Claude Code, What Six Months of Daily Use Actually Delivers

Video: 6 Monate Claude Code: 138 Webseiten, 29 Apps und meine ehrlichen Learnings

Oct 17, 2026 · based on the channel Calvin Hollywood (AI)

Claude Code is Anthropic's terminal-based coding agent, and the most useful way to judge it is by output: in six months of near-daily use, Calvin Hollywood (AI) reports 138 websites, 29 apps and fully AI-edited video for under €300 a month in tooling. The full breakdown, including the tools and the failures, is in his video 6 Monate Claude Code: 138 Webseiten, 29 Apps und meine ehrlichen Learnings, and the accompanying Claude Masterclass is available with a 7-day free trial at the Claude Masterclass on Skool. This article extracts the practical lessons for anyone evaluating Claude Code as a production tool rather than a demo.

What Claude Code is and why it became the hub

Claude Code is a command-line agent that reads a project, plans changes, edits files and runs commands — and the first structural lesson from six months of use is to route everything through it instead of spreading work across chat windows and separate apps. The creator describes consolidating content production, app building and even video work into a single Claude Code environment, with the terminal treated as the entry point rather than a specialist tool. That matters because the agent keeps context about the project: file structure, conventions, earlier decisions. Instead of re-explaining the setup for every task, you describe the goal once and let the agent work inside the existing codebase. The practical consequence is that the learning curve is front-loaded: once the environment, permissions and habits are in place, new projects start from a working baseline rather than from zero.

The output: 138 websites and 29 apps without WordPress

Website production scaled to 138 sites in half a year, and the notable part is the absence of WordPress — pages are generated as code and deployed statically. The workflow is straightforward in principle: Claude Code writes the HTML, CSS and JavaScript, Netlify handles hosting and deployment, and there is no CMS, no plugin maintenance and no database to patch. For content-heavy projects this removes a whole category of recurring work — updates, security fixes, plugin conflicts — that normally eats time without producing anything visible. The 29 apps follow the same pattern: small, purpose-built tools rather than large platforms, each generated and iterated inside Claude Code. A social media control center nicknamed "Social Lab" is one example of an internal tool that would rarely justify a development budget but pays off immediately when building it costs an afternoon instead of a sprint.

The tool stack behind the numbers

Four tools carry most of the workload, and each covers a distinct layer. Claude Code is the production layer where code, content and automation are written. Netlify is the delivery layer: static hosting, fast deploys, no server administration. Higgsfield covers generative images and video, which removes the need for stock libraries or separate design sessions for visual assets. Composio connects Claude to external services, most visibly social media, so publishing and scheduling can be triggered from the agent instead of handled by hand. The combination is deliberately boring: no exotic infrastructure, no custom backend, nothing that requires an on-call engineer. That restraint is the point — the stack stays cheap and legible, and the agent can operate it end to end without a human translating between systems.

Video editing with Claude Code and six free tools

The most surprising result is automated video editing: reels assembled by Claude Code driving a set of six free editing tools, with no human cutter in the loop. One produced reel reportedly reached 110,000 views, which is the relevant benchmark — the output was not merely technically finished, it performed. The workflow depends on the agent orchestrating existing software rather than replacing it: Claude Code handles the sequencing, file operations and repetitive decisions, while the free tools do the rendering. The honest caveat is that this is fragile in the way all automation is fragile — a missing file or a changed export setting breaks the chain — so the process needs to be written down and testable. But the economics are hard to argue with: editing capacity that normally requires a freelancer becomes a script that runs again tomorrow.

Trust, prompting and the backup rule

Reliability comes from process discipline, not from the model being infallible. Three habits recur in the material. First, prompt with military precision: short, unambiguous instructions with explicit constraints beat long conversational requests, because the agent executes what is written, including the parts you left vague. Second, keep backups — before any large refactor or bulk file operation, snapshot the project, since an agent that can rewrite twenty files in a minute can also rewrite the wrong twenty. Third, use skills: write recurring instructions down once as a reusable skill instead of re-explaining them every session, so conventions survive across projects and context windows. The same logic applies to automation with personality — scheduled or triggered jobs that behave consistently because the rules live in a file, not in someone's memory.

The economics: under €300 instead of €15,000

The headline learning is financial: the entire operation runs for under €300 per month, against a quoted €15,000 for the conventional route of freelancers, agencies and licensed software. That gap is not a discount on the same output — it reflects a different production model in which one person plus an agent covers work that previously required several specialists. The monthly figure includes the Claude subscription and the supporting tools, which is why the stack was kept deliberately small. For anyone planning a similar setup, the implication is to budget for subscriptions and hosting first and treat headcount as the last resort, not the default. It also sets a realistic expectation: the savings come from volume and repetition, so the model pays off when you ship many small things rather than one large, risky project.

Common questions from practitioners

Experienced developers and newcomers raise the same concerns from opposite directions. One recurring theme is whether prior programming knowledge is required: the honest answer is that basic literacy helps — knowing what a terminal is, how files are structured, what a deploy does — but the agent closes most of the gap for small projects. Another is whether the subscription tier matters; heavier daily use across many projects pushes toward the larger plans, while occasional use is fine on smaller ones. A third is scope creep: the tool makes it easy to start ten things, so the discipline of finishing one site or app before opening the next is what separates a portfolio from a folder of half-built experiments. Finally, scepticism about quality is fair — generated code still needs review, and the agent's confidence is not a substitute for checking the result.

Conclusion

Claude Code rewards a specific kind of user: someone who ships many small projects, keeps the stack simple and treats the agent as a production system rather than a novelty. The six-month record — 138 websites, 29 apps, automated video editing, under €300 a month — is less about raw model capability than about consistent process: everything in one environment, precise prompts, backups, reusable skills and a short list of dependable tools. Calvin Hollywood (AI) is the channel behind this breakdown, and the full walkthrough is worth watching for the live demos alone. To go further, the Claude Masterclass with a 7-day free trial bundles the tutorials, a fundamentals course and session recordings.

Source: 6 Monate Claude Code: 138 Webseiten, 29 Apps und meine ehrlichen Learnings · Markdown version