Stanford CS153 Frontier Systems | Nikhyl Singhal from Skip on Product Management in the AI Era

Stanford CS153 Frontier Systems | Nikhyl Singhal on Product Management in the AI Era

In a CS153 guest lecture, Professor Mike Abbott shifts from technical topics to product, tracing how software moved from PRD-driven project management to founder-led consumer product building, and arguing AI is blurring the boundaries between design, engineering, and product.

Nikhyl Singhal shares his background founding companies and leading product at Google, Meta, and Credit Karma, then explains four company phases — finding product-market fit, post-fit process and coordination, hypergrowth scaling and expansion, and big-tech-late-stage innovator’s dilemma — and how each requires a different kind of product manager. He argues that in the AI era, the bureaucratic movement-of-information role is dying while a new product builder role is emerging where designers, engineers, and PMs converge.

(Channel description, Stanford Online.)

A 63:14 guest lecture in Stanford CS153 Frontier Systems, taught by Mike Abbott (Stanford CS professor, Sequoia partner, ex-Twitter engineering VP). Guest: Nikhyl Singhal — founder of Skip (talent-agency-for-product-people); ex-VP of Product at Google / Meta / Credit Karma; ~thousand career conversations with executives over the past 15 years. Published 7 May 2026 to the Stanford Online channel. ASR caption track; no chapters.

The wiki’s second Stanford-CS-course-lecture source (after Leskovec CS224W 2023) and third Stanford Online ingest.

TL;DR

Five substantive contributions:

1. The four-company-phases framework — different PM role per phase: a structural taxonomy of product-management’s evolution across a startup’s lifecycle.

   Phase	Stage signal	PM role
   1. Pre-PMF (product-market-fit)	“Rubbing two sticks together hoping to get smoke” — rapid experimentation	No PM. Founder drives all decisions; PM doesn’t make sense here
   2. Post-PMF / process	”Sucking sound — natural pull. The exact thing that got you here (experimentation) now has to stop”	PM emerges as quieter, process-oriented function — multiple teams that don’t talk to each other; PM is the glue
   3. Hypergrowth scaling + expanding	”App store + Facebook ads + internet distribution allows companies to grow very rapidly” (LinkedIn: 10-15 years to 1B users; Uber: 18 months)	Chief Product Officer + large PM teams; scale existing product + expand into adjacencies (Google: new products beyond search/ads; Meta: feed → short-form video → Reels; Credit Karma: credit-score → money button)
   4. Big-tech late-stage / innovator’s dilemma	”Combat innovators dilemma. Need to go create something new. Need to go do zero-to-one when there’s so many reasons not to”	Zero-to-one product builders within large orgs — the hardest PM role

   The wiki’s first operator-canonical four-phase company-lifecycle framework with explicit PM-role differences per phase.

2. The AI-era PM role inversion — product builder, not movement of information: Singhal’s central claim is that the bureaucratic PM role is dying while a new product builder role is emerging where designers, engineers, and PMs converge. “All the parts of my job that I dislike and hate, I can essentially obsolete myself, engineer myself out of. And then the parts that I love — judgment, decisioning, being courageous, testing things in the wild, talking to customers, working with engineers on a really hard problem, partnering with another company on expanding the pie — those are the parts of my job that exist. That is amazing.”

3. Concrete operator-class employment numbers:

   • Big tech doing 30-70% layoffs this calendar year (named: Salesforce, Block, Snap-10%-yesterday).
   • Top-1% PM salaries more than doubled in 18 months.
   • Four product-leader contracts Singhal personally helped negotiate crossed eight figures annual compensation.
   • More open PM roles right now than ever in history — “completely unobvious to folks.”
   • The at-risk cohort: middle-managers aged ~mid-30s, 8-15 years tenure, promoted during 0% interest rates as movement-of-information role, with kids and aging parents — “don’t have time to spend reinventing themselves.”

4. The “brand at all-time low” hiring shift: Singhal claims top employers (named: Anthropic, OpenAI) don’t care what brand of company you’ve been at. They want to know “how modern you are in how you think and how you use your tools.” “They can tell when people are learning the tools at the same time that they’re interviewing. They can tell when people are like ‘look, I’m here because I want to manage.’ It’s like, ‘well, there’s no more management. It’s all about building. It’s all about being hands-on.‘”

5. The career-as-chapters framing — origin of “Skip”: “If you’re going to work for 50 years and you’re going to on average your jobs are going to be between two and three years, math tells me that you’re going to have between 15 and 18 jobs. 15 and 18 jobs means that your career is not like periods in a hockey game. It’s like chapters in a book.” The naming of his company Skip captures the operator-advice: “Think about the chapter after the one you’re thinking about now. Less about ‘what’s my first job out of college’ — what’s your second job out of college and how does your first job set you up to get the maximum amount of opportunity in your second job.”

