Measure reliability before shipping to customers
When shipping a product, measure the reliability of the product before giving it to customers, and aim for around 92% reliability before expecting customers to fully hand over workflows to your agent.
39,871 results · Most relevant
When shipping a product, measure the reliability of the product before giving it to customers, and aim for around 92% reliability before expecting customers to fully hand over workflows to your agent.
Asking for a whole not-fully-thought-out feature set back in one shot means asking without really understanding what you're asking for. Gabriel Martinez advises not to throw a whole not-fully-thought-out feature set back in one shot without really understanding what it is you're asking for.
Gelsinger advises against committing capital to new datacenter/GPU projects without first securing the energy to power them — energy will be an upper bound, and more project defaults (like the Oracle/Comet case) are coming.
When she subtly implies she's in a rush (food, dinner plans), recognize you only have a short window — plan for roughly a five-minute interaction instead of trying to bounce to another location.
If she starts to leave mid-interaction (e.g., to get her food), position yourself slightly off to the side and reopen when she returns. If she completely leaves and avoids you, accept it gracefully — 'that's all good.'
Mention your work indirectly (e.g., 'I use WhatsApp a lot for my business') so she asks what you do, letting you present it in an attractive way. When she asks, deliver a demonstration of higher value in your answer.
After getting the number, text later, ask how dinner's going, talk about what you have going on, and then meet up — even if it's just grabbing a drink or meeting at a venue.
Ask 'What are you doing later?' near the end of the interaction to set up the meetup plan. If she's unsure of her plans, ask whether she'll be in the area so you can gauge feasibility of meeting up.
When passing an interaction to Figma Make, use the intro animation and keep the transitions as-is (honor the transitions).
When asking an LLM to reproduce an animation, request a full breakdown including duration, repeat behavior, value ranges, and easing curves.
Vlad recommends connecting agents across frameworks via a platform that handles runtime-binding, persistency, channels, filtering, security, and observability for you; use a registry so agents can discover each other automatically, do message filtering within the platform rather than the client side, and keep humans in the loop by letting your agent invite you into every conversational space.
Rogan says the first step to engineering society positively is turning your back on the state: anyone could drive a truck to a dumpster and clean up trash instead of waiting for the federal government, give water to people out there (even ones that seem terrifying or possessed), and help without filming it or doing it for any reason.
Philipp Schmid advises that environments alone are not enough: the model needs the right context, information, or tools to work inside those environments. You can provide sources to the environment: GCS buckets, GitHub repositories, or inline files.
Philipp Schmid recommends reusing environments: in a multi-turn conversation you can use the same environment from your previous turn for a new interaction (e.g., research the latest tech news and create a file, then follow up in that same environment); you can share environments between different agents — three agents working inside the same environment can share context through the file system — and you can use sub-agents to pick up work another agent has already done.
Joe Liemandt argues the single best use of AI for education is giving kids individualized lesson plans based on their level to catch them up to grade level — not just giving kids chatbots.
Michael Lee recommends restructuring the organization to leverage machines that work 24-7 and scale with electricity, rather than just giving small AI tools to each human.
Dru Knox describes turning any skill into an automated workflow via a command called teslaunch, which takes a skill that codifies a workflow, sandboxes the agent with appropriate permissions, makes it long-running, and gives it GitHub tokens so it can put up a PR and respond to comments.
endorsed
Dru Knox recommends giving the agent CLI and API access and a way to click through your product in order to build it.
endorsed
Dru Knox recommends tracking manual takeovers and human PR comments as metrics to measure progress toward a software factory. Over time, you want to see more PRs initiated without human input as a sign that you're automating more. During the transition, keep quality constant or accept only a very small dip, expecting better quality in the code produced as the ultimate payoff.
When the agent needs to follow a strict rule, state it with no-exceptions language in the guidelines, for example 'we always use our AppHutter component for navigation, no exceptions'.
Choose sources → to get a library of your own and a daily brief.