Claude, written up by
people who use it daily.
Claude Code, Claude Cowork, Claude Design, what tokens actually cost you, and what each new model changes. Sourced, dated, and free of the breathless stuff.
The Six-Month Claude Catch-Up
Five major Claude releases since February. What actually changed, which defaults moved underneath you, and how to catch a team up without the whiplash.
Managed Agents 101: Building Your First Autonomous Claude Workflow
Claude Managed Agents moved from public beta in April 2026 to a steadily more capable agent platform. What the feature does, and where a training session pays off fastest.
Claude Voice Mode: Now on Opus and Sonnet
Voice Mode expanded beyond Haiku to Opus and Sonnet on 23 July 2026, with mid-conversation model switching and Gmail and Slack access. What changed, and the one boundary to set before rollout.
Claude Web Search and Free Code Execution
Since 17 February 2026, Claude's web search and code execution reached general availability, and running code alongside a search is free. What that pairing unlocks.
Claude's New Admin API for Enterprise
Anthropic launched a dedicated Admin API for Claude Enterprise on 14 July 2026, covering user, role and group management. What it covers, and why IT should learn it first.
Moving Your Team to the Claude 5 Models
A practical checklist for migrating to Sonnet 5 and Opus 5: the tokenizer change, the sampling-parameter break, effort defaults, and the deadline that prices the work.
The Effort Parameter: Claude’s Most Useful Dial
Effort went from beta to the primary reasoning control across the Claude range in six months. What each level means, and how to stop paying for thinking nobody needed.
Steering Claude Mid-Task, Without Starting Again
Mid-conversation system messages let you redirect Claude partway through a task instead of restarting it. How the feature works and the habit that makes it pay.
Multiagent Orchestration: Agent Teams, Not Just Agents
Multiagent orchestration and Outcomes entered public beta on 6 May 2026. What dividing work between Claude agents involves, and where it goes wrong.
What a Million-Token Context Window Actually Changes
The 1M-token window went from premium beta to standard across the Claude range. What that scale means for documents, and the old habits worth unlearning.
Memory and Dreams: What Claude Should Remember
Agent memory reached public beta in April; Dreams, the consolidation feature, followed in May and spread in July. The judgement calls come before the configuration.
When Claude Says No: Refusals and Fallbacks
Claude can now decline a request with a distinct refusal outcome, and you can define what happens next. Designing for the no is what makes workflows reliable.
Claude’s Developer Tools, Explained for Everyone Else
The ant CLI, the Advisor tool and the open-source Claude API skill, explained for people who will never open a terminal, and why that understanding pays.
Claude Opus 4.7 and xhigh: A Power User’s Guide
Opus 4.7 brought the xhigh effort level, task budgets and high-resolution image input, with breaking changes attached. What still matters now Opus 5 exists.
Adopting a Brand-New Claude Model, Safely
Fable 5 and Mythos 5 launched, wobbled, and stabilised inside a month. The June episode is the best recent case study in how to adopt a new model family.
Claude Opus 5: What Changed for Your Work
Opus 5 keeps Opus 4.8 pricing, turns thinking on by default and makes effort the dial. The breaking change and the max_tokens trap to check before you switch.
Claude Sonnet 5: The Price Cut That Is Not
Sonnet 5 launched with introductory pricing and a new tokenizer that counts about 30% more tokens for the same text. Do the arithmetic before 31 August.
What the METR Studies Actually Found About AI Coding Productivity
Developers believed AI made them 20% faster and were measured 19% slower. What the 2025 and 2026 METR studies found, and what closed the gap.
Why So Many People Hide Their AI Use At Work
A 2025 US survey found nearly half of employees who use AI at work hide or downplay it. Why that happens, and what it costs a team.
Why Claude Code Is Rated the Most-Loved AI Coding Tool
Claude Code topped the Pragmatic Engineer survey of 906 engineers. What separates the people getting that result from everyone else.
The Claude Cowork and Claude Code Education Pack
Every free Anthropic course and exam for Claude Cowork and Claude Code, plus the best community guides, curated into one followable path.
Claude Mythos 5: The Model You Cannot Buy
Mythos 5 is Fable 5 with some safeguards lifted, released through Project Glasswing to approved organisations. What that separation means for your planning.
Claude Fable 5: What It Is and What It Costs
A Mythos-class model made safe for general use, at $10 and $50 per million tokens. What the price buys, how the safeguards behave, and when to reach for it.
Your First 30 Days With Claude at Work
A month-long plan for getting Claude into a working week: which three tasks to pick, the criteria that make a task suitable, and what good looks like on day 30.
Claude Design: Prompt to Prototype, Honestly
What Claude Design makes, where the export list says it belongs in your process, the handoff bundle that is the real unlock, and the limits Anthropic admits to.
Claude Pro vs Max: Is $100 or $200 Worth It?
What the 5x and 20x usage multipliers really mean, the rule that beats guessing which tier you need, and how to price your own overflow before you upgrade.
Why Claude Burns Through Tokens So Fast
The seven things Anthropic says drive usage, which one is the real multiplier, and the changes that cut consumption without changing what you get back.
Claude Haiku 4.5: When the Small Model Wins
Haiku 4.5 costs $1 and $5 per million tokens and runs at more than twice the speed of Sonnet 4. Where that trade pays off, and the two limits to plan around.
How Claude Tokens and the Context Window Work
Tokens, context windows, usage limits and length limits are four different things. What each one controls, and the caching lever most people never pull.
How to Learn Claude Code in a Fortnight
A two-week order for learning Claude Code: which repo to start on, why the first three days are read-only, what belongs in CLAUDE.md, and where your usage goes.
Claude Chat vs Cowork: Which to Use
Chat is not being retired, but serious work is leaving it. The copy-paste tax, what long threads cost you, and a simple test for which tool a job belongs in.
What Is Claude Cowork?
Claude Cowork runs multi-step work on your own files and hands back a finished deliverable. What it is, what changes about how you brief it, and who it suits.
Looking for the research write-ups? They are here too, filed under Research, and the free courses and exams worth doing are in the Education Pack.