What was actually ingested

Full 63:14 transcript including Q&A. Abbott opens with a fun-history-of-product-management framing (20 years ago: PRD-driven IBM/Microsoft model; consumer companies typically have no PM — founder drives; “Apple does not have any product managers”). Singhal then traces career background and the four-phases framework. Q&A covers Hangouts as anti-pattern, forward-deployed engineers vs PMs, Skip the company, AI-era trends, layoffs, college-advice-for-CS-students, Meta-and-the-metaverse, hierarchy-flattening, career advice from his own 30-year retrospective.

ASR caption track only; minor name-rendering errors flagged below.

Mike Abbott’s opening — the PRD-driven-to-founder-led arc

Abbott opens by tracing software’s evolution:

• 20 years ago: project-manager builds PRD (product requirements document); hands to engineers; classic IBM / Microsoft playbook.
• Consumer-side anomaly: founders drive product decisions (Twitter, Apple). “Apple does not have any product managers. They just basically build all their products between having a designer and having an engineer.”
• Today: “a designer can now go vibe code a lot, and really these roles of design / engineering / product are merging.” Direct PM-leadership confirmation of the vibe-coding floor-raising thesis from a Stanford-CS-faculty-with-Sequoia-affiliation vantage.

The four-phase framework — detailed

Phase 1 — Pre-PMF

“Product market fit is the equivalent of like you know rubbing two sticks together hoping that you get some smoke. The goal of the organization is to have as many shots on goal as possible. Product management doesn’t exist at this phase. You need a founder. There is no product manager that makes sense when the company is essentially throwing whatever they’re doing away trying to find some resonance.”

Implication: early-stage-company-with-PM = anti-pattern.

Phase 2 — Post-PMF / process

“Some very small percentage of those companies, let’s say 1-2-3-4% of those companies will actually find what we call product market fit. Product market fit means that you’ve got a sucking sound — you’ve built something, all of a sudden there’s a natural pull. The exact thing that got you to where you are, which is experimentation, rubbing two sticks together, finally smoke is coming, you start to feel heat — you need to stop experimenting and throwing things away. What you actually need to do is take a moment and build some resilience, some consistency because the next customer that you bring in can’t have a completely different product. Now product management comes in and is a much quieter function — more process.”

The role-switch is the critical handoff: founder-skills (experimentation) and PM-skills (predictability) are opposite. The transition is “the company growing up.”

Phase 3 — Hypergrowth (scaling + expanding)

“Of those companies, very very very small percentage — again maybe one or 2% — go through hypergrowth. Hypergrowth didn’t really exist when I was in school. When I was in school it took 10-15 years for LinkedIn to get its first billion users. It took 18 months for Uber. And now it even happens even faster. The reason is the app store, Facebook ads, the distribution of the internet allows companies to grow very rapidly.”

Singhal’s worked examples:

• Google: search + ads → trying new products beyond.
• Meta: feed → short-form video → Reels.
• Credit Karma: credit-score company → “the money button on the phone.”

This phase needs large PM teams + Chief Product Officer. Both scaling existing and expanding into adjacent products simultaneously.

Phase 4 — Big-tech late-stage / innovator’s dilemma

“You have to sort of start over again because now you have so much success — whether it was Twitter, whether it’s Facebook or whatever these companies are — your job is to combat innovator’s dilemma. You need to go create something new. You need to go do zero-to-one when there’s so many reasons not to do it, when small businesses don’t compare to the large huge businesses that exist.”

The hardest PM role — zero-to-one inside a large org with sunk-cost-fallacy headwinds.

The AI-era inversion — product builder, not movement of information

Singhal’s central claim about the post-2025 PM role:

“Number one thing I’m noticing is there is more joy for the leaders and people that are in tech than ever before because they don’t have to depend on an engineer, on a designer, on a founder, on their boss in order to get something done. They can all build. And there’s nothing more empowering than being able to whip out cloud code and build something — or more specifically to obsolete yourself from some status report that you hate having to file.”

The role-collapse mechanism:

“In the past, product was essentially a movement of information — no matter how big or small the company was. Your job was to package information for some other decider. And that is a horrid job if you are a — I mean being a bureaucrat sucks if you’re a builder.”

“The vast majority of product people get into product because they like building stuff and then very quickly they realize like ‘hey once you know how to build stuff you ought to organize others that build stuff’ and that’s when things kind of go dark.”

