Guides
Short, original AI guides for devs and EMs — filter by tool or topic, search, and sort. Newest first.
73 guides
Your AI Fluency Collapses Exactly Where the Code Ships
Anthropic classified 9,830 Claude conversations. The ones that produce code get briefed better and audited worse — fact-checking falls 3.7pp. The index publishes no personal score, so here is how to run one on yourself.
ChatGPT Work vs Claude Cowork: the answer is in your admin console
Two workspace agents shipped 48 hours apart. The thing that actually decides which one your team runs isn't the model — it's a toggle an admin flipped before anyone got to try either. Here's the seam, and my routing rule.
Managed Agents: Anthropic Now Runs the Loop, Not Just the Model
Claude Managed Agents moves the agent loop, the sandbox and the session state onto Anthropic's infrastructure. The line nobody reads: it isn't eligible for Zero Data Retention or a HIPAA BAA — and self-hosting the sandbox doesn't buy that back.
A 3-Step AI Cost Audit: Seats, Tokens, and the Hours Nobody Books
Two thirds of companies have an AI dashboard. Barely a quarter can actually see what AI costs them. Here is the 3-step audit I run on a tooling budget: what to list, which number to pull for each line, and the threshold that means kill it.
97% vs 47% — Your Job Title Says Nothing About Your AI Exposure
Anthropic's index found 97% of tasks are automation-dominant through the API and 47% in chat. Same model, overlapping tasks. Exposure follows the deployment, not the job title — so stop forecasting the fog and run the migration test instead.
The CV Prompt That Screens You Out First
88% of employers say qualified high-skills candidates get vetted out for not matching the exact criteria in the posting. So don't ask AI to improve your CV — make it screen you out. The prompt, and the three decisions that keep it honest.
The Mock Interview Prompt That Refuses to Help You
The structured interview is the strongest single predictor of job performance, and also the one with the widest spread, because it can be built well or badly. A chat model builds you the bad one. Here is the prompt that installs the structure instead.
A Claude Code Plugin Adds No New Powers. It Adds Everyone Else.
Everything a plugin does, a loose .claude/ folder already does. What packaging buys you is distribution — and it costs you your command names. Here is the manifest, the install commands, and the rule for when that trade is worth making.
Claude Tag: The Channel Is the Permission, Not the Person
Anthropic put Claude in Slack as a real teammate with its own service accounts. The sentence that should drive your decision is buried in the identity docs — access attaches to the channel, so every member of that channel inherits it. Here's the model, the four setup steps, and where they stall.
Four Places to Run Claude Code — and What Each One Silently Drops
A cloud session clones your repo into a fresh Anthropic VM — so your personal skills, agents and every MCP server you added with `claude mcp add` are simply not there. Surface choice isn't a UI preference, it's a capability choice. Here's the decision rule by task shape, plus the commit that makes your setup portable.
12 Agent Loops Worth Copying (One of Them Silently Does Nothing)
The obvious loop — /loop 20m /code-review — is a no-op: the bundled reviewer refuses model invocation, so a scheduled fire delivers it as plain text. Here are 12 loops that actually run, grouped by the only thing that distinguishes them, plus the caps that quietly kill them.
Your Weekly Setup Review Needs a Delete Column
Adding your 30th skill can quietly make the other 29 harder to find: the skill listing gets ~1% of the context window, and Claude Code drops descriptions from the skills you invoke least. Here is the Friday ritual, the routing table for each lesson, and the subtraction pass everyone skips.
Your Agent's Memory Is an Index, Not a Diary
Claude Code writes its own notes to disk now — but only the first 200 lines of the index ever load, and the folder never leaves your laptop. Here is the file layout, the routing rule for what earns a memory, and the one setting that stops three weeks of learning from dying with a machine.
Three Ways to Build a Second Brain — Pick by Who Can Read It
A hosted project, a local Markdown vault, or a folder in git. The features barely differ; what differs is whether a brand-new agent session can answer from your notes with zero uploads. Here are the three layouts, the honest catch in each, and the structure I'd copy.
A Good Handoff Is Mostly a List of Dead Ends
Compaction and `/resume` both preserve what happened. Neither preserves what you decided against — and the abandoned attempts are the most expensive thing in a long session. Here's the handoff file that fixes it, the prompt that writes it, and the one-liner that generates it from a session you already closed.
Your Knowledge File Fails on Contradictions, Not Gaps
One curated file makes Claude an expert in your domain — until two lines disagree and nothing says which wins. Claude can't detect that, and it can't tell which of your facts expired in March. Here's the skeleton I use, with dates, a precedence block, and the prompt that proves the file was read.
Hallucination Is a Broken Off-Switch, Not a Knowledge Gap
Anthropic traced the circuit that makes Claude decline a question — and found it is ON by default. Fabrication is that switch being overridden, not the model running out of facts. Three prompts that exploit the difference, plus what each one still misses.
Intent Engineering: Stop Writing Steps, Start Writing Done
Anthropic's own docs now say a hand-written step-by-step plan often reasons worse than the words 'think thoroughly' — and that on Claude Opus 5 you should delete the 'verify your answer' line you added last year. Here's the shift, a real before/after, and a four-slot intent template.
Fable 5 Prompting: Five Moves, and the First Is a Deletion
The line most prompt libraries still carry — explain your reasoning — is now a documented way to get refused on Fable 5 and silently downgraded to Opus 4.8. Five model-specific moves from Anthropic's own docs, with the parameter names, the defaults, and where each one backfires.
Build a Claude Skill Three Ways: Write It, Capture It, or Install It
All three paths end in roughly the same 30-line file. What they don't produce equally is the description — the only part of a skill that sits in context permanently, and the only part that decides whether it ever fires. Pick your path by where the knowledge lives today.
When a Claude Skill Never Fires: Six Causes, in Diagnostic Order
The cruelest failure mode is a broken skill whose slash command works perfectly. Malformed frontmatter still loads the body, so /my-skill runs fine while Claude has no description to match against — and you conclude the skill is healthy. Here is the order to check things in.
Turn a PDF Guide Into a Skill by Deleting Most of It
A document explains to a reader. A skill instructs an agent whose every line is a recurring token cost — so a faithful conversion produces a bad skill. What to cut, what to move into bundled files, and the one field your PDF can never give you.
Sonnet 5: The Price Didn't Change, Your Bill Will
Sonnet 5 lists at the same $3/$15 as Sonnet 4.6, but a new tokenizer turns the same input into up to 1.35x the tokens — and the discount hiding that expires August 31, 2026. Here are the specs that decide something, plus the routing rule I run.
Laid Off? These 30 Days Fill Your Calendar, Not Your Résumé
A remote US or EU hiring loop almost never closes inside a month, so a 30-day plan that promises an offer is lying to you. Here is the week-by-week version that ends in a full pipeline instead — and where the free AI credentials honestly belong in it.