35,848 results · Most relevant
The first thing to assess in an athlete wanting to get explosive is whether they are strong enough to hold the required positions. Many athletes chasing sprint speed or power simply lack foundational strength, so no amount of sprint work fixes it. A good foundation of strength is needed to execute movement with good technique.
On back squats in season, volume is moderate to low, accepting that some strength will be lost for most sports. Athletes in higher contact or higher output sports should accept that strength and fitness will decay somewhat during the season. It is important for the coach — and for the athlete psychologically — to accept that gym performance will decline in-season. Strength lost in-season can be rectified later and will come back. If you hit your strength numbers in the off-season and your sport is going well, you should not stress about in-season gym numbers. If you could back squat 250 kg in the off-season but 210 kg moves really slow in-season, it doesn't matter as long as sport performance is going well. Occasionally hitting a fast single at 85-90% of a true max is a bonus, not arequire
Explosiveness is worthless if you cannot use it consistently; sequencing matters more than gripping and ripping. Without conditioning, athletes in jiu-jitsu, judo, or rugby who are no good after the first 20% waste their single big outputs. The second thing many athletes forget is conditioning — do you have work capacity for your sport?
The air bike requires zero technique — fixed feet and hands mean you cannot do it wrong — so any athlete can push conditioning hard immediately, unlike sprints or 400s which require sprint skill to derive conditioning benefit. It is available in many gyms with similar recording standards across companies, and even the guest's parents in their late 60s could use one effectively straight away. It drives extreme discomfort quickly and repeatedly without the danger markers of field sprinting.
Endorsed heuristics plus one anecdote about the guest's parents
Spin is described as a vendor-agnostic CNCF project for building WebAssembly applications meant to run on the server or in the cloud. It consists of three pillars: a CLI for developers' day-to-day inner-loop concerns, a runtime based on Wasmtime that can run Wasm apps locally, on microcontrollers, or in constrained environments, and language-specific SDKs that boost productivity for tasks like HTTP and database interaction. Spin is language agnostic, supporting any language that compiles to WebAssembly including WASI.
Patterson argues that having two out of three elements of the lethal trifecta is completely unacceptable in any secure enterprise environment, and having all three means it is not if but when a security incident will occur.
endorsed
Patterson explains the risk mechanics behind the lethal trifecta: an agent accepts injected content as a source of truth and can do something unexpected; prompt injection or context poisoning can unwittingly escalate an agent's privilege so it can drop tables, make requests, or post things it shouldn't; and unrestricted internet access lets agents go straight to the model provider and bypass oversight.
Look at the magnitude of the strength: every entrepreneur who gets to a top firm is pretty good, but the question is whether they are outstanding — better than everybody else at something — and that is what makes a company and an entrepreneur great.
Horowitz says a16z once had 'urban HR' and the lesson was that it's talent, not black talent — you just have to be able to see the talent. Everyone is talent blind, which is why you need a process and a methodology. The a16z hiring process was created to get around not knowing the skills you don't have, and this approach has been a big advantage in venture capital because they get to people others never reach.
Horowitz's own account and interpretation of a16z's experience
Because entrepreneurs are hard to evaluate, you need the idea to help evaluate the entrepreneur: what they went through in life to get to it, how good the idea is, and how much original work they did.
Analytical people can figure out everything wrong with any idea, company, or entrepreneur — but that's not what matters; what matters is whether their strengths are world-class.
Compliance agents always start in the repo where the agent lives, so they may not be able to take action in other repositories. You need to start the agent in the other repo so that it has permissions in that repository.
Deploy the principle that anything you can do with a human, you should be able to do with an agent: same tools, context, goals, accountability, and visibility, because the primitives are nearly identical even if scale differs; audit the product portfolio for gaps and default to following the same human patterns when a design deviates.
The lesson from the AI-builder slowdown: teams of pure builders fail because everyone rows and no one steers; when everyone is rowing, be clear who is steering and directing, and if the person who is supposed to be steering is vibe coding, they spend no time actually directing the ship.
Users gravitate to the AI closest to their workflow; workflow proximity matters more than model quality when deciding where AI adoption happens, so for existing SaaS products with real users, evolve the AI experience rather than rebuilding from scratch.
Add a chat surface to each SaaS app even when building dedicated UIs: chat reveals the full set of use cases while dedicated UIs serve the best ones, analogous to how the command line persisted as the universal interface while successful products layered dedicated UIs on top of it.
Scale always changes the answer. The same knob can have opposite verdicts depending on which side of the bottleneck sets it. Tuning parameters like zero-copy waste and spreading workers become important only at larger scales and may not manifest as obvious improvements at smaller levels.
Live API calls are easier to build, easier to add on, and cheaper to run than synced context — prefer them when you know you'll have simple lookups or simple retrievals. Cached data is absolutely essential for semantic data, complex lookups, and deep questions about any data. Cached/synced context is more complicated to set up and build, more expensive to run because you need a vector database and constant syncing, and tougher to keep up to date. Building rich data for any question is a mix of some live lookups and some cache lookups coming together.
When a live API lookup is needed, limit the pull to only the tickets or records needed (e.g., five tickets) rather than pulling full history, because live lookups run into timeout risk. Pulling full history during live lookups causes timeout risk.
Cached/synced context is more complicated to set up, more expensive to run, and tougher to keep up to date because you need a vector database and constant syncing. You have to work on your strategy for how you store data in the context layer and how you retrieve from it for the synced cache layer.
Choose sources → to get a library of your own and a daily brief.