The implication: bureaucratic-PM ≈ micro-productivity-trap-of-PM-role. AI obsoletes the bureaucracy; the builder identity reasserts.

The two-question audience test

Singhal’s diagnostic:

1. “How many of you are anxious about getting a job after college?” → Entire audience (Stanford CS153 students, ~100 people).
2. “How many of you are having a lot of fun building things using AI?” → Same group.

Compare to two-years-ago: low anxiety (20-30%), low fun. “Three years ago before the mass layoffs, when interest rates were 0%, every person even if they were average-skilled would have six job offers. And and and there was low anxiety. However, everyone disliked their jobs.”

The inversion: anxiety up, fun up, jobs concentrated at the top of the talent distribution.

The employment numbers — concrete

Singhal supplies the wiki’s first operator-class compensation-and-layoff numbers for the AI-era PM market:

Datapoint	Number	Source
Big-tech layoffs this calendar year	30-70%	Named: Salesforce, Block, Snap (10%-yesterday)
Top-1% PM salary movement (last 18 months)	More than doubled	Singhal’s network
Product-leader contracts crossing eight figures annual	4 (personally helped negotiate)	Singhal’s network
Open PM roles in industry	More than ever in history	Industry data
At-risk cohort	Mid-30s middle-managers, 8-15 years tenure	Singhal’s diagnosis

The wiki’s first first-party-operator data anchor on the bifurcation: top-1% salaries doubling + 30-70% layoffs in the same year. The bimodal-compensation distribution is direct evidence for AI’s employment effects at the senior-knowledge-worker class level.

The at-risk cohort framing:

“Middle managers — these are the people that, you know, were moving information. They then see the cloud codes of the world. They have kids. They have aging parents. They are in their mid-30s. They’re sitting there like ‘I don’t have time to spend reinventing myself.’ Those are the people that are going to get laid off.”

Convergent with Thompson 2026’s Pia Torian and senior-vs-junior code-sense-worry, but at the PM-leadership scope rather than the developer scope.

The “brand at all-time low” hiring shift

“Top quality employers like the Anthropics, the OpenAIs and others, when they interview, they don’t actually care what brand of company you’ve been at. They want to know how modern you are in how you think and how you use your tools. They can tell when people are learning the tools at the same time that they’re interviewing.”

The shift: brand → modernity-of-toolchain as the dominant hiring signal. “A person that’s worked at Google for six years might be dramatically less relevant than the people that are in this room because when you’re at Google you’re in back-to-back meetings all day long and you’ve basically worked on one stack which is the Google stack and it works a certain way. While everyone in here has probably got a max cloud code account.”

The wiki’s first first-party-CHRO-level-network claim on the brand-discount-vs-modernity-premium hiring signal at top AI labs.

The career-as-chapters framing — Skip’s naming

Singhal’s framework:

Variable	Value
Working career length	40-50 years (theoretically growing — “unless I’m mistaken, no one in this room is going to physically labor in their jobs”)
Average tenure per job in tech	2-3 years
Therefore: total jobs per career	15-18
Implication	”Your career is not like periods in a hockey game. It’s like chapters in a book.”

The naming of Skip: “The best career advice is to think about the chapter after the one you’re thinking about now. So it’s less about ‘if all of you are thinking about what’s my first job going to be out of college’ — my ask would be ‘what’s your second job going to be out of college and how do you make sure your first job sets you up to get the maximum amount of opportunity in your second job.‘”

Skip’s properties:

• Founded by Singhal after leaving Meta.
• Talent-agency-for-product-people positioning.
• 125 heads of product from “Anthropic and OpenAI all the way to Meta and others”.
• Highly curated; not designed to scale; not designed to monetize.
• Infinite waitlist.
• Singhal funds it himself.
• Companion: skip.show Substack + content publication.
• 30-year time-horizon plan; Singhal-aged-52 plans to work until 82.

Hangouts as anti-pattern — Singhal’s worked example

A long detour into why Google Hangouts failed:

• Wrong problem: Hangouts tried to consolidate 7 communication codebases (text + voice + video + group + messaging) into one app. “Inside the building, we had seven different code bases — it would make sense if this was all consolidated. Well, it turns out customers didn’t just didn’t care.”
• Lesson 1: “Inside-the-building drama and trauma and challenge doesn’t translate to outside-the-building usage.”
• Lesson 2: “Need to stick with it. Large companies have trouble sticking with things that don’t look like they’re winning from day one.” WhatsApp won by focusing on text-only-in-India for years before adding voice/video.
• Lesson 3 (Iteration speed as moat): “It doesn’t matter how you start, it’s how fast you improve it.” Chrome beat Firefox by shipping every 6 weeks (vs Firefox quarterly, vs IE annually). Android beat iOS by shipping every quarter (vs iOS annually). “They built an organization based on iteration speed.”

