2-1b2b0.1a Prioritize upstream productivity

That which downstream into other productivities (e.g., exercise).

Related:

• 1-1c6a2a1 Be careful with ‘just-in time productivity’
• 2-1a7a1 (1) Increase your productivity; (2) Don’t let your income increase faster than your productivity; (3) Don’t let your debt increase faster than your income
• 2-1b2b ‘Second-order thinking’ - Solve the root cause of a problem (prevention) and not symptoms. Be smart-lazy.

---

E.g.,

• Every’s AI-native engineering philosophy
  • ”If you don’t trust the results, fix the system, instead of compensating by doing everything yourself."
  • "A system that produces code is more valuable than any individual piece of code."
  • "Effective compound engineers write less code than before and ship more.”
    • <> 1-1a5b5 Shipping perfection means you are too late - you have to find a balance of finding the right moment and being fixable
  • ”First attempts have a 95 percent garbage rate. Second attempts are still 50 percent. This isn’t failure—it’s the process… Focus on iterating fast enough that your third attempt lands in less time than attempt one.”
    • <> 9-4b3a Write a bad version 1.0 as fast as you can
  • ”The developer who reviews 10 AI implementations understands more patterns than the one who hand-typed two.”
    • <> 9-1b0a When you have learned something, the way it exists in your mind is the same as the way it exists in the mind of the inventor
  • “Planning, reviewing, and ensuring quality standards is the work. You did the thinking. All the AI did was the writing.”
  • Extract your taste into the system—“That taste usually isn’t documented anywhere. It lives in senior engineers’ heads and is transferred through code review. This neither scales nor lets others on the team learn.” Solution: write preferences into CLAUDE.md/AGENTS.md, build specialized agents, add slash commands, point agent at existing style guides. “Once the AI understands how you like to write code, it’ll produce code you actually approve instead of code you have to fix.”
    • <> 2-1b2e ‘Algorithms’ & ‘Replication’ - Share your problem-situation as clearly possible, so that others can make decisions without you on your behalf (i.e., multiply)
    • <> 5-2b1b Writing down makes it easier for your conscious to see the unconscious
    • <> 9-4c2 ‘Programs should be written for people to read, and only incidentally for machines to execute.’ ‘Design to express algorithms, and only incidentally tell machines how to execute them.’
    • <> 10-1b4e Developing a taste means transitioning from being obsessed with where it came from (analog) to focusing on what it is and what it can do (digital)
  • The 50/50 rule—“allocate 50 percent of engineering time to building features, and 50 percent to improving the system.” Traditional teams do 90/10. “An hour spent creating a review agent saves 10 hours of review over the next year… System improvements make work progressively faster and easier, but feature work doesn’t.”
    • <> 2-1b2b0.1 Give a man a fish and you feed him for a day. Teach a man to fish and you feed him for a lifetime.
  • Trust the process, build safety nets—“Trust doesn’t mean blind faith. It means setting up guardrails such as tests, automatic review, and monitoring that flag issues… When you feel as if you can’t trust the output, don’t compensate by switching to manually reviewing the code. Add a system that makes that step trustworthy.”
  • Make your environment agent-native—“If a developer can see or do something, the agent should be allowed to see or do it too… Anything that you don’t let the agent handle, you have to do yourself manually. The goal should be full environmental parity between human and AI developers."
    • "Every capability you withhold from the AI becomes a task you have to do yourself.” Checklist: can your agent run tests, run the app locally, run linters/type checkers, create branches/commits/PRs, read PR comments, view logs, take screenshots, inspect network requests, access error tracking? Progressive levels: Level 1 (file access + tests + git), Level 2 (browser + local logs + PRs), Level 3 (production logs read-only + error tracking), Level 4 (ticket systems + deployment + external services).
  • Parallelization is your friend—“The new bottleneck is compute—how many agents you can run at once.”
    • <> Naruto and LLMs
  • Plans are the new code—“The plan document is now the most important thing you produce… Fixing ideas on paper is cheaper than fixing code later."
  • "Embrace the discomfort of letting go… imperfect results that scale, rather than perfect results that don’t."
  • "The principles extend beyond engineering to design, research, or even writing—any discipline where codifying taste and context help make future work go faster and easier.”