall amid stronger competition. They also say broader vendor revenue growth depends on productivity gains and willingness to pay across the economy, for which they see limited evidence so far .
Lenny’s 2026 tech job market report lists PMs as “winning in the AI era,” while engineering demand has stalled, design openings are at a record low, and every role now requires AI skills. Chamath’s interpretation is that generalized, abstract problem-solving is increasingly valued over narrow skills, while tech job growth is occurring mostly outside traditional software roles.
Speedrun opened its previously invite-only Insiders program to applications from operators and executives, offering a look at startups it backs before the wider public sees them. Andrew Chen describes the offering as curated startup access through formats including introductions, dinners, and conferences.
A post points to a customized user-story template with an example for business analysts, product managers, and requirements engineers; the template is linked in a Google Doc, but its contents are not reproduced in the post.
Assess a startup CTO co-founder as a product and business partner, not just a coder: advice in the thread describes the role as combining technical leadership with business/startup competence and product strategy; the founder says candidate discussions covered architecture, taking the MVP to production, customer needs, and commercial engagement. For engineering quality, a commenter recommends involving a trusted, highly experienced technical reviewer; the nontechnical founder can separately assess communication, ambiguity handling, and whether they work well together.
- For a small product built alongside frontier-model platforms, commenters advise focusing on one recurring job for one buyer rather than staying broad. They warn that features such as shared memory, roles, model choice, self-hosting, or per-business pricing may be easy for larger platforms to copy; a more durable wedge is a workflow that reliably solves a costly handoff and earns repeat use before expansion.
- One suggested wedge is a weekly chore that an owner-run shop hates, designed to work within the first 10 minutes with no setup; if customers can self-host, setup and support may be the paid value rather than the code. A commenter cautions that trust-focused buyers may be slower and more expensive to sell to early.
- A LegalTech founder said that shifting from pitching to listening to customers helped build trusted, deeply embedded law-firm partnerships; customers felt the product was bespoke and that they had helped shape it.
- The founder reported that a U.S. law firm supplied five tasks on which the product outperformed Harvey, but a commenter questioned which tasks were tested and who judged them, calling it a single demo where the evaluator may have wanted the product to win; treat the claim as anecdotal, not a validated benchmark.
- A commenter advised clarifying the product proposition and growth flywheel before raising, and resisting pressure to hire quickly just to spend the funds.
An applicant considering Eternal’s on-campus associate program manager role asked whether they could move into product management at the same firm within 1–2 years . Replies differed: one called program and product roles “very different” , while another suggested using the role as a stepping stone . A follow-up advised checking for product-team exposure and how realistic internal transitions are before deciding .
My biggest takeaways from @mona_sarantakos (opens in new tab), Head of Product for @Muse (opens in new tab), the most successful AI consumer product of the year:
The first Muse prototype broke an hour before the board meeting pitch. Lead designer @AlexCornell (opens in new tab) built the working prototype in a few days, and the team decided at the last minute to show it at a board meeting. An hour before, it broke. Alex and an engineer raced to fix it, and it came back five minutes before the meeting started. “In some alternate universe we didn’t figure this out and maybe Muse was never born.”
Jolly, the now-iconic Muse avatar, was just a placeholder. Alex created the character for the first prototype and called it Jolly Bot, after his nickname for his daughter. Every few months, someone said they could never launch with a stuffed animal as the logo. “Jolly’s like our cockroach. It just never dies.” A week before launch, the team and Mark Zuckerberg made the call: “Jolly stays.” Mona’s reasoning: “When you’re talking to something that feels so personal, the last thing you want to talk to is a corporate logo.”
Muse began with a ping from @natfriedman (opens in new tab). On a Thursday in late January, Nat messaged a few people internally: OpenClaw was changing his whole view of what’s possible, and they needed to try it. Mona spent the weekend setting it up on a Raspberry Pi (Mac Minis were sold out) and gave it access to her email, calendar, and home automation. On Monday morning it sent her a summary of what was coming up that week, pulled from the school emails she never keeps up with. She pinged Nat back: “This is it.”
Muse took seven months to launch, even though work started before every other AI assistant. After figuring out what the product was going to be, and building a fully functioning end-to-end version in a few months, the team spent the rest of the time on reliability, safety, and security. Their internal motto was “Trust is a must.”
