Lessons from Investing in 700 Companies — Gokul Rajaram on Invest Like the Best
In an era when software can build everything cheaply, judgment — editorial capability about what to build and evaluate — is the one truly future-proof skill. Durability comes from network effects, control points, hardware, scarce assets, or being an essential workflow.
Why this is in the corpus
Rare operator + investor combined interview. Gokul ran core product at Google Ads, Facebook Ads, Square, DoorDash — and has invested in 700+ companies. Unusual framework density (custom audiences origin, three-ways-to-do-ads, two-kinds-of-legacy-SW, Midas-style durability stack, weekly CEO email, self-serve forcing function, idea maze, board buddy).
Summary for skimmers
Gokul names the product manager as "keeper of the why", argues judgment is AI-proof, distinguishes utility-priced legacy SW (Zendesk, Slack — at risk) from data-priced SW (NetSuite, Salesforce — insulated), names 5 durability sources (scarce asset / control point / hardware / essential workflow / network effects), the only 3 ways to build an ads business, North Star metric with paired check-metrics (engagement budget), self-serve as a forcing function, work-project hiring (Tony Xu gave candidates $10-$20 to acquire 1000 DoorDash customers), and founder authenticity via origin story + idea maze.
Briefing
What survives the editorial filter
This page should feel like a smart colleague already listened for you and left only the operating logic worth keeping. Not everything said in the episode makes it through.
Trust signal
Direct episode extraction
Best used for
Decision-grade retrieval metadata not yet added for this episode.
Hold lightly
No explicit downgrade reason stored yet for this episode.
Principles
Durable claims that survive beyond the speaker's biography — each with explicit limits, transferability judgment, and evidence.
Principle
Hiring: work projects over interviews
Interviews let BS-ers through. Work projects surface agency and craft.
Best PM candidates went and talked to 10 customers on the street, came back and said "none of them want this Premium Insights product — we should not build it."
Testable, specific, with Tony Xu canonical example.
Principle
Stay 3-4 years minimum — job-hopping is a red flag
Impact requires tenure. Three to four years is the minimum threshold.
One short stint among a series of jobs is forgivable; two or three in a row is an immediate red flag for hiring managers.
Counter-cultural in Bay Area talent markets.
Principle
Span of control: 10+ or be an IC
Pure middle management is the role most exposed to AI-era cuts.
Gokul: "On the company side, don't hire managers as long as possible. Hire doers, hire builders."
Controversial but concrete guidance.
Principle
Replace the entire system — do not live on top of an API
System-of-record ambition is mandatory; system-of-action alone is no longer viable.
Slack cut Glean's API access. Salesforce cutting third-party agents. Migration tools (1-2 year engineering effort) are now a mandatory investment.
2024-2025 API-lockdown shift; Slack/Glean is the canonical example.
Principle
Outcome-based selling replaces feature-based selling
Lead with the outcome you already delivered, not features. Pricing should follow outcome, not effort.
Outcome-based pricing is the natural endpoint of AI-era software where features are commoditised by foundation models.
Signals a broader pricing shift — complements utility-vs-data legacy-SW diagnostic.
Principle
Lighthouse effect — win one vertical anchor before going horizontal
The reference customer in the target vertical is the single highest-leverage sales asset.
Default go-to-market: land the best one in a vertical, then convert the rest through imitation.
Named effect with memorable JP Morgan / P&G comparison.
Principle
PMs are keepers of the why
Every feature must have a clear hypothesis grounded in customer behavior change — no hypothesis, no ship.
Bottoms-up building with PMs + engineers + researchers + designers writing code together. PM:engineer ratio shifts from 1:3 or 1:10 to 1:20.
Redefines PM role for the AI era.
Principle
Consumer behavior change is the biggest ad-network threat
The risk to incumbent ad networks is not competition — it is users never opening the app again.
If users connect Uber/DoorDash accounts to ChatGPT and let the agent handle repeat transactions, the ad surface evaporates. Incumbents must experiment and measure behavior of early connectors.
The least-discussed but largest risk to incumbent ad networks.
Principle
Outcomes as customer behavior change
Customer behaviors are the leading indicators of every business outcome.
The only question to ask before a launch: why are you launching this? If there is no hypothesis grounded in customer behavior change, do not ship.
Operationalises "product is outcomes" in a testable way.
Principle
Self-serve is a forcing function for better product
Self-serve forces onboarding quality + reveals power-user patterns that direct sales never surfaces.
Larry rejected Google internal-only tools: "Everything you build for large customers must be available to small customers." Smaller customers adopted advanced knobs faster than enterprise. Figma infiltrated Square bottom-up after design team refused top-down mandate.
Anchored by Larry's internal-tool veto at Google.
Principle
Best PMs and designers are editors, not adders
Judgment = editorial capability. Reducing is the unit of product work.
In an AI era, editorial judgment becomes the scarce skill because output is abundant.
Anchors the judgment-is-future-proof principle in a concrete hiring bar.
Principle
Shift risk from onboarding to transaction-level
Late + targeted risk review beats upfront + broad review. Lazy-brilliant onboarding.
Works because most applicants never reach the threshold where risk matters. Saves massive engineering effort on upfront ops.
Counter-intuitive pattern with two independent Google/Square examples.
Principle
Custom audiences — the ad innovation that came from Zynga whales
The best ad-product ideas come from connecting disparate customer frustrations.
