Principle · MoneyKent argues that software engineers have gotten used to just sitting in their corner taking a Jira ticket and turning it into implementation, but that this is not how software engineering ever was. Simply turning tickets into implementations makes you look like an agent — very replaceable. The best software engineers talk to people and understand their problems; that's who builds a system that solves those problems better. The product engineer's job is to connect customer understanding to technical choices so they don't paint themselves into a corner or overbuild.
AI Engineer · Build the Right Thing: Product Engineering (Part 1) — Kent C. Dodds, EpicProduct.engineer
Principle · MoneyKent argues that project validation is critically important and that product engineers, not just product managers, own the responsibility for understanding why a feature exists. If a feature request arrives without validation of its purpose, the product engineer should go back to the PM and ask to understand more about why it exists. This validation serves two purposes: confirming the feature is worth engineering time, and answering other questions about the system. Understanding where a request comes from helps shape the system used to build it.
AI Engineer · Build the Right Thing: Product Engineering (Part 1) — Kent C. Dodds, EpicProduct.engineer
Principle · MoneyWayne Allen's framing, quoted by Kent: building the thing right is downstream of building the right thing — the product concern precedes the engineering concern. Kent adds that engineers should ensure upstream activities deliver valuable work so their engineering effort produces something worth building, and that deciding whether something is worth building is more valuable than the mechanics of building it. A product engineer should also recognize that finishing implementation doesn't guarantee customer value — you can still fail after shipping.
AI Engineer · Build the Right Thing: Product Engineering (Part 1) — Kent C. Dodds, EpicProduct.engineer
Principle · MoneyKent C. Dodds claims the most valuable software engineers at any company are those who can help answer whether something is worth building, and that this applies to internal enterprise work as well as startup apps.
AI Engineer · Build the Right Thing: Product Engineering (Part 1) — Kent C. Dodds, EpicProduct.engineer
Principle · MoneyJim Clark argues sandbox boundaries should be small enough that the agent can only perform the task you want it to execute.
AI Engineer · How Many Credentials Should Your AI Agent Have? Zero. — Jim Clark, Docker
Creator perspective · MoneyJim Clark recommends breaking an effective agent into logical sandboxes so that no single point has both unfiltered read access to 'crazy' context and the ability to act with tools.
AI Engineer · How Many Credentials Should Your AI Agent Have? Zero. — Jim Clark, Docker
Creator perspective · MoneyJim Clark defines a sandbox as a place for an agent to do work, notes that when people first started using agents the sandbox was their entire laptop with all tools, and recommends making sandbox boundaries small enough that the agent can just do the task you want it to execute. A smaller sandbox makes it safer to let the agent run for longer periods of time. The goal is not to sandbox the harness itself but to harness the tools and context flowing into the harness.
AI Engineer · How Many Credentials Should Your AI Agent Have? Zero. — Jim Clark, Docker
Principle · MoneyJim Clark's core heuristic: let agents work freely on problems, but constrain what each individual sandbox can access so the agent loop has less access than the whole workflow.
AI Engineer · How Many Credentials Should Your AI Agent Have? Zero. — Jim Clark, Docker
Creator perspective · MoneyJim Clark recommends that when thinking about agent safety, you figure out what resources are needed to do the job. When analyzing a new task, ask what information and what tools are needed to get it done. If a sandbox has the right resources and tools — and only those — it limits the blast radius of what can go wrong when the agent works.
AI Engineer · How Many Credentials Should Your AI Agent Have? Zero. — Jim Clark, Docker
Principle · MoneyJim Clark's closing principle: give your agents room to solve problems, but limit what individual sandboxes can access — sandboxes are what actually represent intent. Allowing agents to run longer but more sandboxed makes it feel safer; this is where safety comes from.
AI Engineer · How Many Credentials Should Your AI Agent Have? Zero. — Jim Clark, Docker
Creator perspectiveAdler describes rejecting the art world's enforced scarcity model — which he finds pretentious — and instead pricing his products in line with what they cost to make, which he says frees him to create infinitely many things at infinitely many price points.
Design Better · Jonathan Adler returns: The minimalist inside the maximalist
Creator perspectiveAdler argues that a successful gallery artist's currency is a clear, identifiable style: the market requires zeroing in on one style, replicating it with slight variations, and staying not very prolific, since high output could decrease the work's value.
Design Better · Jonathan Adler returns: The minimalist inside the maximalist
Principle · SocialCoach Kyle transitions into qualification by asking 'what makes you unique?', keeps the question light rather than deep or serious, redirects it back to her, and qualifies her by saying specifically what he likes — e.g., valuing a mix of maturity and youthfulness, experienced but smart, resourceful, with young fun energy.
Coach Kyle · A Realistic Approach to a Single Woman Sitting Alone on a Bench (Infield)
Principle · MoneyRita's advice, quoted by Kent: when someone says 'you should build this feature,' your first question back should be 'what are you trying to solve?'
AI Engineer · Build the Right Thing: Product Engineering (Part 2) — Kent C. Dodds, EpicProduct.engineer
Creator perspective · MoneyThe Kano model helps you make the right decision in each category — not just which one to do first, but how much technical expertise or effort to put into it. For basics, focus on reliability, completeness, and ownership — these you absolutely have to nail and do right; you don't want notifications not showing up. For performance needs, these are measurable quality gradients — you figure out how much the metric actually matters and what level of satisfied you need to get to. If your performance metrics are under the line, that's going to be a problem.
AI Engineer · Build the Right Thing: Product Engineering (Part 2) — Kent C. Dodds, EpicProduct.engineer
Principle · MoneyWhen tuning an LLM judge, prioritize recall over precision in most cases — letting wrong things through is worse than missing some correct things. Choose precision over recall only in domains where being absolutely right when it's right matters more than catching everything.
AI Engineer · From Vibes to Production: Evaluating and Shipping AI Agents That Work 101 — Laurie Voss, Arize AI
Report · MoneyLaurie Voss gives sample-size guidance: for workshop-scale experiments, 12 to 20 examples gives you direction and direct feedback. For actual shipping decisions, target 200 to 400 samples — a human-sized number you could possibly accumulate over a couple of weeks, not tens of thousands or millions like you'd use for training an ML model. There are diminishing returns on sample size: to halve your margin of error, you have to quadruple your sample size. At some point you have enough signal to be able to get along — stop growing the sample.
AI Engineer · From Vibes to Production: Evaluating and Shipping AI Agents That Work 101 — Laurie Voss, Arize AI
Principle · MoneyUse LLM evals for semantic questions like whether the analysis is good, whether financial data is accurate, whether the report is complete, and whether recommendations are well reasoned — for semantic questions, you need a judge.
AI Engineer · From Vibes to Production: Evaluating and Shipping AI Agents That Work 101 — Laurie Voss, Arize AI
Principle · MoneyFor workshop-scale experiments 12–20 examples give direct signal; for actual shipping decisions target 200–400 examples — a human-sized number accumulable over a couple of weeks, unlike ML training's tens of thousands. Halving margin of error requires quadrupling sample size, so at some point you have enough signal.
AI Engineer · From Vibes to Production: Evaluating and Shipping AI Agents That Work 101 — Laurie Voss, Arize AI
Principle · MoneyPeople closest to product requirements — PMs, customer service, sales — are best positioned to define success and can contribute eval tasks without writing code; have them describe 'good' in plain English and boil that down into the eval rubric.
AI Engineer · From Vibes to Production: Evaluating and Shipping AI Agents That Work 101 — Laurie Voss, Arize AI