Method · MoneyRob Snyder recommends coming back after the first sales call with a concept and a plan: mirror the prospect's described problem in a small restatement confirmed with 'Is that roughly it?', offer to return in a couple of days with a mock-up of the solution, discuss fit, and then ask 'Based on that, is there anything else you need to know to decide whether this is a fit you want to bring into your organization?'
guest's recommended script
Method, not a verified protocol.
Lenny's Podcast · How to find product-market fit (but, actually) | Rob Snyder (Author, The Power of Pull)
Method · MoneyRob Snyder recommends starting outreach on LinkedIn because the kind of people you want to talk to have interesting LinkedIn profiles, which does a lot of lifting. Add 100-150 people per week (about 20 a day), and message new connections when they accept the connection request.
None stated.
Method, not a verified protocol.
Lenny's Podcast · How to find product-market fit (but, actually) | Rob Snyder (Author, The Power of Pull)
Recommendation · MoneyRob Snyder discourages recruiting design partners: they 'buy into your vision,' give a bunch of feedback, and never buy from you. Their feedback is just them 'playing Steve Jobs' — helping the startup rather than trying to get something done. People who actually have something important to get done don't want to be a design partner; they just want that thing done.
Stated as his observed experience ('I have yet to see it be helpful').
Lenny's Podcast · How to find product-market fit (but, actually) | Rob Snyder (Author, The Power of Pull)
Method · MoneyRob Snyder's typical recommendation for project management in larger organizations is to have a conversation with the buyer/champion asking when the project on their to-do list needs to be showing results by. The buyer's deadline (e.g., 'definitely Q1 next year') is a forcing function: work backwards from it to determine the steps needed to be deployed and showing results by then, and it becomes a forcing function for all the proper steps in the sales process.
guest's recommendation
Method, not a verified protocol.
Lenny's Podcast · How to find product-market fit (but, actually) | Rob Snyder (Author, The Power of Pull)
Method · MoneyWith TurboPuffer, you show up with your permissions logic and conceptual model and do a simple translation of those permissions into TurboPuffer filter language. Permissions can be expressed directly, e.g., Viewers contains one, Groups contains any BC — a code sample that can be directly mapped to an organization's own permission structure.
Method, not a verified protocol.
AI Engineer · Can Your AI Analyst Keep a Secret? - Aaron Linsky, Bridgewater & Nikhil Benesch, turbopuffer
Recommendation · MoneyLevie advises not expecting AI to disrupt go-to-market: consumer products need virality, enterprise needs a go-to-market machine, and technically-savvy enterprise audiences get PLG plus a GTM engine. Don't mistake a viral stunt for distribution — own an audience via LinkedIn, X, Reddit, and wherever your community gathers.
The a16z Show · The New Startup Playbook | Box CEO Aaron Levie
Recommendation · MoneyAutomate subjective metric review by turning it into a numeric score using LLM-as-a-judge, judging a span, trace, or log data against an LLM with a guide. LLM-as-a-judge turns a subjective metric into a score across zero to five, and you can have multiple different labels; a task completion eval, for example, performs binary classification of complete vs incomplete. You're not limited to LLM-as-a-judge: use a code evaluator for deterministic checks and an LLM-as-a-judge for non-deterministic checks, and build evals for each metric that is important to you, creating a testing suite you can go against.
AI Engineer · Your Agent Fix Just Broke Something Else - Sofia Jakovcevic, Arize AI
Recommendation · MoneyYou don't need to synthesize data if you already have real user queries, real inputs, and real tool calls — pull from those directly. Ensure the dataset is representative of the average workflow users push into your systems. Capture known failures by filtering for moments where the agent often errors or fails eval metrics, capture edge cases (maybe not as common, but just as important), and capture successful workflows you want to protect, because you're also testing that you won't break something that was already working. Forgetting to capture successful workflows is a common user mistake.
AI Engineer · Your Agent Fix Just Broke Something Else - Sofia Jakovcevic, Arize AI
Method · MoneyIntrospect requests without data retention: track how many tool calls pass, how many are incorrect, and whether tool calling errors occur. Track how the shape of requests changes over time — different providers may serve different output lengths, which indicates issues. Keep a running score per provider and monitor it request to request. OpenRouter does something similar, e.g., with the exacto model approach.
endorsed
Method, not a verified protocol.
