Agents & Automation

Automating the bottleneck means finding the constraint that sets your pace and pointing Claude at it: four working rules distilled from Anthropic's own engineers.

A skill file is career-long judgment packaged as instructions your agent can run. How to adopt someone's method, and why you scan it like untrusted input first.

AI Firmware is a boot layer of kernel rules, installable skills, and committed memory that makes every new agent repo start senior instead of relearning old lessons.

When an agent hits the same problem twice, encode the fix as a lint rule, CI check, or steering-file instruction that every future session inherits.

One big question in one chat makes the model decide, research, write, and grade itself in a single pass. Map the work instead: parallel research jobs, a checker that did not write the answer, a merge, and your approval before anything expensive.

Buzz is an open-source team chat from Jack Dorsey's Block where AI agents are members of the channel. What that buys you (swap the model, keep the context), what it can already do, and whether your team should try it yet.

A new agent should not run your inbox on day one. Give it one job with read-and-draft access only, check its work until your corrections stop, then add the next tool, and let it run unattended only after it has earned that.

In ChatGPT, any chat can search every other chat, send work into them, and collect the results. How to find a lost conversation by asking, delegate work between chats, publish a result to a shareable page, and reach it all from your phone.

Grok Bot, the new beta app from SpaceXAI and Cursor, gives you a team of AI agents that keep working with your laptop shut, each on its own cloud computer. How work gets routed between them, how recorded skills and triggered routines run themselves, and which work should stay at your desk.

Allie K. Miller runs 34 AI agents, and her day is reviewing what they propose, deciding, and handling escalations. What an agent needs before it can start work you never assigned (written goals, fresh context, tools, triggers), the four-stage ramp from one agent to a workforce, and the watchdog jobs worth assigning first.

Boris Cherny, who created Claude Code, maps four steps between chatting with an AI and a system that starts work you never assigned. Find your step from the bottleneck you feel, and make the one change that moves you up.

Models get repriced, capped, and retired; their way of working can be kept. Extract the top model's discipline into skill files, routing tables, and planner-and-worker setups any model can run.

A routine is a saved instruction Claude runs on its own, on a schedule or an incoming signal. What it replaces, and why an unattended run has to be written tighter than one you watch.

Wrong agent answers usually trace to the files the agent reads. The four ways a growing knowledge base drifts from reality, and the habits (a router file, read-only audits, scheduled pulls, backtracking) that catch the drift.

A rerun loop makes the agent redo its own work until a check passes. Script the repetitive half of a skill, end it with a pass or fail verdict, wrap both in a loop, and keep a human at the checkpoints where errors are expensive.

Anything your agent reads (a web page, a resume, a calendar invite) can carry hidden instructions it will follow. Why the attack has no clean fix, and the five defense layers that shrink what a successful one can do.

An agent sent a discount code to 150,000 people because a to-do item looked like an instruction. A rule in the prompt is a suggestion; a rule built into the tools is a restriction. Scope every key to the least the task needs.

Cody Schneider builds agents that find leads and post content while nobody watches. The method: turn the human process into scripts on a schedule, and call the model only where a judgment is needed. Both of his builds, tool by tool.

Skills built on one person's machine stay there, and the copies you send around go stale the moment anyone edits one. Put the team's skills in one GitHub repository, install it as a plugin, and every edit reaches every teammate's agent.

Agent workflows run one step after another by default, which is the slowest shape. Ask at each step whether it needs the result of the step before it, then re-shape the work: fan out independent steps, route requests down branches, loop until a check passes, and cap what each shape can waste.

Theo let Claude Code save memories on its own, then audited them: 45 files, written three times more often than read, most never opened again. How to run the same audit on your machine, the three ways saved memories rot, the ladder that finds a better home for each lesson, and what a harness built to pass the audit looks like.

A memory store is a workspace-scoped directory your agent reads and writes with normal file tools, carrying preferences and prior mistakes across sessions.