Zuck shadowed the ads team for a year, then generated custom-audiences as an idea from Pincus's whale-targeting complaint.
Specific origin story for a foundational ad-product innovation.
Principle
Judgment is the one truly future-proof skill in the AI era
Humans provide editorial judgment over AI slop — across product, engineering, and design.
Every product director worries about AI slop — thousands of AI engineers generating code without anyone knowing what matters. Humans review the critical code; PMs evaluate outputs; designers own the coherence of the system.
Clean statement of the AI-era defensibility thesis.
Frameworks
Reusable systems and operating models — including when they help and when they break.
Framework
Weekly CEO email — three sections
Repetition of top-of-mind items across weeks is how the message actually lands.
Derived from Jack, Zuck, and Sheryl's weekly emails. Most powerful when candid — candor surfaces ideas from the team.
Simple, reusable template; cited by 15+ CEOs Gokul has advised.
Framework
Five sources of durability in the AI era
Five durability sources — pick at least one or get eaten by horizontal AI platforms.
DoorDash: network effects across restaurants, dashers, consumers. Toast: hardware + payments. Mercury: money flow + regulation. Sierra: Bret Taylor as unique asset.
Cleanest durability checklist in the corpus. Complements Hamilton Helmer 7 Powers.
Framework
Two kinds of legacy software — utility vs data
Pricing model is the diagnostic for AI-era legacy-software survival.
Zendesk: 50 seats can be replaced by 20 seats + 30 AI agents over time. NetSuite: runs your whole business; no one rips it out.
Practical investor diagnostic; public market multiples have not distinguished these two groups.
Framework
Three ways to build an ads business
No other sustainable ad business model exists. Everything else is middlemanning on platforms who will learn and absorb your capability.
You either die or live long enough to become an ads company. ChatGPT is headed there now.
Named framework with canonical examples for each of the three paths.
Framework
North Star metric + check metrics
A single metric is manipulable; pair it with guardrails to prevent Goodhart.
At Facebook, the News Feed team and ads team shared an engagement budget — revenue could rise but only if engagement did not fall more than X percent.
Standard NSM guidance + the under-discussed guardrail practice.
Framework
Founder evaluation — origin story + idea maze
Founder-problem fit emerges from lived experience + historical study. Authenticity + idea-maze depth separate the founders who will last from those who are chasing.
Max Rhodes / Faire: origin in an undergrad umbrella-company distribution struggle. Dylan / Figma: seeped in design.
Reusable interview framework for VC/angel evaluation.
Signals
What appears to be shifting, for whom it matters, and what happens if you ignore it.
Signal
PM:engineer ratio shifting from 1:3-1:10 to 1:20 in AI-native orgs
The traditional product-team composition is being rewritten; headcount plans should account for the 1:20 ratio as the new benchmark.
Designers now manage design systems (small central team); AI does the production design work. PMs shift from prescribers to evaluators of non-deterministic AI output.
Concrete numerical signal on org-shape change.
Opportunities
Only included where there is a buyer, a real wedge, and a plausible revenue path — not vague idea theater.
Opportunity
Migration-tool building as the mandatory AI-native wedge
Migration engineering is the mandatory wedge — the AI-native who does not invest here cannot convert any serious enterprise customer.
Transition tools are how you get past "this data is still there" — without them, the spanking-new system is un-installable.
Concrete startup-investment thesis on the overlooked technical wedge.
Lessons still worth keeping
Useful takeaways that did not fully clear the bar for durable principle status.
Lesson
Larry killed the internal-only customer tools — self-serve became the advantage
Internal-only tools are a tell that self-serve product is under-invested; releasing them unlocks power users.
Self-serve customers exploited the ICS knobs in ways enterprise did not — the product learned faster via the smallest customers.
Specific Larry-Page decision with long tail.
Lesson
Zuck's custom-audiences idea came from Mark Pincus's Zynga-whales frustration
Great product ideas often come from CEOs connecting disparate customer frustrations, not from the owning team.
Custom audiences became the foundation of most ad systems. The ads team did not generate the idea — Zuck did, from Pincus's complaint.
Lineage-dense origin story with long operational tail.
Tensions surfaced
Contradictions and trade-offs the episode raises — judgment calls a thoughtful operator has to navigate.
Tension
Engagement budget vs ad monetisation
Engagement and monetisation are not both-can-go-up variables — the trade must be explicit and budgeted.
Facebook News Feed team + ads team annually agreed an engagement budget. Without that governance, ad teams push until engagement craters.
Named operational discipline; applicable to any ad-exposed consumer product including ChatGPT.
Tension
Utility-priced legacy SW vs data-priced legacy SW
Pricing model is the single most important diagnostic for legacy-SW survival in the AI era.
A Zendesk customer can pay for 20 seats + 30 AI agents instead of 50 Zendesk seats — a two-way-door decision. A NetSuite customer cannot rip out their ERP without career risk.
Actionable investor diagnostic.
Tension
Building on incumbent APIs vs building the full system of record
The API-as-stable-foundation era is over — agent companies must replace the system of record.
Slack cut Glean's API access. Multi-vertical incumbents followed. Migration-tool engineering is now a 1-2 year mandatory investment, not optional.
Names the 2024-25 structural shift explicitly.
Corpus connection
Where this episode fits for retrieval
What kinds of decisions this briefing is best pulled into.
Primary decisions
- • product
- • strategy