Recommendation · MoneySam recommends identifying which of your company's leaders are out there constantly networking and presenting on the company's behalf — speaking tours, pay-to-play panels backed by your chief communications officer, or super-dynamic networkers even at director level — and saving Sales Navigator searches on them to monitor and capitalize on their expanding networks.
30 Minutes to President's Club | No-Nonsense Sales · #618: LinkedIn Sales Navigator: How I Get Warm Intros Every Week
Method · MoneyAl Chen endorses a simple disaster recovery pattern: run ClickHouse in a primary region, a job that backs up to a bucket, and another job that restores that bucket to a DR region, with multiple pods in that region holding replicas of the data.
Method, not a verified protocol.
AI Engineer · Shipping AI Into 5 Major Banks Without a 12-Month Compliance Wait — Al Chen, Cisco
Recommendation · MoneyDoneyli recommends scoping investigations down from thousands of traces to dozens or hundreds for more targeted analysis. Instead of going through thousands of traces to find the most problematic agent activity, filter traces by a specific failure mode (e.g., dispute not resolved) to find instances where the customer's issue wasn't resolved, and save filtered trace views so you don't need to re-filter every time. Focus investigation on the thumbs-down interactions (e.g., the 18%) to understand what's been happening.
AI Engineer · Evals for High-Stakes AI Agents — Doneyli De Jesus & Annabell Schäfer, Langfuse
Recommendation · MoneySupport multimodal evaluation: for contact center calls, run speech-to-text, then evaluate the text for violations, and link the media, transcript, and assistant response in the trace.
endorsed
AI Engineer · Evals for High-Stakes AI Agents — Doneyli De Jesus & Annabell Schäfer, Langfuse
Method · MoneyKade is basically chronically on LinkedIn and interacts with post notifications within five minutes of seeing them; he currently has about 75 prospects' post notifications turned on. He notes there's not really a way to automate LinkedIn — engagement is totally manual — and that commenting on a prospect's post sometimes brings them back to the DM.
Method, not a verified protocol.
30 Minutes to President's Club | No-Nonsense Sales · #615: How to Master Social Selling on LinkedIn to Book Meetings | Kade Hinkle | 30MPC Hall of Fame
Recommendation · MoneyThe host recommends making an offer of value (e.g., 'I can send over a few specific examples') instead of an interest-based CTA like 'do you want to learn more' — per a 300M+ cold email analysis with Gong, offers of value get far more meetings agreed to. He adds that adding genuine value inside the outreach message itself builds trust that you'll add value in the meeting, increasing response rates.
30 Minutes to President's Club | No-Nonsense Sales · #615: How to Master Social Selling on LinkedIn to Book Meetings | Kade Hinkle | 30MPC Hall of Fame
Method · MoneyAlex's video message structure: come in with relevant stuff he knows about the person and their business, connect it to a problem he is hypothesizing, then go into a quick one-sentence pitch of what he sells (a team training program).
Method, not a verified protocol.
30 Minutes to President's Club | No-Nonsense Sales · #615: How to Master Social Selling on LinkedIn to Book Meetings | Kade Hinkle | 30MPC Hall of Fame
Recommendation · MoneyKade Hinkle only comments on a prospect's post when he can add genuine value — never just an emoji or an AI-generated comment. He comments on personal posts rather than company posts, because company content rarely allows a relevant, value-adding comment. When commenting on a hiring post, he tags people in his network who could fill the open role, which adds value and builds goodwill.
30 Minutes to President's Club | No-Nonsense Sales · #615: How to Master Social Selling on LinkedIn to Book Meetings | Kade Hinkle | 30MPC Hall of Fame
Recommendation · MoneyTry the same design-principles exercise on your own site or product: scroll through and ask where your eye goes first, what feels most important, what feels grouped, what feels scattered, whether anything feels off on the page, and why that might be.
NNgroup · 5 Design Principles in Context: Stripe
RecommendationPamela Fox recommends defining functions for LLM tool calling using JSON schema, which is what the LLMs understand. She emphasizes that writing a function's JSON schema is itself prompt engineering: you decide what type to use, what description to have, and what the parameter names should be, because the LLM decides how to call the function based on all that information.