The original Muse team was just 10 people. The core team stayed small on purpose, to keep communication and decisions tight. They worked out of a library-like common area in a Menlo Park building and ate all three meals together each day. “It was like lightning in a bottle. I couldn’t have imagined this product being built with any other team.”
The team didn’t expect shopping to be one of Muse’s top use cases. Mona predicted that personal productivity, health, and personal finance would be the primary use cases. Others bet on home automation, which was big with OpenClaw users. Then internal testers started shopping. Engineering manager @christinesed (opens in new tab) became the top dogfooder, sending shopping bug reports every morning. “All of our evals have been somewhere else,” so the team started building shopping evals.
The team uses “agent swarms” to surface user insights at scale. Muse conversations are long and multi-turn, and no existing tool could analyze them. So the team sends swarms of custom-built agents through anonymized conversations and feedback. The swarms report back what’s working and what’s breaking, down to the 10 people who tried to shop on one site and got blocked. Mona said, “It makes me wonder how we even built products before.”
Muse’s system prompt has to be written by a person. “You cannot use an AI to write a system prompt, any part of it.” The team uses agents everywhere else, including to help write evals. But a human always reviews the data, because “when we don’t, we always have issues.”
Product taste is now the first thing Mona looks for in a PM. She agrees that the PM role has changed more in the past two years than in the two decades before. She can now come up with an idea, build it herself, and have a real conversation about it “in a 24-hour cycle.” What sets a PM apart is representing the customer, holding a high bar, and telling an engineer who loves their own idea: “That’s really cool. I’m glad you really like it. I don’t think anyone else will.”
The next frontier for Muse is ambient presence. Near term, Mona wants to cut the babysitting (“At the point where I’m babysitting an agent, I might as well do it myself”) and move Muse beyond the app into glasses and other devices. But she doesn’t claim to know where this goes: “A year ago, I would’ve never explained Muse to you the way that I did.”
Watch the conversation: https://www.youtube.com/watch?v=WY0cdD4gNoQ (opens in new tab)
In late January, @natfriedman (opens in new tab) pinged a few people at @Meta (opens in new tab). He’d been using OpenClaw, and told them they needed to try it.
Mona Sarantakos couldn’t find a Mac Mini, so she spent the weekend setting it up on a Raspberry Pi and gave it access to her email and calendar. Her husband told her, “That is so not safe.”
But by Monday morning, it had pulled together everything her kids had going on at school that week. She pinged Nat back: “This is it. I’m in.”
The team spent the next seven months building something safe and simple enough for everyone. They called it @Muse (opens in new tab). It became the fastest-growing AI consumer product ever.
Mona is head of product for Muse and VP of product at Meta Superintelligence Labs. In her first podcast interview, she shares the inside story of how they built and launched Muse.
We discuss: 🔸 Why they waited seven months to launch, despite getting started before all of their competitors 🔸 The two levers that actually move activation for an AI agent 🔸 How @natfriedman (opens in new tab), @alexandr_wang (opens in new tab), and @finkd (opens in new tab) each left their imprint on the product 🔸 Why the Muse team relies on “agent swarms” more than dashboards 🔸 Why taste is now the most important PM skill
Listen now 👇 https://www.youtube.com/watch?v=WY0cdD4gNoQ (opens in new tab)
- Muse waited seven months to launch: after building a working end-to-end product in a few months, the team spent the remaining time on reliability, safety, and security, guided by “Trust is a must.”
- Dogfooding overturned the team’s expectations about key use cases: internal testers made shopping a prominent use case, and their bug reports prompted the team to build shopping-specific evaluations.
- To analyze long, multi-turn conversations, Muse used custom agent swarms on anonymized conversations and feedback to find what worked and where users got blocked; people still reviewed the results, and the system prompt had to be written by a person.
- Muse’s original team had 10 people, kept deliberately small to make communication and decisions tighter.
- Head of Product Mona Sarantakos says product taste is now the first quality she looks for in a PM: represent customers, hold a high bar, and be willing to challenge an engineer’s favored idea. She says PMs can now build an idea themselves and discuss it within a 24-hour cycle.