Principle · MoneyFinance and wealth management workflows are complex and need runtime strategies as the process unfolds; you can't script every path in advance. Because workflow data is sensitive, you need a path to move these workflows entirely onto privately hosted models. In a highly regulated industry, a policy violation isn't a bug report; it has severe consequences. Execution of regulated workflows has to be highly reliable and deterministic.
AI Engineer · Reliable, Smart Workflows for Finance, Wealth and Insurance — Karthic Rao, Perpetual
Principle · MoneyIn regulated domains like finance and wealth management, workflows must be reliable enough to meet regulatory standards while still intelligent enough to handle the real world. Reliability and intelligence are not opposites; the designer chooses where to draw the boundary between them.
AI Engineer · Reliable, Smart Workflows for Finance, Wealth and Insurance — Karthic Rao, Perpetual
Principle · MoneyDeterminism should govern control flow, evidence, and side effects: what runs next, what we know, how we know it, and anything that changes the world (a record written, an email sent, a form filed). Intelligence (models) should provide interpretation and judgment: reading an ambiguous instruction, deciding which person is meant, understanding a client's reply. The model answers a typed question and the deterministic system decides what to do with the answer. The interplay between judgment and determinism is recursive: a judgment can open a new, smaller plan that is compiled, gated, and deterministic again. The boundary between deterministic and judgment parts is the design itself.
AI Engineer · Reliable, Smart Workflows for Finance, Wealth and Insurance — Karthic Rao, Perpetual
Principle · MoneyRob Snyder identifies feeling like you need to build the product yourself first before talking to customers as one of the most brutal traps when coming up with ideas. Building before customer conversations delays finding pull and wastes months you could have spent validating. His advice: face the cold wind of reality as soon as possible — don't delay the customer conversations where you can actually find pull.
Lenny's Podcast · How to find product-market fit (but, actually) | Rob Snyder (Author, The Power of Pull)
Principle · MoneyRob Snyder advises that when you show up and offer something that gets a stuck prospect unstuck and helps them get their priority done, that's when they rip the product out of your hands. This ties the offer design directly to the prospect's blocked project rather than to generic product features.
Lenny's Podcast · How to find product-market fit (but, actually) | Rob Snyder (Author, The Power of Pull)
Principle · MoneyIdeas grounded in real pull come from firsthand observation, not brainstorming; some insights wouldn't even emerge in interviews.
Lenny's Podcast · How to find product-market fit (but, actually) | Rob Snyder (Author, The Power of Pull)
Principle · MoneyThe project on the customer's to-do list is already there; the founder's job is to find it, not create it. This reframing was liberating for Rob.
Lenny's Podcast · How to find product-market fit (but, actually) | Rob Snyder (Author, The Power of Pull)
Principle · MoneyRob Snyder explains that the purpose of the pull framework is to find the people for whom the status quo is not okay — the ones already trying to change and who don't need convincing. The host adds context that there's a high bar for people to try and buy something new, and the framework is designed to get over that hump. Rob adds that when someone has pull, it should feel weird for them NOT to buy, and if a prospect has pull, they will convince you to let them buy.
Lenny's Podcast · How to find product-market fit (but, actually) | Rob Snyder (Author, The Power of Pull)
Principle · MoneyRob Snyder says seeing the case study factory in your mind radically simplifies how a startup works, and gives you a way of thinking about your startup and knowing exactly where to focus at any one point in time.
Lenny's Podcast · How to find product-market fit (but, actually) | Rob Snyder (Author, The Power of Pull)
Creator perspective · MoneyAparna Dhinakaran argues that the AI engineer's role really starts with tracing everything — tracing the agents you're actually building — running evals, and being able to learn how to test these products.
AI Engineer · What Does an AI Engineer Do Now? - Aparna Dhinakaran, Arize AI
Creator perspective · MoneySallyAnn frames evals as the primary signal for AI agents, including all evals at Arize. This is the foundational framing for the entire workshop.
AI Engineer · How to Write Your First AI Eval - SallyAnn DeLucia, Arize AI
Report · MoneyWhen a model retrieves a historical revenue number from its training corpus, there is no guarantee it was pulled correctly from the right year and scope, and it cannot be traced back to its original source. This is especially difficult with historical data.
AI Engineer · Do Not Let Models Invent Numbers - Vinoo Ganesh, Kepler
Creator perspectiveMost existing financial benchmarks prepare golden answers and compare them with the agent's answer, but in real-world tasks two analysts can reach two different answers and both could be right. Existing financial benchmarks also use relatively short and simple tasks that only capture fragments of a full unit of analyst work.
AI Engineer · Why Frontier Models Fail as Buy-Side Analysts - Janie Gu, Google DeepMind & Suyeol Yun, LinqAlpha
Creator perspectiveBenchmarks are how the AI industry measures progress, but the current state is fundamentally broken. Frontier labs use them to train and hill climb models, startups use them to choose models, and investors use them to judge technical moats. Because models hill climb exactly what the benchmark measures, measuring the wrong thing makes models worse at what they are supposed to do. Most public benchmarks test synthetic puzzles, academic trivia, and random math problems that bear little resemblance to real production agent work. Running 500 multi-turn agent tasks across 16 frontier models costs tens of thousands of dollars per sweep, and new frontier models launch every week or two, so sweeps must be re-run constantly or the benchmark becomes stale.
AI Engineer · Why Frontier Models Fail as Buy-Side Analysts - Janie Gu, Google DeepMind & Suyeol Yun, LinqAlpha
Creator perspective · MoneySofia Jakovcevic describes the common failure mode where pushing a simple change, like a one-line prompt edit, unexpectedly breaks the whole system, and warns that coding on vibes instead of numeric evaluation suite-backed data leads to shipping unverified changes.
AI Engineer · Your Agent Fix Just Broke Something Else - Sofia Jakovcevic, Arize AI
Creator perspective · MoneySofia Jakovcevic describes the manual investigation loop's bottleneck: humans are still reviewing dashboards, triaging and shipping fixes, and reviewing PRs, which is the ultimate bottleneck. Coding agents pushing PRs without knowing what's happening in production are pushing PRs blind, reliant on static code.
AI Engineer · Your Agent Fix Just Broke Something Else - Sofia Jakovcevic, Arize AI
Principle · MoneyThe real prize is building the foundations to make agents a proactive participant in your process instead of a reactive tool for humans to use — taking agents from a reactive resource that answers a human's question to proactive participants that steer high-value professionals toward the most important areas for their attention.
AI Engineer · How Bridgewater Turned 100 Million Documents Into an Edge - Sam Green & Aaron Linsky, Bridgewater
Creator perspective · MoneyHari Kumar describes showing what the firm actually gains from AI spend — both P&L and operational 'are we getting good ideas' — as a really big challenge, compounded by the time lag before results are visible.
AI Engineer · Proving Alpha: How a Hedge Fund Measures AI's ROI - Hari Kumar, Contour Asset Management
Creator perspective · MoneyEvery 23-year-old wants to start a personal brand because they see it online, but they forget the personal part — starting without real skills, accomplishments, or opinions worth watching produces content with no substance behind it.
Alex Hormozi · How I Would Build a Personal Brand in 2027 (If I Had to Start Over)
Report · MoneyWorking with agents compresses the long-horizon workflow a lot because Arca took humans out of the critical path — nobody has to babysit the goal over days; agents are on it.
AI Engineer · Inside Arca's Long-Horizon Agents That Run the Business - Aman Singh, Arca