Recommendation · MoneyNouha Dziri recommends building the multi-agent harness as a top-down hierarchy with a minimal design, and exposing every part of the harness so you can change one choice at a time, keep everything else fixed, and see how each choice moves reward and cost.
AI Engineer · How to Get Frontier Coding Results at 70% Lower Cost - Nouha Dziri, Cohere
Recommendation · MoneyA heavy prompt that tells the model exactly what to do — follow a plan, delegate agents, create adversarial workers, verify the task — helped a lot. A heavy prompt with detailed coordination instructions (don't step on each other's toes, use this folder, delegate throughout implementation, create adversarial workers, verify) helps both async and blocking teams significantly. Once you tell the lead how to coordinate, the async lead catches up with the blocking lead because it knows when to wait.
AI Engineer · How to Get Frontier Coding Results at 70% Lower Cost - Nouha Dziri, Cohere
Recommendation · MoneyStart with a minimal prompt that just tells the lead about the task, that it should create sub-workers, inform them about the task and tools, and give the lead tools. They started minimal to see how agents behave on their own — whether they need a lot of guidance or if a minimal prompt is enough.
AI Engineer · How to Get Frontier Coding Results at 70% Lower Cost - Nouha Dziri, Cohere
Recommendation · MindRyo Lu recommends picking one place you care about, changing one small thing, and trying it. If there's an opportunity, show your changed version to another person. You don't need to change everything at once, and you don't need to prove you are better than the machine.
AI Engineer · When We Can Do Anything, Do We Still Know What We Want? — Ryo Lu, ex-Cursor | AI Engineer Shanghai
Recommendation · MindIf you're building tools, Ryo Lu suggests starting from three small points: make important assumptions visible, make each modification small enough to feel safe, and make handed-off work retrievable when needed.
AI Engineer · When We Can Do Anything, Do We Still Know What We Want? — Ryo Lu, ex-Cursor | AI Engineer Shanghai
Recommendation · MindWhen you see a very complete answer, first give yourself time to ask which decision in it you haven't thought about yourself — and if there's a chance, show it to another person.
AI Engineer · When We Can Do Anything, Do We Still Know What We Want? — Ryo Lu, ex-Cursor | AI Engineer Shanghai
Recommendation · MoneyThe narrator advises using short-form video (TikTok, Reels, Shorts) as your discovery channel at the top of the funnel, where strangers find you before they'd ever search your name.
Ahrefs · 7 Ways to Get Website Traffic, Explained in 10 Minutes
Recommendation · MoneyThe narrator advises using paid ads for what they're actually good at: speed and testing. Use them to validate an offer or give a page that's already working an extra push.
Ahrefs · 7 Ways to Get Website Traffic, Explained in 10 Minutes
Recommendation · MoneyAdam Nahari advises embedding AI champions from the end-user group in solution design from day one: have them meet with the engineering team every other day, co-design the interface, participate in A/B testing, and drive adoption to their peers — champions adopt because they had a hand in building it.
AI Engineer · Your AI Pilots Won't Move the P&L. Here's What Will - Adam Nahari, Apollo
Recommendation · MoneyYou can easily have a local router (OpenCodex is one) that routes any image recognition to Gemini or Luna while keeping GLM for the main tasks.
Also discussed by Jeff on Y Combinator (unclassified source — not counted until classified). Not counted as independent sources.
AI Engineer · How Revolut Cut AI Coding Costs 90% With Self-Hosted Open Models - David Abitbol, Revolut
Recommendation · MoneyIf running a Blackwell generation GPU, the speaker says you don't want to run anything higher than FP4, and FP4 is best.
Also discussed by Y Combinator (unclassified source — not counted until classified). Not counted as independent sources.
AI Engineer · How Revolut Cut AI Coding Costs 90% With Self-Hosted Open Models - David Abitbol, Revolut
Recommendation · MoneyAll self-hosted endpoints are OpenAI-compatible APIs with response API, so they work inside OpenCode, Codex, and Codex app. Revolut recommends using the OpenAI-compatible endpoints inside coding harnesses.
AI Engineer · How Revolut Cut AI Coding Costs 90% With Self-Hosted Open Models - David Abitbol, Revolut
Recommendation · MoneyJesse Hall notes LiveKit's turn-detection evals and methodology are open source and reproducible, and advises running them against your own dataset to see what interruption-rate numbers you get rather than trusting published numbers.
AI Engineer · How to Build Voice Agents That Don't Fall Apart When You Interrupt - Jesse Hall, LiveKit
Recommendation · MoneyKarthic Rao recommends taking one typed call from the Harper-Burby 9 plan and pointing it at a privately hosted model to verify the same answer comes back.
AI Engineer · Reliable, Smart Workflows for Finance, Wealth and Insurance — Karthic Rao, Perpetual
Recommendation · MoneyKarthic Rao recommends logging every judgment the LLM makes at typed points so it can be inspected, identifying where deterministic code can replace it, and using every logged judgment to need less LLM judgment in the next iteration.
AI Engineer · Reliable, Smart Workflows for Finance, Wealth and Insurance — Karthic Rao, Perpetual
Recommendation · MoneyThe creator endorses making focus mode trigger automatically: start typing and typewriter scrolling happens wherever the cursor is, with the interface fading out, rather than hiding focus mode behind a menu item.
MicroConf · SaaS Product Design: How a B2C App Got to 4% Churn
Recommendation · MoneyThe creator advises positioning your product against a complex incumbent with a simple comparison — 'Dabble is like Scrivener minus the learning curve.' The speaker notes this is especially effective at conferences where people already know Scrivener, and recounts putting the comparison on Dabble's website but eventually taking it down because he felt it was a little too cheeky.
MicroConf · SaaS Product Design: How a B2C App Got to 4% Churn
Recommendation · MoneyThe creator endorses displaying users' books on a shelf with cover art, which connects the writer with their work in progress and makes them excited to write.
MicroConf · SaaS Product Design: How a B2C App Got to 4% Churn
Recommendation · MoneyThe creator endorses placing a rotating loading quote into the app's shell index HTML page so it renders instantly before JavaScript and data load, connecting with users during startup.
MicroConf · SaaS Product Design: How a B2C App Got to 4% Churn
Recommendation · MoneyRob Snyder says AI personalization for cold outreach doesn't work at the early pull-hypothesis stage — it tells prospects something they already know and looks like Claude wrote it. Personalization is not super important in the initial outreach message; a simple message without any personalization can be good enough, like 'Hey, I'm going to X conference. We're going to be pitching. Would you be open to chat?'
None stated.
Lenny's Podcast · How to find product-market fit (but, actually) | Rob Snyder (Author, The Power of Pull)