We can't find the internet
Attempting to reconnect
Something went wrong!
Hang in there while we get back on track
Snyder's method for finding demand and keeping customers
On Lenny's Podcast, Rob Snyder said demand comes from a specific person with PULL:
- Project: something on their to-do list, like "I want to hire more employees," not "buy recruitment software."
- Unavoidable: it has to happen now.
- List: the options they've considered.
- Limitations: the reasons those options aren't good enough .
All four have to be present. A product team with an urgent, required feature that it can build itself "won't buy anything unless their existing options aren't good enough" .
Advice PMs can act on:
- Talk to customers before you build. Snyder calls building first the most "brutal" trap. AI shortens it to weeks, but you should still "face the cold wind of reality as soon as possible" .
- Skip design partners. In his experience they buy into your vision, give feedback "playing Steve Jobs," and never buy. People with a real job to get done just want it done . Offering a partnership puts off buyers who really need the product and attracts people who don't .
- Throw the Hail Mary. If a call is lukewarm, ask what they'd pay to have taken off their plate right now. One startup selling applicant tracking to governments got its first contract from a job-description problem it heard about this way .
- Design for retention. Snyder assumes every customer will churn until they reach a specific outcome, "X." Find X and build onboarding so every customer gets there. For his restaurant retention product, X was preventing an employee from quitting within the first two weeks . If churn is near-total, find out who stays. One horizontal AI-workflows startup found freight forwarders sticking with a single workflow and refocused the whole company on them as a premium product .
- Run a "case study factory." Repeat one customer win across pipeline, sales and delivery, then fix the current bottleneck first .
Will AI push PMs toward strategy or toward constant execution?
A former PM now in grad school says the job has become "all about speed and no one cares about quality." They see two possible futures: PM merging with brand and marketing, or moving into consulting and forward-deployed roles . Replies split three ways:
- Judgment matters more. One PM's biggest worry is companies "confusing the ability to build faster with knowing what to build." They expect PM to move toward strategy, positioning and trade-offs .
- Back to normal. Another hears less 2025-style "product is going away" talk. Building faster, they say, "hasn't necessarily translated into happier customers or more money" .
- Product under strategy. A PM in logistics predicts product orgs will report to CSOs rather than CTOs, with PMs prototyping and testing with users before engineering starts .
The pushback comes from the execution side. A B2B SaaS PM describes doing discovery, delivery and releases "at basically the same time" under constant pressure to ship faster. They call it temporary, driven by leaders' "ridiculous promises" about AI .
Aakash Gupta makes a related point about strategy. Decentralized decisions (even CSMs are "shipping to production") and PMs' career incentives are spreading AI across products. A strategy has to help teams say no, and AI only belongs where it advances the core value proposition .
Pay and backgrounds of PMs at AI model labs
Gupta reports that open model-PM roles posting a US salary range have a $345K median base, versus $216K for PM roles outside AI. He reviewed 316 product people at 22 model-training companies:
- 85% were PMs before joining the lab.
- 64% have a technical degree.
- 28% founded a company first.
- Median career length is 13 years .
Only 6% started in data or research . His advice: take the AI feature on your team, learn to build evals, and write the behavior spec yourself .
Turning tacit knowledge into specs
A non-technical founder with 230+ paying users built a financial-services product largely with AI. He is weighing a developer hire with a $35K recruiter fee, and says explaining the domain often feels slower than building it himself . The top advice: spend a week writing the financial rules as test cases (input, expected output, and why). That becomes "the spec, the regression suite and the onboarding doc at once" . Another builder says their real bottlenecks after AI are decisions and testing, not generating specs .
Also notable
- Hiten Shah: you learn a lot by watching someone use a product without the builder explaining anything .
- TBM 447: John Cutler argues that "our work and our workplace are not the same thing." Protecting the work can mean accepting a workplace's limits, choosing your battles, or moving on .