The iteration-speed-as-moat point is convergent with Hu 2026’s thousand-times-faster-than-incumbents startup-edge claim.

Other substantive points

Forward-deployed engineers ≈ AI-era PM

Singhal on forward-deployed engineers (Palantir’s coinage): “A forward-deploy engineer is extremely helpful to pull out [customer] insight, but the reality is now we are able to even move a step beyond that because there’s so many signals that contribute to what should we build and why should we build it.”

AI-era equivalent: agents that summarize every customer service chat / every sales call / every survey response into a prioritized order with money-generation and implementation-complexity weighting. “In the past, like if I was to come a year ago and have this conversation, that would sound like science fiction.”

Meta-and-the-metaverse — Mark’s belief-driven investment

Singhal’s diagnosis of the Metaverse failure: “Mark felt that in order for the company to get to the next level, he wanted to be the innovator of [the next] platform. Meta as a company was not an innovator on mobile. It leveraged mobile. It leveraged the web. … Now we’re 5 years in and we’re not seeing that.” Mark’s strategy: “I’ll spend 10 years trying to get there because I believe I can drive that next platform.” Apple-like culture; not Google-like (consensus-driven).

Connects to a broader observation: “To do big innovation you can’t do it by consensus.”

The Stanford-CS-faculty-quality knock

Singhal’s most pointed observation: “The teaching was not particularly great. The assignments were incredibly difficult. Often built like for the class at the seat of the time. And yet the students that were in the course were so good — you had to learn how to take an unstructured problem and work with your peers to understand even what the hell was being asked, let alone how to solve it and use the tools. That skill turned out to be the most important skill … My computer science teachers were world class — I think worse than Foothill is my suspicion.”

The structural takeaway: unstructured-problem-solving with peers is the durable skill; structured teaching is not. “You’re moving into a world where managers are going to be a dirty word. Nobody’s going to want to be a manager anymore. And so now you’re not going to get any feedback. You’re not going to get any structure. You’re going to be given an impossible problem.”

Career advice — three durable skills for CS students

1. Be current and modern in toolchain. “You have to be using the tools. You have to have an opinion. You have to be pushing hard.”
2. Network. “Those passive relationships can be very very valuable. Both directions — you can help each other, work for each other, advise each other.” Singhal: “I stay in touch with 25 people that are undergrad classmates.”
3. Systems-programming mindset. “You have to sort of understand what is the system you’re building and how does it evolve. Every few years the platforms evolve — assembly language became compiled programming languages became scripting languages now is prompted languages. Every few years you see the stacks get smarter.” The skill is engineering abstraction — “is this the right thing to build? Is it working? Does it fit into our system design?”

Convergence and contradictions

Source	Connection
Thompson 2026	Singhal’s mid-30s-middle-managers-at-risk is the PM-leadership-cohort analog of Thompson’s junior-vs-senior code-sense worry. Both name the at-risk class with tenure-and-life-stage specificity
Hu 2026	Singhal’s product-builder convergence of design/engineering/PM is the role-side counterpart to Hu’s three-archetypes (IC / DRRI / AI founder type). Both predict role-collapse; Singhal frames it from the PM-leader vantage
Tan 2026 (YC President)	Singhal’s “a designer can now go vibe code” is the Stanford-faculty articulation of Tan’s AI-founder-type-running-10-15-parallel-sessions worked example. Both observe the same role-collapse from different organizational vantages
Jassy 2025 (Amazon CEO)	Jassy’s +15% IC-to-manager-ratio target is the operator-implementing-side counterpart to Singhal’s middle-managers-are-the-at-risk-class diagnosis. Direct convergence on the flatten-management consequence of AI
Karpathy 2026	Singhal’s “a designer can now go vibe code” is the wiki’s first PM-leader-vantage citation of the vibe-coding term. Confirms the term has spread from coinage (Karpathy April 2026) → developer-side observed (Thompson April 2026) → product-leadership-circles (Singhal May 2026) → Stanford-CS-course-content (this source) in <13 months
Fung 2026 (Anthropic)	Fung shows what an AI-native engineering org looks like operationally; Singhal shows the AI-native PM/leader career-trajectory implications at the labor-market layer. Two-layer view of the same shift
Sternfels 2026 (McKinsey)	Singhal’s judgment / decisioning / courage / customer-conversation / system-design-mindset durable-skills cluster is convergent with Sternfels’ aspiration-setting / judgment / discontinuous-leap-thinking / human-to-human-skill. Different sectors (PM-leadership vs management-consulting), overlapping but non-identical durable-skills clusters
Globerson et al. 2026	Singhal’s named durable-skills are the PM-vantage version of Globerson’s collaboration / creativity / critical thinking psychometric framework
Karpathy 2026	Singhal’s “every few years the platforms evolve — assembly language → compiled → scripting → prompted languages” is direct citation of Karpathy’s Software 1.0 / 2.0 / 3.0 progression by analogy, from a PM-faculty vantage
Leskovec 2023 (Stanford CS224W)	Second Stanford-CS-lecture in the wiki. Different courses (CS153 Frontier Systems vs CS224W Machine Learning with Graphs), same Stanford Online channel, but distinct vantages (PM/product-leadership vs ML-foundations academic)

