Claude Projects, Styles & Artifacts: the field guide we actually use

What “repeatable” actually means

A chat is disposable. You can produce something excellent in it, but the moment the tab closes, the instructions, examples and judgement calls made along the way close with it. The next person — or the same person next week — starts from a blank page and re-explains the brief.

That’s the test worth applying to each feature below: does it only make this one answer better, or the next twenty, without you explaining again?

Claude Projects: the workspace that remembers your brief

A Project is a self-contained workspace inside Claude with its own chat history and knowledge base. You upload what matters — brand guidelines, a style guide, approved past copy, a pricing sheet — and write project instructions that set the tone and role Claude takes every time someone opens a chat inside it. From then on, nobody pastes the brand voice guide into the chat before writing.

The workflow we actually run: a “Brand Brain” project per client, loaded with guidelines, tone examples and previously approved work. When we took on SEO content for a luxury hotel chain across 70-plus pages, the brief was to hold a consistently premium tone while multiple writers worked in parallel — normally a losing battle. With the brand guidelines and prior approved content sitting in one project, every writer started from the same base instead of guessing at house style, and the production timeline for that content compressed from around two weeks to three days. We’ve written the longer version of that story in why we chose Claude over ChatGPT, and the exact 45-minute build in how to build a Campaign Brain.

Projects are available to everyone, including free accounts — Free is capped at five projects, which is plenty for testing the idea before you commit a team to it. Paid plans (Pro, Max, Team, Enterprise) get a real upgrade underneath the hood: once a project’s knowledge base approaches its context limit, Claude automatically switches on retrieval so the effective capacity expands by up to 10 times, without you managing it. On Team and Enterprise, projects can also be shared across the organisation with two permission levels — “can use” for people who need to work inside it, and “can edit” for people who maintain it — which is what turns one person’s tidy setup into a whole account team’s shared resource.

Styles: becoming Skills, and why that’s not a problem

Styles let you set how Claude communicates — a tone and structure preset applied to every message in a chat. You could pick a built-in preset (Normal, Concise, Formal, Explanatory) or build a custom style either by uploading writing samples or describing what you want in plain language. It’s a genuinely useful idea: match a specific executive’s LinkedIn voice, or keep a whole content team inside one client’s tone without retyping instructions every session.

Here’s the current-state nuance that a lot of guides are going to miss: Anthropic announced on 1 June 2026 that Styles are being folded into Skills. Any custom style you’ve built gets migrated automatically — it will show up under Customize > Skills, disabled by default, so you’ll need to switch it on to keep using it. Most default styles (Concise, Explanatory, Formal) are being retired outright; Learning survives as a skill in its own right. Until the migration fully completes, the old styles menu and the new skills list sit side by side, so you can move across at your own pace rather than losing anything overnight. Once you do switch, a migrated style is triggered with a slash command in the format /{style-name}-style.

The practical takeaway: if someone on your team built a custom style for client tone, check Customize > Skills before assuming it’s gone. Setting this up fresh today, build it as a skill (or fold the guidance into your standing instructions) rather than investing in the styles menu as it stands — Anthropic has already told us where this is heading. Both are available from Free through Enterprise, provided code execution is switched on. The hour-long walkthrough is in Build Your First AI Skill.

Artifacts: the part your client actually sees

An Artifact is what Claude produces when the output is substantial enough to stand on its own — typically 15-plus lines you’d want to edit, reuse, or hand to someone else: a document, a code snippet, a single-page HTML site, an SVG, a diagram, or an interactive component. Instead of scrolling a long chat to find it, the content opens in its own panel with version history, and can be edited in place — highlight a section of a Markdown artifact, click “Edit with Claude,” and the change lands exactly where you pointed.

The workflow: build an interactive scorecard or a pricing calculator for a client pitch, publish it, and send the link. Nobody on the receiving end needs a Claude account to open it and use the basic functionality — only customising it or touching an AI-powered feature prompts a sign-up. That’s a materially different deliverable to a static PDF, and it costs nothing extra to distribute: if the artifact has AI built in and 200 people use it, that usage draws on each of their own Claude limits, not yours.