AI Engineer · The Model Swap Workshop: GPT-5.5 vs Kimi vs DeepSeek vs Mistral — Pamela Fox, Microsoft
Anecdote · MoneyJohn's features included an active investor or significant shareholder publicly calling for a supported sale within 12 months, share price at least 60% below its five-year high, or the company being named Dominion. The top competitor (Zane) used a hard-coded exclusion list of specific companies — an example of overfitting to the known answer set. John's model also showed some overfitting, though the speaker argues it would probably perform better. The speaker regrets removing the holdout set, which he'd dropped to keep the eval simple, because it enabled this overfitting.
AI Engineer · Backtest Your Agent Against Yesterday's Web — Ryan Ahern, Exa
Reported positionWhen NBC moved Seinfeld to Thursday, Jerry's response was to accept it: 'Thank you very much. We will take the new time slot.' Larry David objected: 'If they didn't watch us on Wednesdays, I don't want them watching us on Thursdays. We've earned this weird niche audience that is our people.' The hosts endorse Jerry's choice as a good business decision — accepting a bigger audience/distribution opportunity over protecting a niche audience.
Acquired · Seinfeld
RecommendationAmeisen says the on-ramp into interpretability is mostly self-studying and reading the key papers, then experimenting on open-source interpretability platforms like neuronpedia.org, which hosts models and lets you run interpretability experiments on them. He notes the field is so new that it doesn't require a lot of prior knowledge, and if you're at all curious you should just give it a try because there's so much to explore.
Lenny's Podcast · Inside the black box of AI: How AI models actually think | Emmanuel Ameisen (Anthropic)
Recommendation · MoneyA speaker describes using unit tests around model-written JSON to verify in different ways, catch edge cases, and confirm the output is a real representation of what you expect from a system-one model — and keeping those tests around as an ongoing check so the code doesn't drift over time.
anecdotal stance
AI Engineer · Teach Your Agent to Prove It's Done — Nick Nisi & Zack Proser, WorkOS
Recommendation · MoneyUse unit tests around model-generated JSON to verify it is a real representation of what you expect from a System 1 model, catch edge cases, and keep those tests as an ongoing check to make sure output doesn't drift over time.
AI Engineer · Teach Your Agent to Prove It's Done — Nick Nisi & Zack Proser, WorkOS
Method · MoneyRun the doctor command, add an email (ada@example.test), then check storage to verify persistence. If persistence fails, run reload and then snapshot to verify that storage is actually empty. Run a snapshot command to capture the page as JSON and verify content is present, and take a screenshot to make verification easier and prove state back to the human.
Consequential creator advice requires independent verification before Today can promote it.
Method, not a verified protocol.
AI Engineer · Teach Your Agent to Prove It's Done — Nick Nisi & Zack Proser, WorkOS
Recommendation · MoneySurface data to the model in different ways so it can interpret it and keep the model in the loop on what's going on. For example, print out the coordinates of a ball at every point in a canvas game to surface bugs the model cannot see visually — if you surface that to the model, it can help you debug a canvas it doesn't actually have eyes to see.
AI Engineer · Teach Your Agent to Prove It's Done — Nick Nisi & Zack Proser, WorkOS
Recommendation · MoneyDenys Linkov recommends talking to internal government and regulatory teams as part of handling regulations and reading the NAIC reports, since that organization publishes a number of different reports on AI regulation in insurance.
AI Engineer · Insurance Claims Are AI's Messiest Data Problem — Denys Linkov, Wisedocs
Method · MoneyOnboarding starts with a few sample files worked through in various formats — typically files customers are immediately working on or trying to back-test. A review process compares what the customer extracted versus what the system extracted, aligning with the information the customer wants to see and their standards (e.g., date of injury vs report date is a matter of aligning business objectives). Then process a few standard cases; if things work well, start a broader pilot, and after a successful pilot, integrate with upstream providers via API integrations, dealing with all the messiness, apart from hand-uploaded documents.
Method, not a verified protocol.
AI Engineer · Insurance Claims Are AI's Messiest Data Problem — Denys Linkov, Wisedocs
Recommendation · MoneyThe creator recommends converting user intent and business intent into a bounding box of what agents can do, and calculating the intersection of that to say, here's what you can actually do.
AI Engineer · An AI Agent Got Its User Banned From Resy — Keycard, JPMorgan Chase & AAuth