What changed and why it matters to you

Anthropic launched Claude Academy on August 20 — 355 free courses, tutorials, and role use cases. It is the best product training any AI company has published, so we mapped the whole catalog rather than competing with it: which six resources matter for each of five roles, in what order, and what each track leaves you to solve alone. 175 of our articles now carry a link to the official course covering the same ground. The position this settles: Anthropic teaches the product, we cover the job around it — what breaks in month six, what it costs, who has to approve it, and when to pick something else.

The comparison pages were still benchmarking GPT-4o and Gemini 1.5. All seven are rebuilt against current models with a visible "last verified" date, and the format changed as well as the facts: verified specs first, then pricing with the caveats that make sticker prices misleading, then performance claims with attribution to whoever published them, then a method to settle the question on your own workload. Where the honest answer is "these are close and the model is not your problem," the page now says so. Three findings worth the click: OpenAI meters long context at roughly double its headline rate while Anthropic does not, Gemini Flash pricing doubles on January 1 2027, and Sonnet 5 is now only 2x Haiku 4.5 rather than 3.75x — which inverts the standard advice about dropping a tier.

Two things we had wrong. Our Claude Sonnet 5 article told readers to budget for a 50% price increase on September 1 — Anthropic cancelled that increase on August 10 and made $2/$10 permanent. Corrected. The cost calculator was also still priced against Claude Opus 4 at $15/$75, overstating current flagship cost by three times; it now runs on Opus 5, Sonnet 5, Haiku 4.5, and Fable 5, reading from the same pricing file as the comparison pages so the two can never disagree. We audited all 254 articles for stale model references while we were in there.

Computer use went generally available alongside a new browser use toolset, with the Skills API and Files API GA the same day. Auto mode became the default in Claude Code on Pro, Max, and Team plans. Managed Agents gained hard per-session spend budgets, an advisor model the primary thread can consult mid-turn, inference geo pinning, and Skills loaded straight from a GitHub repository. Eleven new timeline events covering August 6 to 20, including Gemini 3.7 Flash.

Four articles completing the developer production path: deploying to production (secrets, rate limits, cost controls, observability checklist), production error handling (API error taxonomy, output format resilience, hallucination detection, user-facing errors), end-to-end Next.js + Claude chatbot (streaming SSE, Vercel deploy, conversation history), and what to actually build with Claude as a first product (the 2x2, structural advantages, the test before you build).

Four new articles: prompt caching implementation (cache_control, multi-block caching, verifying hits, 80% cost reduction patterns), cutting Claude API costs (model routing, batch API, output length control, context management), building a chatbot with persistent memory (user facts, session summaries, extraction patterns), and using Claude for note-taking and knowledge management (Projects, capture workflows, retrieval patterns).

Four articles completing the developer and founder content: tool use implementation deep dive (defining tools, parsing responses, parallel tool calls, production failure modes), multi-agent orchestration basics (orchestrator + subagents, parallelism, state management, checkpoints), customer discovery with Claude (interview simulation, assumption mapping, synthesis), and meeting prep patterns.

An 8-step implementation path for developers: API basics, streaming, RAG pipelines, tool use, evals in CI, prompt caching, and agentic loops. Assumes you can code — no business-case framing. Visually distinct from the operator paths.

Four technical articles: your first Claude API call (auth, streaming, structured output, errors), building a RAG pipeline from scratch (chunking strategy, hybrid retrieval, reranking, evals), writing evals that catch regressions (test case structure, LLM-as-judge, CI integration), and streaming implementation patterns (SSE to browser, error handling, UX).

A 7-step path for solo founders and early-stage builders: validate your idea, choose the right stack, avoid the failure modes, price your product, get your first ten customers, and pitch to investors. No ML background required.

Seven new articles targeting the zero-person startup: the solo founder operating system, idea validation without fooling yourself, build/buy/prompt stack decisions, AI product failure modes, investor pitch narrative, pricing your AI product, and getting your first ten customers.

Four new articles: how to give Claude precise connector instructions (the difference between "find that doc" and an instruction that reliably works), managing email with Claude, the 20-minute weekly review, and using Claude to declutter your digital life.

A 10-step path for IT leads and administrators: from evaluating Claude to running it across an organisation. Three stages — evaluation, deployment, ongoing management — each with dedicated articles.

20 Claude features grouped by capability area — core interface, skills & connections, automation & agents, research & intelligence, admin & scale. Each with a definition and a guide.

A full batch of operator-focused articles covering: admin setup, plan selection, org-wide Projects architecture, usage policy, system prompt writing, role-type field notes (finance, product, legal, data, engineering, executives), change management, token control, security, and more.

Practical, role-specific articles covering what AI actually looks like for each function — what works, what doesn't, and how to set up Claude for that team.

Two structured paths: "Figuring out AI for your company" for founders and ops leaders, and "Getting your team actually using AI" for department heads. Each walks through the key decisions in order.

Glossary terms for all major Claude features: Projects, Skills, Connectors, Artifacts, Memory, Cowork, Dispatch, Deep Research, Computer Use, Managed Agents, Claude Plans, Claude Code, MCP, and more.

AI Codex launched with a glossary, articles, and the first learning path. Built for operators and admins implementing Claude at work — plain-English explanations, practical guides, no hype.