Contradictions

None substantive. Singhal’s framing is operator-prescriptive and largely convergent with the wiki’s existing AI-employment / vibe-coding / durable-skills clusters from a new vantage.

Linked entities and concepts

Existing wiki entities reinforced:

• Stanford Online — third source on this channel (after Leskovec 2023 and CS224W); both academic-foundation and frontier-systems courses now ingested.
• Anthropic / OpenAI / Google — named as hiring exemplars + Skip community participants.
• Andrej Karpathy — implicit via vibe-coding term usage and prompted-languages citation by analogy.

Concept pages affected:

• vibe-coding — PM-leader-vantage usage of the term; Stanford-CS-faculty-level mainstream-adoption signal.
• ai-employment-effects — concrete operator-class numbers (30-70% layoffs / top-1% salaries doubled / 4 contracts >$10M annual).
• ai-deskilling — mid-30s middle-managers as named at-risk cohort.
• durable-skills — PM-vantage durable-skills cluster.
• micro-productivity-trap — bureaucratic-PM-role as instantiation.
• agentic-engineering — product builder convergence at the role-design level.

Dangling (single-source first-mention, deferred):

• Nikhyl Singhal — Skip founder; ex-Google / ex-Meta / ex-Credit Karma VP-of-Product.
• Mike Abbott — Stanford CS153 professor; Sequoia partner; ex-Twitter engineering VP.
• Skip — Singhal’s talent-agency-for-product-people company; 125-head-of-product community; 30-year plan.
• Skip.show — Substack / content publication.
• CS153 Frontier Systems — Stanford course.
• Credit Karma — Singhal’s pre-Skip employer; mentioned as a hypergrowth-phase exemplar.
• Block / Salesforce / Snap — named in the 30-70% layoffs context.
• Horizon Worlds / Mark Zuckerberg — referenced in the Meta-metaverse anti-pattern discussion.

Concept candidates surfaced:

• Four company phases framework — Singhal’s structural taxonomy. Strong promotion candidate on second-source mention.
• Product builder (vs PM/designer/engineer-as-silos) — Singhal’s role-collapse framing. Promotion candidate.
• Career-as-chapters / 15-18-jobs-in-50-years — Singhal’s framing. Single-source.
• Brand-at-all-time-low / modernity-of-toolchain-as-hiring-signal — Singhal’s hiring-criteria-shift claim. Promotion candidate on second-source mention.
• Iteration-speed-as-moat — convergent with Hu 2026’s startup-edge claim. Two-source threshold met (Hu + Singhal); promotion candidate.

Open questions raised by this source

• The Chrome-vs-Firefox iteration-cadence-as-moat claim — Singhal cites it from memory; primary-source target on the actual cadences (Chrome was every 6 weeks, Firefox quarterly, IE annual — historically true but worth confirming).
• Skip community membership data — “125 heads of product from Anthropic and OpenAI all the way to Meta and others”. First-party data on the composition would substantiate the brand-at-all-time-low claim.
• The four-product-leader-contracts-crossed-eight-figures — anonymised but specific number Singhal personally helped negotiate. First-party verification would substantiate.
• The 30-70% big-tech-layoffs-this-calendar-year — Singhal cites Salesforce / Block / Snap. Primary-source data (e.g. layoffs.fyi) would substantiate the range and identify additional companies.
• The brand-discount-vs-modernity-premium hiring criteria at top AI labs — would benefit from a controlled comparison: e.g. an Anthropic / OpenAI hiring-data disclosure on time-to-hire / acceptance-rate by prior-employer-brand. Open primary-source target.
• Skip.show Substack — primary-source ingest target for Singhal’s published PM-leadership essays.