The exact language changes by plan. On Free, Pro and Max, you “publish” an artifact — a public link anyone can open. On Team and Enterprise, you “share” instead — it stays inside your organisation, and viewers must be logged into your account (plus have access to the source project, if it came from one). Publishing also gives you an embed code with an allow-list for which sites can host it. One thing worth knowing before you publish anything sensitive: unpublishing is permanent — you can’t bring the same artifact back, only create a fresh one.

Artifacts themselves are available from Free right through Enterprise, but two of the more powerful capabilities are gated to paid plans: connecting an artifact to outside tools via MCP (Asana, Google Calendar, Slack, or a custom connector) and giving it persistent storage so it remembers data across sessions both require Pro or above. And if your team works inside Claude Code, Team and Enterprise plans can publish a coding session’s output as a live, interactive page — a pull-request walkthrough or a dashboard that updates as the session continues — shareable inside the org, though not published to the public web.

The supporting cast: Skills, connectors and Cowork

Three more pieces are worth knowing, because they turn Projects, Styles/Skills and Artifacts from separate conveniences into one connected system.

Skills are folders of instructions Claude loads only when relevant, rather than every message — the opposite of a Project’s “always loaded” knowledge base. Anthropic ships built-in skills for tasks like Excel, Word, PowerPoint and PDF creation, and your organisation can build its own: a report format, a compliance checklist, a brand style guide. On Team and Enterprise, an Owner can push a skill to everyone, on or off by default, so the whole team works from the same procedure without manual setup.

Connectors, built on Anthropic’s open Model Context Protocol (MCP), are what let Claude reach your other tools — Google Drive, Gmail and Calendar for everyone; Team and Enterprise extends to GitHub, Microsoft 365 and Slack, switched on by an org Owner first. Artifact MCP connections (Asana, Calendar, Slack, or a custom connector) need Pro or above. This is what turns a skill or artifact from clever-in-isolation into wired-into-your-actual-client-data. New to the word? Agents, MCP and Skills, explained.

Cowork is where this comes together for genuinely multi-step work. It brings the agentic engine behind Claude Code to non-coding tasks, no terminal required: describe an outcome, step away, and come back to a formatted spreadsheet, a finished deck, or an organised folder. It’s a paid-plan feature (Pro and above, not Free), expanding beyond desktop to web and mobile from 7 July 2026, in beta, starting with Max. A Friday reporting run is a fair example: Cowork pulls from a connected client folder, references the brief in a Project, applies a provisioned skill for formatting, and delivers a finished deck while the account manager is still in a client call. Our Claude for Marketing training covers agents, MCP and skills hands-on if you want to build this rather than just read about it.

Why this adds up to a system, not a feature list

Any one of these features alone is a nice-to-have. A Project without a Skill still means someone remembers the output format by hand. A Skill without a Project means re-explaining the client’s context every time. An Artifact without a connector is a one-off, however polished. Stacked together — Project holding the brief, Skill holding the format, Artifact holding the shareable output, Cowork and connectors doing the fetching — the work becomes something your team can repeat without you standing over it.

That’s the actual reason LOKAL bothered to learn this properly rather than treat Claude as a fancier search box. We’ve trained more than 5,000 people across the Philippines and Australia since 2017, at a 98% satisfaction rate, and one 4,000-person enterprise rollout hit 72% weekly AI use within 30 days — numbers that only happen when a tool is set up as a system, not a clever trick one person knows.

Starting this week: pick one recurring deliverable — a client report, a content brief, a proposal — build a Project around it with the reference material that matters, then decide whether it needs a skill or a connector on top. Adding fancier layers before the basic workspace exists is how most rollouts stall. The plan and governance version of this is in our Claude for business starter guide.

Keep going: the AI guides hub has the rest of the Claude deep-dives, and the Insights archive carries everything else the team publishes.