Y Combinator · Not All Open Model Tokens Are the Same | YC Paper Club
Recommendation · BodyDr. Scott Forbes recommends choosing creatine monohydrate over creatine HCL when starting creatine. Creatine HCL works just as well as monohydrate but lacks the same safety profile and is more expensive.
Garage Strength · What Actually Happens To Your Body on Creatine w/ Dr. Scott Forbes
Recommendation · BodyDr. Forbes says high creatine doses of 20-25 grams are probably not necessary to get cognitive benefits, and taking that much daily means wasting money by ingesting double or triple the required amount.
Consequential creator advice requires independent verification before Today can promote it.
Garage Strength · What Actually Happens To Your Body on Creatine w/ Dr. Scott Forbes
Recommendation · SocialCoach Kyle recommends practicing a simple greeting ('Hey, what's up, bro') consistently across environments — gym, train, commute, social settings — saying it warms you up, some interactions pop open into full conversations, and it can lead to friendships. He demonstrates by greeting a passing man (who was on the phone, so it went nowhere).
Coach Kyle · A Realistic Approach on the Train (Infield)
Recommendation · MoneySam Green advises aligning all ingested data to an internal ontology with enrichments such as countries, concepts, markets, and the time period discussed in the document.
AI Engineer · How Bridgewater Turned 100 Million Documents Into an Edge - Sam Green & Aaron Linsky, Bridgewater
Recommendation · MoneyAaron Linsky advises benchmarking search quality by collecting real questions and hand-labeling the best documents, including distractors that look helpful but aren't; this was done in less than a week.
AI Engineer · How Bridgewater Turned 100 Million Documents Into an Edge - Sam Green & Aaron Linsky, Bridgewater
Recommendation · MoneyBen Horowitz says there is no training for being a CEO — nobody will train you, so don't wait for a five-step process that doesn't exist. Look for a mentor who is a conversation partner who knows what they're talking about, not an instructor — someone you can get to a specific stuck point with and get functional, situational advice. A mentor doesn't have to be an older, senior person — a peer or someone just one step ahead of you can work. Avoid generic advisor advice (like 'no asshole rule' or frameworks 'heard at Harvard Business School') — everything is situational: your company, your product, your people, and how they work.
The a16z Show · What It Takes to Win as a Founder | Ben Horowitz
Recommendation · MoneyDane Knecht describes Cloudflare's practice of writing down engineering principles (RFC-style) covering how to build secure and available software, then distilling them into a short set of 'commandments' that AI agents can follow and enforce automatically in every part of the code review process.
described Cloudflare practice plus endorsed recommendation
No Priors: Artificial Intelligence | Technology | Startups · Democratizing Security for the Agentic Web with Cloudflare CTO Dane Knecht
Method · MoneyDane Knecht describes setting up a low-friction intake alias (agents@cloudflare) where employees email their jobs-to-be-done and get back a skill file, instead of hiring a consulting firm to catalog every task. The alias was actually staffed by interns helping people write skill files — people thought it was fully AI-automated. This bootstraps organizational AI adoption cheaply and reveals which tasks feel 'harder than it needs to be' without a consulting-firm audit.
described Cloudflare practice; the intern-staffing reveal is an anecdote
Method, not a verified protocol.
No Priors: Artificial Intelligence | Technology | Startups · Democratizing Security for the Agentic Web with Cloudflare CTO Dane Knecht
Recommendation · MoneyAndrey Rybka recommends checking agent memory for malicious sleeper instructions (memory integrity checks), and ideally starting agents fresh with no memories whatsoever, plus injection protection, isolation, and circuit breakers.
AI Engineer · Real AI Agent Attacks and What Would Have Stopped Them — Andrey Rybka, Bloomberg
Recommendation · MoneyAndrey Rybka recommends using a deterministic invocation broker, least-privileged tools, and egress controls for package and tool ingestion.
AI Engineer · Real AI Agent Attacks and What Would Have Stopped Them — Andrey Rybka, Bloomberg
Recommendation · MoneyAndrey Rybka recommends using input validation, injection protection, provenance tagging, and human approval for any high-value action to defend against marketplace-style attacks.
AI Engineer · Real AI Agent Attacks and What Would Have Stopped Them — Andrey Rybka, Bloomberg