Goals · 7 methods

AI at work

Use AI to do more and better work.

Dan Zhang· Fast Company· Nov 21

Your best people are training your AI. Pay them for it

ClickUp argues companies should select employees with strong process knowledge to build useful AI skills, measure verified impact, and reward them for making their expertise reusable.

  1. Choose builders with strong performance and knowledge of where work processes break.
  2. Classify each AI skill by impact, record a baseline, then compare results and subtract tool and maintenance costs.
  3. Set rewards by verified value and how long it lasts: use bonuses for one-off gains and raises for ongoing contributions.

Try todayList one recurring process bottleneck and record its current cost, time, or failure rate before testing an AI skill.

Gergely Orosz· The Pragmatic Engineer· Nov 17

Inference engineering for software engineers

The article explains how to move LLM workloads from frontier-lab APIs to open models, from testing providers through API replacements to dedicated GPUs when usage demands it.

  1. Use evals to select an open model for your use case, then compare inference providers and deployment options.
  2. Start with a multi-tenant API swap; track latency, cache hits, rate limits, and usage costs.
  3. Move to dedicated GPUs when rate limits, inconsistent performance, or high usage costs justify it; size capacity for your workload.

Try todayRun an eval on one workload to compare your current model with an open-weight alternative.

Lenny Rachitsky· Lenny's Newsletter· Nov 16

🎙️ How I AI: Fable 5.2 Review + How I fight token fraud as a solo founder + Building a Claude Code family OS

Scott Farrar uses Claude Code and Obsidian to build a searchable household memory, organize family tasks, and compare insurance quotes. He favors persistent, editable files over relying on long AI chats.

  1. Store reusable context in editable Markdown files so you can inspect and correct it across AI sessions.
  2. Photograph appliances and let AI organize model numbers, serial numbers, and warranty details into a searchable inventory.
  3. Capture household tasks in one master list, then have AI organize them by priority, cost, and supporting information.

Try todayPhotograph one appliance and use AI to record its model, serial number, and warranty details.

Gergely Orosz· The Pragmatic Engineer· Nov 16

Software engineering after code becomes cheap with Satya Nadella

Satya Nadella argues that as AI makes code generation cheaper, teams need stronger verification, flexible model infrastructure, and new ways to improve how they work.

  1. Keep the agent harness, memory, and model separate so you can change providers without lock-in.
  2. Build private evaluations to compare models, and verify agent outputs as if agents were insider risks.
  3. Route work between local models and cloud models based on capability and cost.

Try todayList which parts of your AI setup are coupled, then identify one you could make easier to replace.

Anne Neuberger· a16z· Nov 12

Super Intelligence-Era Critical Infrastructure Cyber Defense Plan

The plan focuses national cyber defense on critical operators and shared vendors, using AI to find and fix vulnerabilities while setting measurable, accountable sector targets.

  1. Prioritize operators and shared vendors whose failure could cascade across power, payments, water, communications and hospitals.
  2. Use AI to find vulnerabilities and test patches or workarounds, including for legacy systems that cannot be patched.
  3. Remove internet exposure from control systems, segment networks and test whether operations can continue manually for 72 hours.

Try todayMap your critical systems and vendors, then flag exposed control systems and dependencies that could cause cascading failures.

Claire Vo· Lenny's Newsletter· Nov 11

Claude Code + Obsidian: My family’s AI brain | Scott Farrar

Scott Farrar organizes household information in Markdown files for Claude Code and Obsidian, then uses the shared system for tasks, insurance shopping, and a family display.

  1. Create a Markdown knowledge base for household records such as appliances, warranties, and insurance documents.
  2. Share a household to-do file, and display priority tasks and recent wins where the family can see them.
  3. Have students teach an AI tutor, keeping the children responsible for doing the thinking.

Try todayCreate a Markdown file with one household task and the information needed to complete it.

Lenny Rachitsky· Lenny's Newsletter· Nov 9

🎙️ How I AI: How this founder built a better Duolingo with Codex

Founder William Imoh used AI agents to build native language-learning apps and support company operations. He shares practical ways to structure prompts, assign agents roles, and review their work.

  1. Write agent requests as epics and user stories tied to specific user problems.
  2. Give each agent a clear role and connect it to relevant work context, while protecting sensitive data.
  3. When an agent stalls, start a fresh session, break the task into steps, and ask it or another agent to review the result.

Try todayRewrite one AI task as a user story tied to a specific problem, then ask the agent to review its output.

The Daily Brief. Five ideas, read by an editor, in your inbox by 6 a.m.

We read the founders, operators and researchers worth reading, keep the few ideas that hold up, and tell you how to use them. Free, and one click to leave.